Will Bases de La Toma de Desiciones

ANRV314-NE30-21 ARI 21 May 2007 13:44
The Neural Basis of

Decision Making
Joshua I. Gold1 and Michael N. Shadlen2
Annu. Rev. Neurosci. 2007.30:535-574. Downloaded from www.annualreviews.org
1
Department of Neuroscience, University of Pennsylvania, Philadelphia,
Pennsylvania 19104-6074; email: jigold@mail.med.upenn.edu
2
Howard Hughes Medical Institute and Department of Physiology and Biophysics,
University of Washington, Seattle, Washington 98195-7290;
email: shadlen@u.washington.edu
by HINARI on 11/23/12. For personal use only.
Annu. Rev. Neurosci. 2007. 30:535–74 Key Words

The Annual Review of Neuroscience is online at psychophysics, signal detection theory, sequential analysis, motion
neuro.annualreviews.org
perception, vibrotactile perception, choice, reaction time
This article’s doi:
10.1146/annurev.neuro.29.051605.113038 Abstract
Copyright c 2007 by Annual Reviews. The study of decision making spans such varied fields as neuro-
All rights reserved
science, psychology, economics, statistics, political science, and com-
0147-006X/07/0721-0535$20.00 puter science. Despite this diversity of applications, most decisions
share common elements including deliberation and commitment.
Here we evaluate recent progress in understanding how these basic
elements of decision formation are implemented in the brain. We
focus on simple decisions that can be studied in the laboratory but
emphasize general principles likely to extend to other settings.
535
ANRV314-NE30-21 ARI 21 May 2007 13:44
of this approach for inferring principles of

Contents higher brain function.
INTRODUCTION . . . . . . . . . . . . . . . . . 536
Elements of a Decision . . . . . . . . . . . 536 Elements of a Decision
Conceptual Framework . . . . . . . . . . . 538
The decisions required for many sensory-
EXPERIMENTS . . . . . . . . . . . . . . . . . . . 542
motor tasks can be thought of as a form
Perceptual Tasks. . . . . . . . . . . . . . . . . . 542
of statistical inference (Kersten et al. 2004,
Simple Motor Latencies: Deciding
Rao 1999, Tenenbaum & Griffiths 2001, von
When to Initiate an Action . . . . . 556
Helmholtz 1925): What is the (unknown)
Value-Based Decisions . . . . . . . . . . . . 560
state of the world, given the noisy data pro-
CONCLUSIONS . . . . . . . . . . . . . . . . . . . 561
vided by the sensory systems? These decisions
select among competing hypotheses h1 . . . hn
(often n = 2) that each represent a state of

the world (e.g., a stimulus is present or ab-
INTRODUCTION sent). The elements of this decision process
A decision is a deliberative process that results (see Figure 1) are described in terms of prob-
in the commitment to a categorical proposi- ability theory, as follows.
tion. An apt analogy is a judge or jury that must The probability P(hi ), or prior, refers to
take time to weigh evidence for alternative the probability that hi is true before obtain-
interpretations and/or possible ramifications ing any evidence about it. In the courtroom
before settling on a verdict. Here we evalu- analogy, priors correspond to prejudices that
ate progress in understanding how this pro- can bias jurors’ judgments. Bayesian infer-
cess is implemented in the brain. Our scope ence prescribes a more positive role for pri-
is somewhat narrow: We consider primarily ors, which are necessary to convert measur-
studies that relate behavior on simple sensory- able properties of evidence (the values it can
motor tasks to activity measured in the brain attain when hi is true) to inferred ones (the
because of the ability to precisely control sen- probability that hi is true given a particular ob-
sory input, quantify motor output, and target servation). For a sensory-motor task, a prior
relevant brain regions for measurement and typically corresponds to the predicted prob-
analysis. Nevertheless, our intent is broad: We ability of seeing a particular stimulus or re-
hope to identify principles that seem likely ceiving a particular reward on the upcoming
to contribute to the kinds of flexible and nu- trial, which can be instructed (e.g., Carpenter
anced decisions that are a hallmark of higher & Williams 1995, Basso & Wurtz 1998, Dor-
cognition. ris & Munoz 1998, Platt & Glimcher 1999) or
The organization of this review is as fol- inferred from its relative frequency of occur-
lows. We first describe the computational el- rence on previous trials (Sugrue et al. 2004).
ements that comprise the decision process. The evidence (e) refers to information that
We then briefly review signal detection the- bears on whether (and possibly when) to com-
ory (SDT) and sequential analysis (SA), two mit to a particular hypothesis. A strand of hair
related branches of statistical decision theory found at a crime scene can be used as evidence
that represent formal, mathematical prescrip- if it supports or opposes the hypothesis that a
tions for how to form a decision using these certain person was present at that location.
computational elements. We then dissect sev- For a perceptual task, neural activity that rep-
SDT: signal
detection theory eral experimental results in the context of this resents immediate or remembered attributes
theoretical framework to identify neural sub- of a sensory stimulus can be used as evidence.
SA: sequential
analysis strates of decision making. We conclude with However, like hair at a crime scene, this sen-
a discussion of the strengths and limitations sory activity is evidence only insofar as it bears
536 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
In the world In the brain
Context Motivation to
e.g., instruction perform the task
Contextual cues and
prexisting knowledge
Consideration of
Two possible states two propositions
(hypotheses)
{up, down}
h1 :up or h2 : down
Sensory data
One state holds
e.g., up x = { x1 , x2 , ... }
Information flow
for each decision Useful form
of evidence
e
Consequence
of action & state Statistical knowledge
(4 possible) Decision variable

e.g., y ≈ ... likelihoods: P ( e hi )
P ( e h1 )
l12 (e) ≡ or logLR12 ≡ log [ l12 (e)] priors: P ( hi )
P ( e h2 )
values: v (Hj hi )
Apply decision rule Establish

One action e.g., choose left if decision rule
e.g., answer “up” l12 (e) ≥ criterion based on goals
Evaluation Experience payoff or cost

value or utility
vij∈{1,2}
Figure 1
Elements of a simple decision between two alternatives. The left side represents elements of the world.
The right side represents elements of the decision process in the brain. Black elements establish context.
Red elements form the decision. Blue elements evaluate and possibly update the decision process.
on a hypothesis. Thus e is useful when it can Value (v) is the subjective costs and benefits
be interpreted in the context of conditional that can be attributed to each of the potential
probabilities such as P(e | h i ), the “likelihood” outcomes (and associated courses of action)
function describing the values that e can attain of a decision process. Value can be manip-
when hi is true. Perceptual tasks are useful for ulated by giving explicit feedback or mone-
studying decision formation in part because tary rewards to human subjects or preferred
of the ability to control precisely the quan- food or drink to nonhuman subjects. Value
tity and quality of the sensory evidence and can also reflect more implicit factors such as
measure the impact on likelihood functions the costs associated with wasted time, effort,
obtained from relevant sensory neurons. and resources. Here we make no distinction
www.annualreviews.org • Decision Making 537

ANRV314-NE30-21 ARI 21 May 2007 13:44
between value and utility, disregarding their tion (Cisek 2007, Clark 1997, Merleau-Ponty
technical meanings and opting instead for a 1962, O’Regan & Noë 2001). A key practi-
more general concept that describes subjec- cal implication is that the parts of the brain
DV: decision
variable tive influences on the decision process. responsible for selecting (or planning) certain
The decision variable (DV) represents the behaviors may play critical roles in forming
accrual of all sources of priors, evidence, and decisions that lead to those behaviors.
value into a quantity that is interpreted by the The goals of a decision maker are to
decision rule to produce a choice. It is a con- achieve desired outcomes and avoid unde-
ceptual entity corresponding to the delibera- sired ones. Desired outcomes include “get-
tions in a trial leading up to the verdict. Note ting it right” or maximizing the percentage
that here “deliberations” does not imply that of correct responses in tasks that have right
the DV is necessarily computed rationally and and wrong answers or, more generally, maxi-
without emotion; rather, it emphasizes that mizing expected value (Green & Swets 1966).
the DV is capable of accounting for multiple Undesired outcomes include getting it wrong,
sources of information (priors, evidence, and minimizing value and wasting time, effort,
value) that are interpreted over time. Thus, or resources. Goals are critical because the
the DV is not tied to the (possibly fleeting) decision process is assumed to be intended,
appearance of stimuli but spans the time from and perhaps even optimized, to achieve them.
the first pieces of relevant information to the Indeed, optimality can be assessed only in
final choice. Also, unlike the choice, which is the context of a goal. Thus, behavior that is
discrete, it is often best thought of as an ana- “suboptimal” with respect to certain objective
log quantity. We spend much of this chapter goals such as maximizing accuracy might in
refining the concept of a DV and describing fact be optimal with respect to the idiosyn-
efforts to find its neural correlates. cratic goal(s) of the decision maker.
The decision rule determines how and Evaluation, or performance monitoring,
when the DV is interpreted to arrive at a com- is necessary to analyze the efficacy or op-
mitment to a particular alternative Hi (the timality of a decision with respect to its
choice associated with hypothesis hi ). The rule particular goals. For laboratory tasks, evalu-
causes the jury to declare, “we have a verdict.” ation can occur with or without explicit feed-
A conceptually simple rule is to place a cri- back (e.g., Carter et al. 1998, Ito et al. 2003,
terion value on the DV. This rule requires a Ridderinkhof et al. 2004, Schall et al. 2002,
DV whose magnitude reflects the balance of Stuphorn et al. 2000). In either case it is likely
support and opposition for a hypothesis. Such to play a critical role in shaping future deci-
a rule allows the decision maker to achieve at sions via learning mechanisms that, in princi-
least one of several appealing long-term goals, ple, can affect every aspect of the process, from
including maximizing accuracy or reward or incorporating the most appropriate priors, ev-
achieving a target decision time. idence, and value into the DV to establishing
The course of action that follows the com- the most effective decision rule.
mitment to an alternative is often necessary
to reap the costs and benefits associated with
that alternative. In these cases, the decision Conceptual Framework
itself might be best thought of not as an Signal detection theory. SDT is one of the
abstract computation but rather as the ex- most successful formalisms ever used to study
plicit intention to pursue (or avoid) a par- perception. Unlike information theory and
ticular course of action. This idea is a form other biostatistical tools commonly used for
of “embodiment” that places high-order cog- data analysis, SDT prescribes a process to
nitive capacities such as decision making in convert a single observation of noisy evidence
the context of behavioral planning and execu- into a categorical choice. Early applications
538 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
allowed psychologists to infer from behavior Green & Swets (1966), where these expres-
properties of the underlying sensory repre- sions are derived.
sentation (Green & Swets 1966). Later, pi- SDT thus provides a flexible framework
LR: likelihood ratio
oneering work in retinal and somatosensory to form decisions that incorporate priors,
logLR: logarithm of
physiology established SDT as a valuable tool evidence, and value to achieve a variety of
the likelihood ratio
to relate the measured responses of sensory goals. Unfortunately, this flexibility also poses
neurons to the limits of detection and discrim- a challenge to neurobiologists. The above ex-
ination (for reviews see Parker & Newsome pressions were obtained assuming that the DV
1998, Rieke et al. 1997). More recently, it has is the likelihood ratio (LR), l 12 (e). However,
begun to shed light on decision mechanisms. equivalent expressions (that is, those that will
According to SDT, the decision maker ob- achieve the same goals) can be obtained (by
tains an observation of evidence, e. In per- scaling β) using any quantity that is monoton-
ceptual psychophysics, e is derived from the ically related to the LR. In other words, these
senses and might be the spike count from equations do not constrain the priors, e, value,
a neuron or pool of neurons, or a derived the DV, or β to take on any particular form,
quantity such as the difference between spike only that they interact in a certain way. Thus
rates of two pools of neurons. It is caused it is difficult to assign a quantity measured in
by a stimulus (or state) controlled by the ex- the brain to any one of these elements with-
perimenter; e.g., h1 (stimulus present) or h2 out knowing how the others are represented.
(stimulus absent). If e is informative, then its One powerful approach to unraveling this co-
magnitude differs under these states. How- nundrum is to exploit differences in the time
ever, e is also corrupted by noise. Thus e is scales of these elements in decision formation.
a random variable described by a distribution
whose parameters (e.g., the mean) are set by h1 Sequential analysis. SA is a natural exten-
or h2 . These conditionalized distributions de- sion to SDT that accommodates multiple
scribe the likelihoods P(e | h 1 ) and P(e | h 2 ). pieces of evidence observed over time. SA
Unlike standard statistical methods, the ob- assumes that the decision has two parts: the
ject of SDT is not to determine whether the usual one between h1 and h2 , and another
parameters describing these distributions are about whether it is time to stop the process
different but instead to decide which of the and commit (Figure 2). In its most general
states gave rise to the observation e. form, SA allows the procedure for construct-
The decision requires the construction of ing the DV and the decision rule to be adjusted
a DV from e. For binary decisions, the DV is with each new sample of evidence. However,
typically related to the ratio of the likelihoods many decisions can be understood by assum-
of h1 and h2 given e: l 12 (e) ≡ P(e | h 1 )/P(e | h 2 ). ing fixed definitions for these elements. A sim-
A simple decision rule is to apply a criterion to ple DV constructed from multiple, indepen-
the DV; e.g., choose h1 if and only if l 12 (e) ≥ β, dent pieces of evidence, e 1 , e 2 , . . . , e n , is the
where β is a constant. A strength of SDT is logarithm of the LR (logLR, or “weight of ev-
that a variety of goals can be reached by simply idence”), which is just the sum of the logLRs
using different values for the criterion. If the associated with each piece of evidence:
goal is accuracy and the two alternatives are
equally likely, then β = 1. If the goal is accu- P(e 1 , e 2 , . . . , e n |h 1 )
log L R12 ≡ log
racy and the prior probability favors one of the P (e 1 , e 2 , . . . , e n |h 2 )
hypotheses, then β = P (h 2 )/P(h 1 ). If the goal
n
P (e i | h 1 )
is to maximize value (where vij is the value as- = log . 1.
i=1
P (e i | h 2 )
sociated with choice Hj when hypothesis hi is
22 + v21 )P (h 2 )
true), then β = (v(v11 + v12 )P (h 1 )
. For more details, A simple stopping rule is to update this DV
the reader should refer to the first chapter of with new pieces of evidence until reaching a

ANRV314-NE30-21 ARI 21 May 2007 13:44
a Sequential analysis framework
e0 → f0 ( e0 ) ⇒ Stop
or
→
e1 → f1 ( e0 , e1 ) ⇒ Stop
or
→
e2 → f1 ( e0 , e1 , e2 ) ⇒ Stop
or
…
en → f1 ( e0 , e1 , …, en ) ⇒ …
b
Symmetric random walk

Choose H 1
A
e
for h1 over h2
Accumulated
Mean of e depends
evidence
0 on strength of
evidence
Mean drift rate = mean of e
-A
Choose H 2
c Race model
Choose H1 Choose H 2
evidence for h1
B
evidence for h2
A
Accumulated
Accumulated
Figure 2
Sequential analysis. (a) General framework. The decision is based on a sequence of observations. After
each acquisition, a DV is calculated from the evidence obtained up to that point; then more evidence can
be obtained or the process can be terminated with a commitment to H1 or H2 . In principle, both the
fi (· · ·)s, which convert the evidence to a DV, and the criteria can be dynamic (e.g., to incorporate the cost
of elapsed time). e 0 can be interpreted as the evidence bearing on the prior probability of the hypotheses.
(b) In random walk models, the DV is a cumulative sum of the evidence. The bounds represent the
stopping rule. If e is a logLR, then this process is the SPRT (see The Sequential Probability Ratio Test).
When the evidence is sampled from a Gaussian distribution in infinitesimal time steps, the process is
termed diffusion with drift μ, or bounded diffusion. (c) In the race model, two or more decision processes
represent the accumulated evidence for each alternative. When there are two alternatives and the
accumulations are inversely correlated, the race model is nearly identical to a symmetric random walk.
540 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
positive or negative criterion (the bounds in

Figure 2b). THE SEQUENTIAL PROBABILITY RATIO
Together, this DV and stopping rule com- TEST
prise the sequential probability ratio test
(SPRT) (see The Sequential Probability Ratio Consider the following toy problem. Two coins are placed in a
Test), which is the most efficient test for de- bag. They are identical except that one is fair and the other is
ciding between two hypotheses on this kind a trick coin, weighted so that heads appears on 60% of tosses,
of problem: It achieves a desired error rate on average. Suppose one of the coins is drawn from the bag,
with the smallest number of samples, on av- and we are asked to decide whether it is the trick coin. We
erage (Wald & Wolfowitz 1947). This proce- can base our decision on a series of any amount of tosses. The
dure played a prominent role in allowing Alan SPRT works as follows. Each observation (toss) e i is converted
Turing and colleagues to break the German to a weight of evidence, the logLR in favor of the trick coin
enigma cipher in World War II (Good 1979, hypothesis. There are only two possible values of evidence,
1983). Their success depended on not only heads or tails, which give rise to weights (wi ):
deducing the contents of intercepted mes- ⎧
⎪ P (e i = heads|h 1 : trick coin)
⎪
⎪ log
sages correctly, but also doing so in time for ⎪
⎪ P (e i = heads|h 2 : fair coin)
⎪
⎪
the information to be of strategic use. ⎪
⎨ = log 0.6 = 0.182 if heads
⎪
SA in numerous guises has been a valu- wi = 0.5
⎪
able tool for psychophysical analysis, particu- ⎪

⎪ P(e i = tails|h 1 : trick coin)
⎪
⎪log
larly for studying the trade-off between speed ⎪
⎪ P (e i = tails|h 2 : fair coin)
⎪
⎪
and accuracy (Luce 1986, Smith & Ratcliff ⎩ 0.4
= log = −0.223 if tails
2004). In recruitment or race models, evi- 0.5
dence supporting the various alternatives is According to SPRT, the decision variable is the running sum
accumulated independently to fixed thresh- (accumulation) of the weights. After the nth toss, the decision
olds (Audley & Pike 1965, LaBerge 1962, variable is

n
Logan 2002, Reddi et al. 2003, Vickers 1970). yn = wi
i=1
In other models that are more closely re-
We apply the following rules:
lated to the SPRT, a weight of evidence is
1−α
accumulated to support one alternative ver- if yn ≥ log answer “trick”
α
sus another (Busemeyer & Townsend 1993, β
Diederich 2003, Laming 1968, Link 1992, if yn ≤ log answer “fair”
1−β
Link & Heath 1975). These models mirror β 1−α
the mathematical description of a random if log y n log get more evidence
1−β α
walk or diffusion process (Ratcliff & Rouder where α is the probability that a fair coin will be misidentified
1998, Ratcliff & Smith 2004, Smith 2000, [i.e., a type I error: P (H1 | h 2 )] and β is the probability that
Smith & Ratcliff 2004): The accumulation a trick coin will be misidentified [a type II error:P(H2 | h 1 )].
of noisy evidence creates a virtual trajectory For example, if α = β = 0.05, then the process stops when
equivalent to the dancing movements of a tiny |yn | ≥ log(19). The criteria can be viewed as bounds on a
particle in Brownian motion (Figure 2). random walk. To achieve a lower rate of errors, the bounds
SA promises to play an important role in must be moved further from zero, thus requiring more sam-
the neurobiology of decision making. First, ples of evidence, on average, to stop the process.
investigators continue to develop neurobio-
logically inspired implementations of SA that
will help to identify where and how the brain momentary, whereas the DV evolves in time.
carries out the underlying computations (Lo Changes to either can affect accuracy or de-
& Wang 2006, Usher & McClelland 2001, cision times differently and can, in principle,
SPRT: sequential
Wang 2002). Second, SA provides a means to be distinguished in neural recordings (Hanks probability ratio test
distinguish evidence from the DV. Evidence is et al. 2006). Third, SA includes a termination

ANRV314-NE30-21 ARI 21 May 2007 13:44
rule. Analogous mechanisms in the brain are Perceptual Tasks

required to make decisions and commit to al-
Vibrotactile frequency (VTF) discrimina-
ternatives on a time frame that is not governed
VTF: vibrotactile tion. Developed in the 1960s by Mount-
frequency by the immediacy of sensory input or motor
castle and colleagues, the VTF paradigm
output, a hallmark of cognition.
requires the subject, typically a monkey, to
compare the frequency of vibration of two
tactile stimuli, f 1 and f 2, separated by a
EXPERIMENTS
time gap (Figure 3a). The range of fre-
Below we summarize key experimental results quencies used (∼10–50 Hz) does not acti-
that shed light on how the brain implements vate specialized frequency detectors but in-
the elements of a decision. We focus first on stead requires the nervous system to extract
perceptual decisions and how to distinguish the intervals or rate of skin depression (Luna
sensory evidence from the DV and the deci- et al. 2005, Mountcastle et al. 1990). This in-
sion rule. We then describe simple motor tasks formation is used to decide whether the fre-
that appear to engage similar decision mech- quency is greater in the first or second in-
anisms. Finally we discuss value-based deci- terval. The monkey communicates its answer
sions that weigh expectation and preference by pressing a button with the nonstimulated
as opposed to sensory evidence. hand.
500 ms
Base (f1) Comparison (f2) KU PB
b 1
f1 = 20 Hz
0.8
Proportion
0.6
of choices
f2 > f1 0.4
0.2
0
10 15 20 25 30
Frequency of comparison (Hz)
Figure 3
Neural correlates of a decision about vibrotactile frquency. (a) Testing paradigm. A test probe delivers a
sinusoidal tactile stimulus to the finger at base frequency f 1. After a delay period, a comparison stimulus
is delivered at frequency f 2. Then the monkey must decide whether f 2 > f 1, a decision it indicates by
releasing a key (KU) and pressing a button (PB) with its free hand. (b) Psychometric function. The task is
difficult when the base (20 Hz) and comparison frequencies are similar ( f 2 ≈ f 1). (c) Response of a
typical S1 neuron. Rasters show spikes from individual trials, grouped by the combination of base and
comparison frequencies (left). The neuron responds to the vibration stimulus in both the base and the
comparison periods. The firing rate encodes the vibration frequency similarly in the base and comparison
periods (brown and purple graphs, respectively), but it does not reflect the monkey’s choice (black and white
lines and data points in the lower panels). The neuron is uninformative in the interval between base and
comparison stimuli. (d ) Response of a neuron in ventral premotor cortex. Same conventions as in (c).
This neuron carries information about the base frequency during the interstimulus interval (blue). In the
comparison epoch (purple), the neuron is more active when f 2 < f 1. Note that for all trials, the base and
comparison frequencies differ by 8 Hz. Adapted with permission from Hernandez et al. (2000), Romo
et al. (2004).
542 Gold · Shadlen

ANRV314-NE30-21
Figure 3
(Continued )
ARI
21 May 2007
13:44
www.annualreviews.org • Decision Making

543
ANRV314-NE30-21 ARI 21 May 2007 13:44
Signals that encode the VTF stimulus have and VPC), more closely resembles a DV.
been traced from the periphery into the pri- Many neurons in these areas persist in firing
mary somatosensory cortex (S1). For most through the delay period between f 1 and f 2
S1 neurons, the average firing rate increases (Figure 3d, blue insert). Moreover, during the
monotonically with increasing stimulus fre- second interval, the activity of some of these
quency. For many S1 neurons, firing rate neurons reflects a comparison between f 2 and
modulations also follow the periodicity of the f 1 (Figure 3d, purple insert). However, identi-
stimulus (Figure 3c). However, the average fying the nature of this comparison can be dif-
firing rate is thought to represent the evi- ficult. For the example neuron in Figure 3d,
dence used to perform the task. First, behav- it is unknown whether the activity during
ioral sensitivity more closely matches the dis- the comparison period reflects the difference
criminability of S1 responses when average f 2 − f 1 or merely the sign of the difference
rates, rather than periodic modulations, are (which is more closely related to the decision
used. Second, trial-to-trial variations in the outcome than the DV) because in every case
rates, but not the periodic modulations, of f 2 − f 1 = 8 Hz. A less direct analysis using
S1 neurons predict to a slight but significant choice probability suggests that some neurons
degree the monkeys’ choices (Salinas et al. in these brain areas might represent the differ-
2000). This relationship, termed the choice ence between f 2 and f 1 (see Hernandez et al.
probability (Parker & Newsome 1998), is ex- 2002, Romo et al. 2004).
pected for neurons that provide noisy evi- This extraordinary body of work provides
dence for the decision, especially if the same to date the most complete picture of the di-
trial-to-trial variations are shared by other S1 versity of brain areas that contribute to de-
neurons ( Johnson 1980a,b; Kohn & Smith cision formation on even a simple sensory-
2005; Shadlen et al. 1996; Zohary et al. 1994). motor task. For some areas, their role in task
Third, replacing the VTF stimulus in the first performance seems clear. S1 provides the sen-
and/or second interval with electrical micro- sory evidence. Primary motor cortex helps to
stimulation of S1, in some cases using ape- prepare and execute the behavioral response.
riodic stimuli that lack regular intervals However, for the remainder of these brain ar-
between activations, elicits nearly the same eas (and doubtless others yet to be studied)
behavioral responses as does the physical stim- that lie at intermediate stages between sensory
ulus (Romo et al. 1998, 2000). input and motor output, many challenges lie
The firing rate of S1 neurons thus repre- in the way of an equally precise recounting of
sents the sensory evidence that underlies the their roles in decision formation.
decision. Why the evidence and not the DV? One challenge is to understand the appar-
To make a decision, the brain must compare f 2 ent redundancy. For example, memory traces
with f 1. This comparison cannot occur until of f 1 and f 1/f 2 comparisons are both found
f 2 is applied, and it must incorporate informa- in S2, VPC, MPC, and dlPFC. Do the subtle
tion about f 1 that has been held in working differences in how those computations man-
memory. S1 responses reflect the f 1 stimulus ifest in the different brain areas indicate sub-
during the first interval and the f 2 stimulus tly different roles in these processes? Or is
during the second interval (Figure 3c). They there simply a continuous flow of informa-
therefore do not provide the comparison. tion through these circuits, such that each per-
Activity in several brain areas, including forms a unique role but has continuous access
the second somatosensory cortex (S2; Romo to the computations performed by the other
et al. 2002) and the dorsolateral prefrontal circuits?
cortex (dlPFC, or Walker area 46; Brody et al. Another challenge lies in linking neural ac-
2003, Romo et al. 1999) but especially the tivity to a particular element of the decision
medial and ventral premotor cortices (MPC process. For example, in several brain regions,
544 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
but especially in the dlPFC, delay-period ac- further established a causal link for MT ac-
tivity resembles a working memory trace of tivity and task performance (Ditterich et al.
the evidence provided by f 1. An alternative 2003; Newsome & Paré 1988; Salzman et al.
RDM: random-dot
explanation is that the delay-period activity 1990, 1992). motion
represents the DV: a prediction of the deci- Two aspects of the task facilitate the study
sion, given f 1. Specifically, if f 1 is low, the of the neural mechanisms of decision forma-
answer is (or might seem) likely to be f 2 > f 1. tion. The first aspect is that the time needed
If f 1 is high, f 2 < f 1 might seem more likely. to make the decision is particularly long for
This DV would later be updated by the f 2 perceptual tasks, typically many 100s of ms.
evidence. Consistent with this idea, manipu- Thus researchers have characterized neural
lating properties of the f 1 stimulus can bias correlates of the decision process as it un-
choices under some conditions (Luna et al. folds in time. For a version of the RDM
2005). Contrary to this idea, dlPFC delay- task in which motion viewing time (t) is con-
period activity reflects f 1 when using a stim- trolled by the experimenter, performance im-
√
ulus set in which the value of f 1 cannot be proves as roughly t (Figure 4b), the rela-
used to predict f 2 (Romo et al. 1999). How- tion expected if at each successive moment the
ever, little is known about the task-related ex- brain acquired (i.e., integrated) an indepen-
pectations that might be represented during dent sample of noisy evidence. Performance
the delay period. on other perceptual tasks, including even a

version of the RDM task, can show little or
Random-dot motion (RDM) direction dis- no improvement with prolonged viewing du-
crimination. Similar to the VTF paradigm, ration (Ludwig et al. 2005, Uchida et al. 2006,
the RDM paradigm (Figures 4 and 5) was Uka & DeAngelis 2003, Watson 1986). In this
developed to study the relationship between sense the RDM task may be less representa-
sensory encoding and perception. Unlike the tive of perception than of cognitive decision
VTF task, the RDM task requires a single making, which can involve multiple sources of
stimulus presentation and thus eliminates the evidence acquired over a flexible time scale.
need for working memory. The monkey de- The second benefit of the RDM task is
cides between two possible (opposite) direc- the imposed link between the direction de-
tions of motion that are known in advance. cision and a particular course of action, the
Task difficulty is controlled by varying the eye-movement response. This link enables in-
percentage of coherently moving dots. The vestigators to treat the decision as a prob-
direction decision is typically indicated with lem of movement selection. Thus, the search
an eye movement. for the DV has focused on parts of the brain
The evidence used to form the direction involved in the selection and preparation of
decision has been traced to neurons in the eye movements, including the lateral intra-
middle temporal area (MT/V5) tuned for the parietal area (LIP), superior colliculus (SC),
direction of visual motion. SDT analyses of frontal eye field (FEF), and dlPFC (Horwitz &
the strength and variability of MT responses Newsome 1999, 2001; Kim & Shadlen 1999;
provided a foundation for understanding be- Shadlen & Newsome 1996, 2001).
havioral accuracy (Britten et al. 1992, 1993; In one experiment (Figure 4), motion
Shadlen et al. 1996). Choice probabilities in- viewing was interrupted at a random time
dicated that individual MT neurons weakly during decision formation by turning off the
but significantly predict the monkey’s direc- RDM stimulus and applying a brief electrical
tion decisions, including errors (Britten et al. current to the frontal eye field (FEF) (Gold &
1996). Lesion and microstimulation studies, Shadlen 2000, 2003). The microstimulation
exploiting the systematic organization of MT caused a short-latency saccade whose ampli-
with respect to motion location and direction, tude and direction were determined by the

ANRV314-NE30-21 ARI 21 May 2007 13:44
a Evoked
Voluntary
Motion saccade
saccade
Fixation
Time
b c
25.6
Threshold (% coh)
Y position (º)
Up
Fix
12.8 0 Down
-5
6.4
100 200 400 800 -5 0 5 10

Viewing duration (ms) X position (º)
d 1.5 51.2% coh

Amount of deviation (º)
25.6%
6.4%
1.0
0%
0.5
0
100 300 500
Viewing duration (ms)
Figure 4
Representation of an evolving DV by the motor system. (a) Interrupted direction discrimination task.
The monkey decides the net direction of motion, here shown as up versus down. Task difficulty is
governed by the fraction of dots that move coherently from one movie frame to the next (% coherence).
The motion viewing is interrupted prematurely, and on a fraction of trials, a brief current is applied to the
FEF to evoke a saccade. The monkey makes a second, voluntary movement to a choice target to indicate
his decision. (b) Decision accuracy improves as a function of motion-viewing duration. Psychophysical
√
threshold is defined as the motion coherence supporting 82% correct. Threshold falls by t (slope of
line fit on log-log plot = –0.46; 95% CI: –0.59 to –0.33). These data are from trials in which no
stimulation occurred. Similar data were obtained on stimulated trials. (c) Examples of eye movement
trajectories. Fixation point is at the origin. The two larger circles are the choice targets. The random-dot
stimulus (not shown) was centered on the fixation point. The symbols mark eye position in 2-ms steps.
FEF stimulation during fixation, in the absence of motion and choice targets, elicited a rightward saccade
(trace marked “Fix”). Stimulation while viewing upward and downward motion induced saccades that
deviated in the direction of the subsequent, voluntary eye movements. (d ) The average amount of
deviation depends on motion strength and viewing time. The amount of deviation toward the chosen
target was estimated using the evoked saccades from 32 stimulation sites (14,972 trials). This result shows
that the oculomotor system is privy to information about the evolving decision, not just the final outcome
of the decision process. Adapted from Gold & Shadlen (2000, 2003) with permission.
546 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
site of stimulation, a defining feature of the Lynch et al. 1985; Paré & Wurtz 1997). This
FEF (Bruce et al. 1985, Robinson & Fuchs area has been implicated in other high-order
1969). This evoked saccade tended to deviate processes involved in the selection of saccade
RT: reaction time
in the direction governed by the eye move- targets, including working memory, alloca-
MST: medial
ment associated with the monkey’s ultimate tion of attention, behavioral intention, spa-
superior temporal
choice. The amount of deviation, even when tial inference, and representation of bias, re- area
measured early in the decision process, par- ward, expected value, and elapsed time (Assad
alleled the evolution of a DV that explained & Maunsell 1995, Chafee & Goldman-Rakic
accuracy as a function of motion strength and 2000, Dorris & Glimcher 2004, Eskandar
viewing time. This result is inconsistent with & Assad 1999, Friedman & Goldman-Rakic
the notion that a central decision maker com- 1994, Janssen & Shadlen 2005, Leon &
pletes its operation before activating the mo- Shadlen 2003, Platt & Glimcher 1999, Sugrue
tor structures to perform the necessary ac- et al. 2004). Moreover, neural activity in
tion. Instead, it implies that, at least under LIP—particularly in its ventral subdivision,
some conditions, information flow from sen- termed LIPv (Blatt et al. 1990, Lewis & Van
sory neurons to motor structures is more or Essen 2000b)—reflects decision formation on
less continuous (Spivey et al. 2005). a fixed-duration version of the RDM task
To measure the correspondence more pre- (Shadlen & Newsome 1996, 2001).
cisely in time of neural activity with elements Figure 5c,d illustrates the responses of
of the decision process, a reaction-time (RT) LIP neurons during the RT paradigm. In
version of the RDM task was developed. This these experiments, one of the choice targets
task allowed the monkey to indicate its choice (Tin) is in the response field (RF) of the
as soon as a commitment to one of the alterna- LIP neuron; the other target (Tout) and the
tives is reached. Figure 5b shows examples of RDM stimulus lie outside the neuron’s RF
choice and RT functions for a monkey. Mean (Figure 5a). Thus, these neurons are stud-
RT increases as task difficulty increases. For ied under conditions in which they are apt to
easy stimuli, RT varies with stimulus strength signal the monkey’s choice (via the associated
even when choice accuracy is perfect. It ul- action). What is more interesting is that their
timately approaches an asymptote that rep- activity reflects the decision process that leads
resents time that is not used on the decision to that choice.
per se. This nondecision time (328 ms in this Aligning the responses to stimulus onset
data set) includes visual and motor latencies (t = 0 on the left side of Figure 5c) pro-
and possibly other processing stages that are vides a glimpse into the brain’s activity in
less understood (see below). For difficult stim- the epoch when the animal is forming the
uli, the nondecision time is relatively short decision but has yet to commit overtly to a
compared with the RT, implying long deci- choice. Initially, there is a brief dip in the
sion times. In contrast, this nondecision time firing rate followed by a rise in activity that
often takes up the lion’s share of RT for sim- is independent of the direction and strength
pler tasks, an important caveat for interpreting of motion or the monkey’s ultimate choice.
many RT studies. Then, after ∼220 ms, the average response
Studies of neural mechanisms underlying begins to reveal differences in the evidence
the decision process on the RT task have fo- and outcome of the decision. On trials that
cused on area LIP. LIP is anatomically po- end in a Tin choice, the firing rate rises like a
sitioned midway through the sensory-motor ramp, on average. On trials that end in a Tout
chain, with inputs from MT and MST and choice, the firing rate meanders or tends to
outputs to the FEF and SC (Andersen et al. decline. This dependence on choice is evident
1990, 1992; Asanuma et al. 1985; Blatt et al. even when the stimulus is ambiguous (0%
1990; Fries 1984; Lewis & Van Essen 2000a; coherence).

ANRV314-NE30-21 ARI 21 May 2007 13:44
Aligning the responses to saccade initiation strength, they achieve a common level
tion (Figure 5c, right) reveals a correlate of of activity ∼70 ms before saccade initiation
commitment: a threshold rate of firing be- (arrow in Figure 5d ). Thus the decision
fore Tin choices. When separated by motion process appears to terminate when the neu-
strength, the curves overlap considerably just rons associated with the chosen target reach
prior to the saccade and thus make it im- a critical firing rate. When the monkey
possible to identify a single point of conver- chooses Tout, another set of neurons—the
gence because each motion strength leads to ones with the chosen target in their RFs—
a broad distribution of RTs. When these same determines the termination of the decision
responses are grouped by RT instead of mo- process.
b Behavior
1
a Saccade
Percent correct
0.8
Motion
0.6
Targets
0.4
Fixation RF 900
Mean RT (ms)
800
)
RT
e( 700
im
nt 600
tio
ac
Re 500
400
0 2.5 5 10 20 40
Motion strength (% coh)
c
Motion
Eye
70 strength movement
Motion
on 51.2
12.8
0 Select Tin
60
d RT (ms)
400
70
500
Firing rate (sp s–1)
600
50 700
60 800
900
Select Tout
40 50
40
30
45
MT 30
20 5
0 200 400 600 800 -200 0 –1000 –500 0
Time (ms) Time from saccade (ms)
548 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
The pattern of LIP activity matches pre- Studies using electrical microstimulation
dictions of diffusion/race models (Figure have helped to establish further the causal
2b,c). The coherence-dependent rise appears roles of MT and LIP in representing the ev-
to reflect an accumulation of noisy evidence. idence and DV, respectively (Ditterich et al.
This evidence comes from a difference in ac- 2003, Hanks et al. 2006). Earlier studies us-
tivity of pools of MT neurons with oppo- ing a fixed-duration version of the RDM task
site direction preferences (Figure 5c, shaded showed that microstimulation of direction-
insert), which is thought to approximate the selective MT neurons causes monkeys to bias
associated logLR (Gold & Shadlen 2001). their decisions in favor of the preferred direc-
However, some caveats should be noted. The tion of the stimulated neurons (Salzman et al.
dashed curves of Figure 5c (right) do not end 1990, 1992). This result could be attributed
in a common lower bound and instead look to a perturbation of the evidence, the DV, or
like the average of paths of evidence for, say, both. The RT task makes it possible to distin-
right when the left choice neurons win in a guish these alternatives (Figure 6). MT mi-
race model (Mazurek et al. 2003). Also, the crostimulation has strong effects on all sub-
DV is represented only from ∼220 ms after sequent choices and RTs, biasing the monkey
motion onset until ∼70–80 ms before saccade toward more, faster choices in the preferred
initiation. Reassuringly, these times taken to- direction of the stimulated neurons and fewer,
gether nearly match the nondecision time slower choices in the opposite direction. This
from fits of the diffusion model to the choice- finding is consistent with an additive offset of
accuracy behavioral data. Whereas the initial the evidence, which causes an increased rate
∼220-ms latency is long in comparison with of rise of its integral, the DV. Conversely, LIP
latencies of neural responses in MT (which microstimulation has small effects on choice
show direction selectivity ∼100 ms after onset and modest effects on RT, which is consistent
of a RDM stimulus; Figure 5c, insert) and LIP with an additive offset of the DV that does
(which can indicate the presence of a target in not affect its rate of rise but rather pushes it
<60 ms; Bisley et al. 2004), behavioral mea- closer to (or further from) the threshold for
surements also indicate that evidence does not terminating the decision (Hanks et al. 2006).
affect decisions until after this critical waiting A further test of the idea that LIP neu-
time (Huk & Shadlen 2005, Kiani et al. 2006). rons represent the DV used weak motion
←−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
Figure 5
Neural mechanism of a decision about direction of motion. (a) Choice-reaction time (RT) version of the
direction discrimination task. The subject views a patch of dynamic random dots and decides the net
direction of motion. The decision is indicated by an eye movement to a peripheral target. In the RT task,
the subject controls the viewing duration by terminating each trial with an eye movement whenever
ready. The gray patch shows the location of the response field (RF) of an LIP neuron. (b) Effect of
stimulus difficulty on accuracy and decision time. Solid curves are fits of the diffusion model (see Palmer
et al. 2005), which accounts simultaneously for choice and decision time. (c) Response of LIP neurons
during decision formation. Average firing rate from 54 LIP neurons is shown for three levels of difficulty.
Responses are grouped by motion strength and direction of choice, as indicated. Left: The responses are
aligned to onset of random-dot motion. Averages are shown during decision formation (curves truncated
at the median RT or 100 ms before the eye movement). Shaded insert shows average responses from
direction selective neurons in area MT to motion in the preferred and antipreferred directions. After a
transient, MT responds at a nearly constant rate. Right: The responses are aligned to the eye movement.
The LIP firing rates approximate the integral of a difference in firing rate between MT neurons with
opposite direction preferences. (d ) Responses grouped by RT. Only Tin choices are shown. All trials
reach a stereotyped firing rate ∼70 ms before saccade initiation (arrow). Adapted with permission from
Shadlen et al. (2006) and Roitman & Shadlen (2002); insert from online database used in Britten et al.
(1992), http://www.neuralsignal.org database nsa2004.1.

ANRV314-NE30-21 ARI 21 May 2007 13:44
a Stimulate rightward MT neurons
Relatively large effect on choice and RT:

equivalent to added rightward motion
Reaction time rightward choices

Proportion
Momentary evidence in MT DV in LIP
Bound for right choice
+A
Mean
Evidence 0 stim adds cumulatively
for right 0
(R – L) 0
Motion strength
Strong Strong
–A leftward rightward
Bound for left choice
b Stimulate right choice LIP neurons

Small effect on choice,
modest effect on RT:
not equivalent to added rightward motion
Reaction time rightward choices
Proportion
Momentary evidence in MT DV in LIP

Bound for right choice
+A
mean
Evidence 0
for right Stim adds constant
0
(R – L)
0
–A Motion strength
Bound for left choice Strong Strong
leftward rightward
Figure 6
Effects of microstimulation in MT and LIP. In both areas microstimulation (red curves) causes a change in
both choice and RT. The schematic shows the consequences of adding a small change in spike rate to the
evidence or to the DV. The graphs on the right are theoretical results obtained using the bounded
diffusion model. They resemble the pattern of data in Hanks et al. (2006). (a) MT microstimulation
mimics a change in stimulus strength (evidence). (b) LIP microstimulation mimics an additive offset to
the DV (or, equivalently, the height of the bounds).
550 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
perturbations in the background of the RDM what happens when it is reached are unknown.
display (Huk & Shadlen 2005). Each pertur- Because the criterion controls the trade-off
bation was a subtle boost or decrement in between speed and accuracy (Palmer et al.
motion energy lasting just 100 ms. The ef- 2005), parts of the basal ganglia sensitive to
fects on RT and choices were consistent with both reward and movement onset have been
a process of integration. The motion pulses suggested as possible substrates (Lo & Wang
affected choices that occurred up to 800 ms 2006). An alternative possibility is that a sin-
after the pulse, and they affected RTs through gle neural circuit can represent the DV and
a sustained effect on the DV (like MT micro- its conversion to a binary choice, which would
stimulation). LIP neurons also registered suggest that the criterion is an intrinsic prop-
these brief motion perturbations with long- erty of LIP, FEF, or SC (Machens et al. 2005,
lasting changes in firing rate consistent with a Wang 2002, Wong & Wang 2006).
process of integration.
We do not know how or where this in- Heading discrimination. Optic flow is the
tegral is computed. It might be computed pattern of motion that occurs when we move
in LIP itself, or it might be computed else- through the environment (Gibson 1950). A
where and merely reflected in LIP. A model natural candidate for the momentary evidence
of LIP can achieve integration by mix- used to infer heading direction from optic flow
ing feedback excitation—using N-methyl-D- is in the medial superior temporal cortex (area
aspartate (NMDA) channels with relatively MST). MST neurons are tuned for expansion,
long conductance times—within neuronal rotation, translation and other large-field mo-
pools that share a common RF with an in- tion patterns that comprise optic flow (Duffy
hibitory antagonism between pools represent- & Wurtz 1991, 1995, 1997; Graziano et al.
ing opposite directions (Wang 2002). This 1994; Lagae et al. 1994; Saito et al. 1986;
hybrid of biophysical and large-scale neural Tanaka et al. 1986; Tanaka & Saito 1989).
modeling was originally designed to simulate Indeed, for a one-interval heading direction-
working memory. That the model can also discrimination task using a RDM flow field,
form a DV in a manner consistent with many signal-to-noise measurements of MST activ-
aspects of the physiological results, includ- ity correlate with behavioral sensitivity in a
ing deviations from perfect integration (Wong manner similar to results from MT using the
et al. 2005, Wong & Wang 2006), suggests RDM direction task (Britten & van Wezel
that this kind of persistent activity might serve 2002, Heuer & Britten 2004).
multiple roles in the brain. Indeed, the ques- However, microstimulation experiments
tion of how neurons or neural circuits can in- have provided only weak evidence for a causal
tegrate is not limited to the study of decision role of MST neurons for the heading deci-
making and working memory but extends to sion (Britten & van Wezel 1998). Microstim-
motor control and navigation as well (Major ulation can cause the monkey to bias choices
& Tank 2004). Progress in this area is likely for one heading direction, but often it is in
to shed light on cognitive functions that oper- the direction opposite the preferred heading
ate on time scales longer than biophysical and of the stimulated neurons. This might be ex-
signaling time constants in single cells. plained by a lack of a clustered organization of
We also know little about how the criterion neurons tuned to similar optic flow patterns
is applied to the DV. Neurons that achieve a (Britten 1998). A more tantalizing explana-
threshold level of activity in anticipation of tion comes from the fact that many MST neu-
a saccadic eye movement on RT tasks have rons receive visual and vestibular inputs that
been found not just in LIP but also in FEF both contribute to heading sensitivity (Gu
and SC (Hanes & Schall 1996, Ratcliff et al. et al. 2006) but, for about half these neurons,
2003). However, how this criterion is set and have opposite direction preferences. Thus,

ANRV314-NE30-21 ARI 21 May 2007 13:44
microstimulation intended to bias judgments a RDM stimulus viewed stereoscopically

in a particular direction using visual cues showed that the sensitivity of MT neurons
might instead activate a local circuit domi- to noisy perturbations in disparity rivaled the
nated more by vestibular tuning in the op- behavioral sensitivity of the monkey (Uka &
posite direction. Consistent with this idea, DeAngelis 2003). Electrical microstimulation
choice probabilities (the weak correlations be- experiments, aided by the clustered organiza-
tween the variable discharge of MST neurons tion of MT with respect to disparity, further
and the trial-to-trial variations in the mon- established their causal role in providing ev-
key’s decisions) also appear to depend on the idence for decisions about depth (DeAngelis
vestibular tuning of the neuron (DeAngelis et al. 1998, 1999; Krug et al. 2004).
et al. 2006). Can MT neurons that provide evidence
We do not know where the DV is repre- about depth provide, in a different con-
sented for this task. One possibility is MST it- text, evidence about direction? To test this,
self, which contains many neurons with spik- monkeys were trained on a RDM direction-
ing activity that persists even in the absence discrimination task in which the dots could
of a stimulus. However, the lack of build- be presented with task-irrelevant disparities.
up activity as the decision is formed and Electrical microstimulation at sites with weak
weak choice probability support the idea that disparity tuning tended to have the largest
MST represents only the momentary evi- effects on performance. Microstimulation at

dence (Heuer & Britten 2004). Another pos- sites with strong disparity tuning had weaker
sibility is the ventral intraparietal area (VIP), effects on performance, even when those sites
which receives direct input from MT and were strongly tuned for direction (DeAngelis
MST and contains many neurons with re- & Newsome 2004). The results suggest that
sponse properties similar to MST on op- the DV tended to discount MT neurons
tic flow and heading tasks (Bremmer et al. whose responses were sensitive to disparity,
2002a,b; Zhang et al. 2004; Zhang & Britten possibly because that variable was irrelevant
2004). However, as in MST, it is difficult to to the direction task. This kind of context-
tell if VIP represents a DV or the momen- dependent read-out implies a critical role for
tary evidence used by other brain structures to learning in establishing the flow of informa-
decide whether, say, an object is nearing the tion from neurons that represent the evidence
head and needs to be avoided (Colby et al. to neurons that form the DV (Freedman &
1993, Duhamel et al. 1998, Graziano et al. Assad 2006, Law & Gold 2005).
1997). In our view, the DV for the heading Less is known about the role of MT neu-
task is likely to be represented in structures rons in decisions that require both depth
that provide high-level control of the behav- and motion information. Nevertheless, a task
ioral (eye movement) response; e.g., area LIP. requiring a decision about the direction of
If the monkey were trained to reach for but- rotation of a transparent, sparsely textured
tons, likely candidates would be parietal and cylinder has provided a striking result for
prefrontal cortical areas that provide analo- MT choice probabilities. An observer view-
gous control of reaching movements. ing such a cylinder sees the texture (e.g.,
dots) on the front and back surfaces move
Disparity discrimination. MT neurons are in opposite directions. The opposing motion
selective for not only motion direction but gives rise to a distinct pattern of firing rates
also the binocular disparity of images pre- among direction-selective MT neurons that
sented to the two eyes, a cue for depth can be unambiguously associated with trans-
(Maunsell & Van Essen 1983, Uka & parency. However, if not for the depth cues,
DeAngelis 2006). Recent studies using a the cylinder can appear to rotate in a clock-
one-interval depth-discrimination task with wise or counterclockwise direction depending
552 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
on whether the two motion directions are as- order of psychophysical threshold. This find-
sociated with the front and back or vice versa ing provided the first direct evidence that face-
(Ullman 1979). selective IT neurons play a causal role in the
MT neurons that are both disparity and perception of faces. It suggests that IT activity
direction selective furnish evidence to remove represents the evidence used to solve the task
this ambiguity (Bradley et al. 1998). For exam- but does not rule out the possibility that IT
ple, an MT neuron preferring near disparities represents a DV or even the outcome of the
and rightward motion responds best when the perceptual categorization (Sheinberg & Lo-
nearer plane is moving to the right. Such a gothetis 1997, 2001).
neuron provides positive evidence for the in- In a related study, measurements of blood
terpretation that a vertical cylinder is rotating oxygen level differences (BOLD) in fMRI
in accordance with a right-hand rule (front were used to identify correlates of a DV that
surface to the right) (Dodd et al. 2001). A reads out object categorization evidence from
strong trial-to-trial correlation exists between IT (Heekeren et al. 2004). Human subjects
the variable discharge of these neurons and were trained to perform a one-interval dis-
the monkeys’ rotation judgments (Krug et al. crimination between faces and houses masked
2004). In fact, this choice probability is larger by noise. The two sets of stimuli were used
than for any of the decision tasks reviewed in because distinct regions of IT are activated
this article, leading us to suspect that it is a for unmasked images from the two categories
sign of feedback from elements in the brain (Haxby et al. 1994, Kanwisher et al. 1997). A
that have rendered a decision, as opposed to candidate DV was found in the dlPFC. Activ-
feedforward variations of noisy evidence that ity in this area was strongest when the sensory
underlie difficult decisions near psychophysi- evidence was strongest and tended to covary
cal threshold. with the magnitude of the difference in the
BOLD signal measured on single trials in the
Face/object discrimination. Neurons se- “face” and “house” areas. Such fMRI experi-
lective for images of faces and other com- ments provide an essential link between mon-
plex objects are found in the ventral stream key neurobiology and human brain function,
“what” or “vision for perception” pathway although they lack the spatial and temporal
(Goodale & Milner 1992, Ungerleider & resolution to characterize fully the neuronal
Mishkin 1982). Recent studies have begun to dynamics that distinguish evidence from the
examine how these neurons contribute to per- DV.
ceptual decisions by comparing brain activity Electroencephalography (EEG), which is
and behavior (Allred et al. 2005, Baylis et al. not burdened by the same temporal resolu-
2003, Dolan et al. 1997, Freedman et al. 2002, tion problem, has been measured in human
2003; Grill-Spector et al. 2000, Op de Beeck subjects for similar one-interval categoriza-
et al. 2001, Rainer et al. 2004). In one study, tion tasks requiring a discrimination be-
monkeys performed a one-interval discrimi- tween pictures of faces and pictures of cars
nation of face versus nonface images masked (Philiastides et al. 2006, Philiastides & Sajda
by white noise (Afraz et al. 2006). Electri- 2006, VanRullen & Thorpe 2001). Two sig-
cal microstimulation applied during stimulus nal kernels could best differentiate the car and
viewing to clusters of neurons in the infero- face stimuli on single trials. An early potential
temporal (IT) cortex that showed a preference that appeared ∼170 ms after stimulus onset
for faces biased decisions toward face versus was selective for faces and only weakly predic-
nonface. The magnitude of the bias was com- tive of errors, a possible correlate of sensory
parable to that found using MT microstimu- evidence. A later potential appearing ∼300 ms
lation on direction and disparity tasks, equiv- after stimulus onset appeared to reflect the
alent to a change in stimulus strength on the difficulty of the decision linking the sensory

ANRV314-NE30-21 ARI 21 May 2007 13:44
evidence to the subject’s choice, a quality of for combined psychophysical and physiologi-
the DV. However, localization of the signals is cal experiments on olfactory decision making
hampered by the limitations of EEG record- in rodents.
ing and the powerful technique used in this
study to combine signals across electrodes. Detection. Detection experiments are
the archetypical SDT paradigm, yet they
Olfactory discrimination. To date, there present serious challenges to neurobiologists
have been few experiments on the neurobi- trying to understand the underlying decision
ology of decisions in animals beside the mon- process. According to SDT, detection begins
key. However, the availability of techniques with a sample of evidence generated by
to study molecular and cellular mechanisms either a signal plus noise or noise alone. The
in rodents hints at tremendous possibilities DV based on this sample is monotonically
if they could be trained on decision-making related to the LR in favor of h 1 : “S present”
tasks. Several recent studies have begun to and against h 2 : “S absent”. The decision
make progress by using tasks that exploit two is H1 : “Yes” if the DV exceeds a criterion,
of their natural strengths: foraging behavior which is set to achieve some desired goal,
and olfactory processing. and H2 : “No” otherwise. This procedure
In one study, rats were trained to discrim- seems straightforward but glosses over an
inate between a pair of odors using a one- important question: When should the DV
interval task with mixtures of the two odors incorporate evidence if there is temporal
(Uchida & Mainen 2003). The patterns of ac- uncertainty about when the stimulus will
tivation in the olfactory bulb were distinctive appear? Obtaining samples at the wrong time
when the stimulus was behaviorally discrim- might miss the signal [i.e., high P (H2 | h 1 )].
inable and less distinctive otherwise, a first In contrast, accumulating samples over time
step toward identifying signals comprising the will accumulate noise alone in all epochs
sensory evidence. However, the rats appeared that do not contain the signal, causing more
to form their decisions in a single sniff and misses in signal-present conditions and more
did not benefit from more time (Kepecs et al. false alarms [P(H1 | h 2 )] in signal-absent con-
2006, Uchida et al. 2006). Such a short time ditions. The consequence of these errors will
scale clouds the distinction between evidence be a loss of sensitivity (Lasley & Cohn 1981).
and DV and thus exposes possible limitations This problem has several possible solu-
of this model for the study of decision making. tions. One is to use an integrator that leaks,
However, it seems unlikely that rats can- causing irrelevant information to affect the
not accrue evidence in time. A recent study decision process for only a limited amount of
demonstrated a trade-off between speed and time (Smith 1995, 1998). Other approaches
accuracy in a similar odor-discrimination task include taking a time derivative of the evi-
(see also Abraham et al. 2004, Rinberg et al. dence to identify changes or using knowledge
2006). A window of integration of ∼400 of the spatial and temporal structure of the
ms was identified, which rivals the inte- stimulus to guide a more directed search for
gration times for monkey’s performing the the evidence. Some studies in the psychophys-
RDM direction-discrimination task. Com- ical literature hint of such mechanisms (e.g.,
bined with the recent discovery of the ol- Henning et al. 1975, Nachmias & Rogowitz
factory receptor genes (Buck 1996, Buck & 1983, Verghese et al. 1999, Schrater et al.
Axel 1991) and subsequent progress in under- 2000), but to date there is little understanding
standing the stable, topographic organization of their general role in detection.
of the olfactory bulb (e.g., Rubin & Katz 1999, Despite these challenges, several recent
Meister & Bonhoeffer 2001, Mombaerts et al. studies have begun to shed light on de-
1996) these results suggest a promising future tection mechanisms. In one study, monkeys
554 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
a
Coherent
Time motion
at one
b
1
patch
“yes” responses
(750 ms)
Proportion
0% Coherent
(500 - 8000 ms)
Fixation
point
0
Receptive Static cue 0 10 20 30
field Motion coherence
MT VIP
c d
411 406
377 374 440
35 446 35 309 483
492 549 702
312 561

694
Threshold
Missed
Missed False alarms
False alarms
10 10
0 600 0 600
Time (ms) Time (ms)
Figure 7
Motion detection. (a) Detection task. The monkey views a RDM stimulus without any net direction of
motion and must release a bar when the motion becomes coherent. Task difficulty is controlled by the
intensity of the motion step (% coherence). (b) Probability of deciding Yes plotted as a function of
stimulus intensity. (c, d ) Average firing rates from neurons in MT and VIP. The responses are aligned to
onset of the motion step and grouped by RT. Yes decisions are predicted by a rise in firing rate in both
areas. The horizontal dashed line is a criterion derived to match the animal’s performance. The missed
detections (motion step, but decision = No) are explained by a failure of the firing rate to reach this
level. According to SDT, the false alarms (no motion step, but decision = Yes) should reach this level,
but with responses aligned to the lever release at 750 ms, they do not; a possible explanation is a variable
relationship between the end of the decision process and the time of the lever release. Adapted with
permission from Cook & Maunsell (2002a,b).
were trained to detect the onset of partially likely to result, in part, from the temporal
coherent motion in a RDM display (Figure 7) uncertainty problem described above. Never-
(Cook & Maunsell 2002b). Activity in both theless, population analyses using a leaky ac-
MT and VIP was correlated with trial-by- cumulator model provide some insight into
trial detection performance and RT. However, the decision process. For MT, population re-
single-unit responses were far less reliable de- sponses on hit trials deviated from baseline
tectors of the stimulus than were the mon- at a time that was closely coupled with mo-
key subjects. This finding contrasts results tion onset and then rose steadily with a rate
from discrimination experiments and seems of rise that was correlated with RT. Responses

ANRV314-NE30-21 ARI 21 May 2007 13:44
on miss trials were similar but failed to attain These results suggest that MPC represents
the same level of activation as hits. In con- the DV that converts the evidence into a
trast, responses on false alarm trials deviated choice. However, inspection of Figure 8e
little from baseline. These results suggest that suggests that different stimulus intensities
MT provides a preliminary form of evidence. cause primarily differences in latency, the ex-
In contrast, VIP population responses were pected behavior of a neuron that simply re-
less coupled to motion onset and were more sponds stereotypically after the decision has
stereotypical with respect to the behavioral re- ended. Nevertheless, even if MPC neurons
sponse, suggesting they represent either the represent the decision outcome and not the
DV or simply the outcome of the decision DV, they do not do so in a trivial matter.
process. The responses are not simply associated with
Similar detection experiments were con- the button press. Moreover, electrical mi-
ducted using the VTF stimulus (Figure 8) crostimulation of MPC without presentation
(de Lafuente & Romo 2005). As for the dis- of the tactile stimulus leads to similar per-
crimination task, S1 activity appears to rep- ceptual reports. These results have been used
resent the sensory evidence and not the DV to support the bold (and currently unverifi-
or choice. Increasingly intense stimuli (deeper able) assertion that MPC responses provide
skin depressions) lead to high firing rates that a neural correlate of the subjective percep-
are more easily differentiated from neural action of the tactile stimulus (see also Ress et al.
tivity when no stimulus is present (Figure 8c, 2000, Ress & Heeger 2003 for a similar asser-
d ). Moreover, trial-to-trial variations in the tion about activity in early visual cortex, mea-
responses to a weak stimulus are correlated, sured using fMRI, for a contrast-detection
albeit weakly, with the monkey’s yes and no task).
decisions. This weak choice probability is sim-
ilar in magnitude to that found for MT neu-
rons on the RDM direction-discrimination Simple Motor Latencies: Deciding
task (Britten et al. 1996). When to Initiate an Action
In contrast to S1, MPC responses are mod- In this section, we consider easy decisions that
ulated by not only stimulus intensity but also do not tax perceptual processing but instead
the monkey’s behavioral report (Figure 8e). simply trigger a movement. Commonly used
−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−→
Figure 8
Vibrotactile detection. (a) Task. The VTF probe is placed on the finger pad. After a random delay, a
20-Hz sinusoidal indentation is applied on half the trials. The monkey must decide whether the stimulus
was present. Detection is indicated by removing a finger of the other hand from a key and pushing a
button. (PD, probe down; KD, key down; PU, probe up; MT, movement to the response button).
(b) Psychometric function. The monkey is more likely to decide Yes at larger vibration amplitudes. The
false alarm rate is ∼8%. (c) Response of a typical S1 neuron. The top half of the raster shows trials in
which vibration was applied. Red marks indicate missed-detection errors. The bottom half of the raster
shows trials in which no stimulus was shown. Red marks indicate false-alarm errors. (d ) Distributions of
firing rates in S1 during the stimulus period. Each curve represents a frequency distribution associated
with vibration at one intensity (indicated by color). The vertical line shows a possible criterion that the
brain could apply to decide yes or no. (e) Response of a typical MPC neuron. Same conventions as in
panel c. The neuron responds strongly when the monkey reports Yes even when there is no stimulus
present (false alarms). Notice that on some trials on the top half of the graph, the neuron seems to
indicate a detection decision before the stimulus is applied. These trials appear as correct detections, but
they are probably lucky mistakes. ( f ) Average firing rate of S1 and MPC neurons as a function of
stimulus intensity. The firing rates are for the epoch of stimulus presentation. Only correct responses
(Yes when amplitude >0) are included. Adapted with permission from de Lafuente & Romo (2005).
556 Gold · Shadlen

ANRV314-NE30-21
ARI
21 May 2007
13:44
www.annualreviews.org • Decision Making

557
ANRV314-NE30-21 ARI 21 May 2007 13:44
tasks require a saccadic eye movement to a Westheimer 1954). By comparison, the time
visual target that might appear at an unpre- it takes to register visual input in visuo-motor
dictable time or location but is unambiguous areas such as LIP, FEF, and SC can be <50 ms
once it appears. Insights into the underlying (Bisley et al. 2004, Pouget et al. 2005,
decision process have come primarily by an- Schmolesky et al. 1998) and to elicit saccadic
alyzing the distributions of latencies from vi- eye movements via electrical microstimula-
sual instruction to saccade initiation. tion of FEF or SC is <40 ms (Bruce et al. 1985,
Motor latencies for these tasks average Robinson 1972, Robinson & Fuchs 1969,
∼200 ms but can range from ∼90 ms to Schiller & Stryker 1972, Tehovnik 1996). To-
>400 ms (Carpenter 1988, Sparks 2002, gether, these measurements do not account
for much of the length and variability of RTs,
prompting speculation that the underlying
a Reaction time
decision process is more sophisticated than
distribution
a mere trigger, perhaps necessarily including

Threshold ST
procrastination to allow for censorship of an
Rate of
rise (r) action and indeterminacy to make actions less
predictable (Carpenter 1981).
Start level S 0
Latency A simple model can account for these mo-
tor latencies (Figure 9). Latency distributions

Infinite-time typically have a single mode with a longer
intercept
tail toward larger values. The distribution
b of reciprocal latency thus appears Gaussian
95 (Carpenter 1981, 1988; Carpenter &
90 Williams 1995). Accordingly, the cumulative
Percent cumulative
distribution of reciprocal latencies plotted on

frequency
Median ←−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
50
Figure 9
A simple model for “go” reaction times. (a) The
LATER model. Instead of accumulating random
10
draws of momentary evidence, the DV is a ramp
5 with a slope drawn from a Gaussian distribution.
100 200 500 1000 inf
The movement starts when the ramp reaches the
Reaction time (ms) threshold (ST ). The distribution of predicted RT
is skewed; the reciprocals would obey a Gaussian
distribution. (b) Reciprobit plot. Predicted
c cumulative distribution of RTs is a line on
Changing S0 causes reciprobit swivel transformed axes. The abscissa scale is RT−1 ; the
ordinate scale is the z-transformed (i.e., inverse
Normal) probability. Extrapolation of the graph
predicts a point intersecting the vertical line at
RT = ∞. The height of this point corresponds to
S0 the probability that the slope of the ramp is ≥0.
(c) Shape of the reciprobit plot as a diagnostic aid.
Changing μ causes reciprobit shift If the threshold changes, the median RT shifts, but
the graph retains the same extrapolated value at
RT = ∞. The graphs swivel about this point. If
there is a change in the rate of rise (mean or
standard deviation), the median RT shifts and the
μ = mean(r) graph simply translates along the abscissa. Adapted
with permission from Sinha et al. (2006).
558 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
a probability scale (a reciprobit plot) is nearly in the distributions of RT for different pri-
linear (Figure 9b). To produce this distribu- ors conformed to the “swivel” prediction, sug-
tion, assume that a DV begins at level S0 at gesting a change in the threshold (relative to
t = 0 then rises linearly with slope r. The the starting point) and not in the rate of rise
saccade is initiated when the DV reaches of the DV. The threshold appeared to change
ST at t = t1 . Critically, r is not determined as a linear function of the logarithm of the
precisely but is rather a random number prior, which suggested that the DV has units
drawn from a Gaussian distribution with log(P ). This idea is important because it im-
mean μ and standard deviation σ . Because on plies a form of probabilistic reasoning, with
any trial ri = (ST − S0 )/ti , the reciprocals of the DV representing a level of certainty that
the response latencies (1/ti ) obey a Gaussian the prepared movement should be executed.
distribution. The harmonic mean of the RT Further studies have shown that different de-
is the threshold height (ST − S0 ) divided cision strategies that favor urgency, certainty,
by μ. This model is called LATER (linear or prior expectations seem to trade off in these
accumulation to threshold with ergodic rate), units of log(P) (Reddi et al. 2003, Reddi &
which emphasizes that the DV is simply the Carpenter 2000).
accumulation of a constant. It makes testable Several physiological results provide qual-
predictions about how the distribution of RT ified support for the LATER model. First,
should change for two different mechanisms: the rate of rise (r) in the activity of FEF
changes in the threshold height (ST − S0 ) or movement cells (Bruce & Goldberg 1985)
the mean rate of rise (μ) (Figure 9c). prior to a saccade is variable and predicts RT,
This formulation has some shortcomings. whereas the final level of activation (ST ) is rel-
It typically deals with only one alternative atively fixed regardless of RT (Hanes & Schall
and short RTs, conditions under which evi- 1996). However, other studies of the FEF,
dence accumulation can be reduced to a single the SC, and primary motor cortex have found
number. Moreover, several possible sources of that prestimulus activity (S0 ) is more predic-
variability are not explored as alternate hy- tive than r of trial-to-trial variability of RTs
potheses for resulting RT distributions, in- (Connolly et al. 2005, Dorris et al. 1997,
cluding latencies outside the decision process Dorris & Munoz 1998, Everling & Munoz
(e.g., sensory and motor delays), changes in 2000, Lecas et al. 1986, Riehle & Requin
the variance of the rate of rise (r), and vari- 1993). Second, priors affect the responses of
ance in the starting point or threshold. These build-up cells in the SC (Basso & Wurtz 1997,
factors can cause real data not to conform 1998; Dorris & Munoz 1998). In general, the
neatly to the prescriptions for lines, swivels, higher the probability is of making a saccade
and shifts illustrated in Figure 9. When they to a particular target, the higher the level
do, however, the overall simplicity of LATER is of activity that occurs just before the tar-
is appealing, and it has been used to explain get appears (S0 ) and the shorter the RT is.
several interesting phenomena. Third, behavior on a countermanding task,
One study tested the idea that prior prob- in which the subject must occasionally with-
ability affects the threshold for initiating an hold a planned saccade, is consistent with a
action (Carpenter & Williams 1995). Priors race between two independent LATER pro-
were manipulated by changing the probabil- cesses representing “stop” and “go” (Hanes
ity that the target would be shown to the right & Carpenter 1999, Logan et al. 1984). These
or left of fixation. Not surprisingly, the laten- two processes have correlates in the activity
cies for eye movements to the target at the of fixation- and saccade-related cells, respec-
more probable location were faster, on aver- tively, in the FEF and SC (Hanes et al. 1998,
age, than to the other location. Differences Munoz & Wurtz 1993, Paré & Hanes 2003).

ANRV314-NE30-21 ARI 21 May 2007 13:44
Value-Based Decisions LIP that represent the DV on perceptual tasks

might represent the DV on value-based tasks,
Decisions that are based primarily on the
as well. Indeed, according to SDT and SA,
subjective value associated with each of the
value can, in principle, be treated the same
possible alternatives are the focus of the
way as priors and sensory evidence in forming
nascent field of neuroeconomics (for reviews
a decision and can be applied to either the DV
see Glimcher 2005, Sanfey et al. 2006, Sugrue
or the criterion. Thus a neural circuit that rep-
et al. 2005). A multitude of approaches, in-
resents a DV by integrating sensory evidence
cluding behavior and imaging in human sub-
on a perceptual task might be equally suited
jects and behavior and electrophysiology in
to integrate value in a different context.
non-human primates, are being used to ex-
Of course, this line of inquiry has serious
amine how the brain assigns, stores, retrieves,
challenges. Little is known about the units in
and uses value to make decisions. Here we
which value is represented in the brain, al-
limit our remarks to a few key concepts from
though some studies suggest that quantities

perceptual decisions that seem relevant to the
like expected utility—the product of subjec-
study of value-based decisions.
tive reward value and probability—might be
Neurobiological correlates of value have
represented directly (Breiter et al. 2001, Dor-
been described in orbitofrontal and cingulate
ris & Glimcher 2004, Knutson et al. 2005,
cortex and the basal ganglia, areas of the brain
Padoa-Schioppa & Assad 2006, Platt & Glim-

traditionally associated with reward-seeking
cher 1999, Schultz 2004). Moreover, unlike
behavior (Kawagoe et al. 2004; Lauwereyns
for sensory evidence on a perceptual task, the
et al. 2002; McCoy et al. 2003; Schultz 1992,
time course of a value representation is not
1998; Schultz et al. 1997; Tremblay & Schultz
easily defined or manipulated, making it dif-
1999, 2000; Watanabe 1996; Watanabe et al.
ficult to identify.
2003). Some neurons in orbitofrontal cor-
Scholars also debate whether the basic
tex appear to represent value independently
mechanisms we have described for perceptual
from evidence, choice, and action (Padoa-
decisions even apply to value-based decisions.
Schioppa & Assad 2006). Anterior cingulate
This debate is concerned with randomness.
cortex is thought to represent negative value
Choices on both perceptual and value-based
(Carter et al. 1998, Gehring & Willoughby
tasks often appear to be governed by a random
2002, Yeung & Sanfey 2004). Recent stud-
process. For perceptual tasks, this randomness
ies have shown additional representations of
is typically explained by considering the evi-
reward size or probability in parietal and pre-
dence as a mixture of signal plus noise. The
frontal association areas in the same neu-
DV and decision rule are both formulated to
rons implicated in perceptual decision mak-
minimize the effects of this noise in pursuit
ing (Kobayashi et al. 2002, Leon & Shadlen
of a particular goal. However, for value-based
1998, Platt & Glimcher 1999). In LIP and
decisions, the randomness is often assumed to
dlPFC, the representation of value seems to
be part of the decision process itself. That is, a
be dynamic, adjusted on the basis of the re-
subjective measure like utility is used to assign
cent history of choices and their consequences
the relative desirability of each choice. The
(Barraclough et al. 2004, Dorris & Glimcher
decision rule is then probabilistic: a random
2004, Sugrue et al. 2004).
selection weighed by these relative measures
It is tempting to try to analyze these phe-
(Barraclough et al. 2004, Corrado et al. 2005,
nomena in the context of the elements of a de-
Glimcher 2005, Lee et al. 2005, Sugrue et al.
cision listed above. An obvious place to start
2005).
would be to try to distinguish representations
One argument supporting this idea comes
of the DV from representations of its raw
from the theory of games (Glimcher 2005,
materials (reward/value/utility). Regions like
von Neumann & Morgenstern 1944). In
560 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
competitive games, subjects outsmart their behavior that appears, on average, to be ran-
opponents by making choices that appear to dom reflects a decision mechanism that ex-
an opponent as random. Not doing so would plicitly generates randomness or instead en-
make them vulnerable to an opponent able to acts a rule to achieve a goal but is faced with
exploit predictability. A second argument is noisy input (Herrnstein & Vaughan 1980, Lau
that randomness facilitates exploration, which & Glimcher 2005). The former is certainly
is essential for discovery of nonstationary fea- possible: We can explicitly decide to try to
tures of the environment and is thus found generate random behavior. However, the lat-
in many learning algorithms (Kaelbling et al. ter mechanism is central to our perceptual
1996). A third argument comes from the sim- abilities. It can, in principle, deal with the
ple observation that average behavior is ap- appropriate kind of input—in this case infor-
parently random under many circumstances, mation about value, expected outcomes, and
such as when following a matching law in for- dividends/costs associated with exploration—
aging (Herrnstein 1961). and produce the best possible choice. Thus,
A recent study provided an intriguing anal- a series of value-based choices might appear
ysis to support this idea (Corrado et al. 2005). random for the same reason that a series of
The study examined the sequences of choices perceptual decisions appears random under
made by monkeys on a simple oculomotor conditions of uncertainty.
foraging task (Sugrue et al. 2005). The mon- Reconsider the speed-accuracy trade-off.
keys made eye movements to one of two vi- It arises because each observation, e, is equiv-
sual targets, each of which was rewarded on a ocal; e by itself cannot be used to distinguish
dynamic, variable-interval schedule. Consis- perfectly the alternative hypotheses. Thus,
tent with previous reports, the monkeys typ- the decision maker is left in a quandary. Gath-
ically exhibited matching behavior, in which ering more observations might improve accu-
the fraction of choices to one of the targets racy, but at the cost of speed. What is the right
matched the fraction of total rewards they thing to do? The correct answer is, it depends.
earned for that choice. A model that suc- If speed is valued, gather less evidence. If ac-
cessfully described the monkeys’ behavior and curacy is valued, gather more evidence. If both
could generate realistic choice sequences was are valued, attempt to maximize quantities like
based on a deterministic, noise-free calcula- the rate of reward (Gold & Shadlen 2002).
tion of the DV (in this case describing ex- The point is that even perceptual decisions
pected reward) based on the recent history have, at their core, value judgments. Noisy in-
of rewards, followed by a random (Poisson) put leads to imperfect output. Because no uni-
process that generates a choice based on the versal prescription exists for which imperfec-
DV. Critically, the values of the parame- tions are acceptable and which are not, their
ters of the model that provided the best fit relative values must be weighed and then used
to the data were very close to optimal in to shape the decision process. It thus seems
terms of maximizing the average reward re- reasonable to posit that value-based decisions
ceived per trial. However, alternative mod- can exploit these same mechanisms. This re-
els that assumed that noise was present in mains to be seen.
the DV instead of (or in addition to) the
decision rule were not tested, so it is diffi-
cult to assess which model is the more likely CONCLUSIONS
implementation. In this review we have evaluated progress in
In general, it is not clear why core princi- understanding how the brain forms decisions.
ples of decision making that apply to percep- Our focus was intentionally narrow, consider-
tual tasks should be abandoned to account for ing only decisions on simple sensory-motor
these phenomena. The key issue is whether tasks that are amenable to behavioral and

ANRV314-NE30-21 ARI 21 May 2007 13:44
neurophysiological studies in the laboratory. volving many alternatives. Recent theoretical

We have presented a theoretical framework work has begun to explore algorithms to solve
from statistical decision theory that describes these decisions [e.g., the multiple sequential
how to form decisions using priors, evi- probability ratio test (MSPRT); McMillen &
dence, and value to achieve certain desir- Holmes 2006] and how they might be imple-
able goals. We have used the elements of mented in the brain (Gurney & Bogacz 2006,
this framework—particularly the distinction Roe et al. 2001, Usher & McClelland 2001),
between evidence and the DV—to analyze but much work remains to be done.
experimental results from tasks requiring A third qualifying factor is the explicit link
perceptual decisions (discrimination and de- between the decision and a specific course of
tection), simple motor decisions, and value- action (e.g., an eye or hand movement) en-
based decisions. forced in most tasks. As described above, many
Even these simple sensory-motor tasks re- neurophysiological studies exploit this design
quire nuanced and flexible mechanisms that by treating the decision process as a problem
seem likely to play general roles in decision of movement selection. The search for neu-
making. However, one must consider sev- ral correlates of the decision can thus focus
eral qualifying factors. First, there is a strong on parts of the brain known to select and pre-
degree of automaticity in subjects perform- pare the associated movement. This approach
ing these tasks over and over, in some cases has shown that the mechanisms of movement
for weeks or months at a time. This is in selection appear to incorporate all the ele-
stark contrast to many of the decisions we en- ments of a deliberative decision. However, it
counter in real life, such as deciding whom to leaves open the question of how and where
marry, which require more deliberation and the brain forms decisions that are not used to
often have few or no similar experiences from select a particular movement. One possibility
which to draw. In fact, the case has been made that maintains this intention-based architec-
for two distinct decision-making systems: one ture is that abstract decisions are formed by
“intuitive,” which controls simple behaviors circuits involved in abstract forms of behav-
learned through repeated experience, and the ioral planning, e.g., flexible rules that involve
other “deliberative,” which is designed to future contingencies, of the sort thought to
achieve goals in a dynamic environment (Kah- be encoded in areas of the prefrontal cortex
neman 2002). However, there is little evidence (Wallis et al. 2001).
for these distinct mechanisms in the brain The path from simple decisions to com-
(Sugrue et al. 2005). As we have argued, even plex ones may be more straightforward than
the simplest sensory-motor decisions seem to it appears. Consider a simple decision about
be based on deliberative elements. the direction of RDM that is not tied to a par-
A second qualifying factor is that many ticular action (Gold & Shadlen 2003, Horwitz
tasks require a selection between two alter- et al. 2004). For this abstract decision, the DV
natives. This design, long favored by psy- and decision rule are likely recognizable but
chophysicists, allows for rigorous quantifica- are carried out in circuits linked to working
tion of the relationship between stimulus and memory, long-term planning, or behavioral
response. It is also consistent with decision al- contingencies (e.g., the context/motivation
gorithms, such as SPRT, that are based on the boxes in Figure 1) as opposed to specific ac-
value of a ratio (e.g., logLR) that is an explicit tions. Other complex decisions are made us-
comparison of the two alternatives (see The ing sources of evidence that, like priors and
Sequential Probability Ratio Test). However, value, do not come from the senses but in-
it is unclear how or even if the mechanisms stead derive entirely from memory (Wagner
responsible for these simple comparisons can et al. 1998). Indeed, the speed and accu-
generalize to more complex decisions in- racy of certain decisions that require memory
562 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
retrieval are consistent with the framework of from simple sensory-motor paradigms to the
SA (Ratcliff 1978). These considerations may kind of complex decisions that comprise the
one day allow us to extend insights obtained fabric of cognition.
SUMMARY POINTS
1. A decision is a process that weighs priors, evidence, and value to generate a commit-
ment to a categorical proposition intended to achieve particular goals.
2. Signal detection theory and sequential analysis provide a theoretical framework for
understanding how decisions are formed. They describe specific, mathematical oper-
ations that correspond to key decision elements including deliberation and commit-
ment.
3. Studies that combine behavior and neurophysiology, typically in monkeys, have begun
to uncover how the elements of decision formation are implemented in the brain.
4. Perceptual tasks have been particularly useful for distinguishing between sensory
evidence, which transiently encodes information from the senses, and a decision vari-
able, which accumulates and stores evidence over time until the final commitment is
reached.
5. The speed-accuracy trade-off on perceptual tasks and variable reaction times on simple
motor tasks can be explained by a basic mechanism that appears to be central to many
forms of decision making: a decision rule equivalent to comparing an evolving decision
variable to a fixed criterion.
FUTURE ISSUES
1. How and where in the brain does information pertaining to priors, sensory evidence,
and/or values combine? Which units are used?
2. Is the decision variable simply a useful abstraction or an explicitly represented quantity
that is critical for decision formation?
3. How do neural circuits integrate information as a function of time?
4. How and where in the brain is the decision rule (e.g., a bound crossing) implemented?
5. How does the brain form decisions that involve more than two alternatives?
6. Under which conditions does the decision rule explicitly invoke randomness when
making the final choice?
7. How and where are decisions formed that are not tied to specific behavioral output?
8. How does experience optimize decision mechanisms to achieve particular goals?
ACKNOWLEDGMENTS
We thank many colleagues for helpful discussions of the issues in this paper: Ken Britten, Roger
Carpenter, Eric Cook, Greg DeAngelis, Paul Glimcher, Tim Hanks, Ben Heasly, Bharathi
Jagadeesh, Roozbeh Kiani, Victor de Lafuente, Brian Lau, Jeff Law, John Maunsell, Bill
Newsome, Ranulfo Romo, Xiao-Jing Wang, and Tianming Yang.

ANRV314-NE30-21 ARI 21 May 2007 13:44
LITERATURE CITED
Abraham NM, Spors H, Carleton A, Margrie TW, Kuner T, Schaefer AT. 2004. Maintaining
accuracy at the expense of speed: Stimulus similarity defines odor discrimination time in
mice. Neuron 44:865–76
Afraz SR, Kiani R, Esteky H. 2006. Microstimulation of inferotemporal cortex influences face
categorization. Nature 442:692–95
Allred S, Liu Y, Jagadeesh B. 2005. Selectivity of inferior temporal neurons for realistic pictures
predicted by algorithms for image database navigation. J. Neurophysiol. 94:4068–81
Andersen RA, Asanuma C, Essick G, Siegel RM. 1990. Corticocortical connections of anatom-
ically and physiologically defined subdivisions within the inferior parietal lobule. J. Comp.
Neurol. 296:65–113
Andersen RA, Brotchie PR, Mazzoni P. 1992. Evidence for the lateral intraparietal area as the
parietal eye field. Curr. Opin. Neurobiol. 2:840–46
Asanuma C, Andersen RA, Cowan WM. 1985. The thalamic relations of the caudal inferior
parietal lobule and the lateral prefrontal cortex in monkeys: divergent cortical projections
from cell clusters in the medial pulvinar nucleus. J. Comp. Neurol. 241:357–81
Assad J, Maunsell J. 1995. Neuronal correlates of inferred motion in primate posterior parietal
cortex. Nature 373:518–21
Audley RJ, Pike AR. 1965. Some alternative stochastic models of choice. Br. J. Math. Stat.
Psychol. 18:207–55
Barraclough DJ, Conroy ML, Lee D. 2004. Prefrontal cortex and decision making in a mixed-
strategy game. Nat. Neurosci. 7:404–10
Basso MA, Wurtz RH. 1997. Modulation of neuronal activity by target uncertainty. Nature
389:66–69
Basso MA, Wurtz RH. 1998. Modulation of neuronal activity in superior colliculus by changes
in target probability. J. Neurosci. 18:7519–34
Baylis V, Salter L, Locke R. 2003. Pathways for continence care: an audit to assess how they
are used. Br. J. Nurs. 12:857–63
Bisley JW, Krishna BS, Goldberg ME. 2004. A rapid and precise on-response in posterior
parietal cortex. J. Neurosci. 24:1833–38
Blatt GJ, Andersen RA, Stoner GR. 1990. Visual receptive field organization and cortico-
cortical connections of the lateral intraparietal area (area LIP) in the macaque. J. Comp.
Neurol. 299:421–45
Bradley DC, Chang GC, Andersen RA. 1998. Encoding of three-dimensional structure-from-
motion by primate area MT neurons. Nature 392:714–16
Breiter HC, Aharon I, Kahneman D, Dale A, Shizgal P. 2001. Functional imaging of neural
responses to expectancy and experience of monetary gains and losses. Neuron 30:619–39
Bremmer F, Duhamel JR, Ben Hamed S, Graf W. 2002a. Heading encoding in the macaque
ventral intraparietal area (VIP). Eur J. Neurosci. 16:1554–68
Bremmer F, Klam F, Duhamel JR, Ben Hamed S, Graf W. 2002b. Visual-vestibular interactive
responses in the macaque ventral intraparietal area (VIP). Eur J. Neurosci. 16:1569–86
Britten KH. 1998. Clustering of response selectivity in the medial superior temporal area of
extrastriate cortex in the macaque monkey. Vis. Neurosci. 15:553–58
Britten KH, Newsome WT, Shadlen MN, Celebrini S, Movshon JA. 1996. A relationship
between behavioral choice and the visual responses of neurons in macaque MT. Vis.
Neurosci. 13:87–100
Britten KH, Shadlen MN, Newsome WT, Movshon JA. 1992. The analysis of visual motion:
a comparison of neuronal and psychophysical performance. J. Neurosci. 12:4745–65
564 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
Britten KH, Shadlen MN, Newsome WT, Movshon JA. 1993. Responses of neurons in
macaque MT to stochastic motion signals. Vis. Neurosci. 10:1157–69
Britten KH, van Wezel RJ. 2002. Area MST and heading perception in macaque monkeys.
Cereb. Cortex 12:692–701
Britten KH, van Wezel RJ. 1998. Electrical microstimulation of cortical area MST biases
heading perception in monkeys. Nat. Neurosci. 1:59–64
Brody CD, Hernandez A, Zainos A, Romo R. 2003. Timing and neural encoding of somatosen-
sory parametric working memory in macaque prefrontal cortex. Cereb. Cortex 13:1196–207
Bruce CJ, Goldberg ME. 1985. Primate frontal eye fields. I. Single neurons discharging before
saccades. J. Neurophysiol. 53:603–35
Bruce CJ, Goldberg ME, Bushnell MC, Stanton GB. 1985. Primate frontal eye fields. II. Phys-
iological and anatomical correlates of electrically evoked eye movements. J. Neurophysiol.
54:714–34
Buck L, Axel R. 1991. A novel multigene family may encode odorant receptors: a molecular
basis for odor recognition. Cell 65:175–87
Buck LB. 1996. Information coding in the vertebrate olfactory system. Annu. Rev. Neurosci.
19:517–44
Busemeyer JR, Townsend JT. 1993. Decision field theory: a dynamic-cognitive approach to
decision making in an uncertain environment. Psychol. Rev. 100:432–59

Carpenter RHS. 1981. In Eye Movements: Cognition and Visual Perception, ed. DF Fischer, RA
Monty, JW Senders, pp. 237–46. Hillsdale, NJ: Lawrence Erlbaum
Carpenter RHS. 1988. Movements of the Eyes. London: Pion
Carpenter RHS, Williams MLL. 1995. Neural computation of log likelihood in control of
saccadic eye movements. Nature 377:59–62
Carter CS, Braver TS, Barch DM, Botvinick MM, Noll D, Cohen JD. 1998. Anterior cingulate
cortex, error detection, and the online monitoring of performance. Science 280:747–49
Chafee MV, Goldman-Rakic PS. 2000. Inactivation of parietal and prefrontal cortex re-
veals interdependence of neural activity during memory-guided saccades. J. Neurophysiol.
83:1550–66
Cisek P. 2007. Cortical mechanisms of action selection: the affordance competition hypothesis.
Philos. Trans. R. Soc. B. In press
Clark A. 1997. Being There: Putting Brain, Body, and World Together Again. Cambridge, MA:
MIT Press. 269 pp.
Colby CL, Duhamel JR, Goldberg ME. 1993. Ventral intraparietal area of the macaque:
anatomic location and visual response properties. J. Neurophysiol. 69:902–14
Connolly JD, Goodale MA, Goltz HC, Munoz DP. 2005. fMRI activation in the human frontal
eye field is correlated with saccadic reaction time. J. Neurophysiol. 94:605–11
Cook EP, Maunsell JH. 2002a. Attentional modulation of behavioral performance and neu-
ronal responses in middle temporal and ventral intraparietal areas of macaque monkey. J.
Neurosci. 22:1994–2004
Cook EP, Maunsell JH. 2002b. Dynamics of neuronal responses in macaque MT and VIP
during motion detection. Nat. Neurosci. 5:985–94
Corrado GS, Sugrue LP, Seung HS, Newsome WT. 2005. Linear-nonlinear-Poisson models
of primate choice dynamics. J. Exp. Anal. Behav. 84:581–617
DeAngelis GC, Cumming BG, Newsome WT. 1998. Cortical area MT and the perception of
stereoscopic depth. Nature 394:677–80
DeAngelis GC, Ghose GM, Ohzawa I, Freeman RD. 1999. Functional micro-organization of
primary visual cortex: receptive field analysis of nearby neurons. J. Neurosci. 19:4046–64

ANRV314-NE30-21 ARI 21 May 2007 13:44
DeAngelis GC, Gu Y, Angelaki DE. 2006. Role of area MSTd in cue integration for head-
ing discrimination: II. Analysis of correlations between neural responses and perceptual
decisions. J. Vis. 6(Abstr. 408):408a
DeAngelis GC, Newsome WT. 2004. Perceptual “read-out” of conjoined direction and dis-
parity maps in extrastriate area MT. PLoS Biol. 2:E77
de Lafuente V, Romo R. 2005. Neuronal correlates of subjective sensory experience. Nat.
Diederich A. 2003. MDFT account of decision making under time pressure. Psychon. Bull. Rev.
10:157–66
Ditterich J, Mazurek M, Shadlen MN. 2003. Microstimulation of visual cortex affects the speed
of perceptual decisions. Nat. Neurosci. 6:891–98
Dodd JV, Krug K, Cumming BG, Parker AJ. 2001. Perceptually bistable three-dimensional
figures evoke high choice probabilities in cortical area MT. J. Neurosci. 21:4809–21
Dolan RJ, Fink GR, Rolls E, Booth M, Holmes A, et al. 1997. How the brain learns to see
objects and faces in an impoverished context. Nature 389:596–99
Dorris MC, Glimcher PW. 2004. Activity in posterior parietal cortex is correlated with the
relative subjective desirability of action. Neuron 44:365–78
Dorris MC, Munoz DP. 1998. Saccadic probability influences motor preparation signals and
time to saccadic initiation. J. Neurosci. 18:7015–26

Dorris MC, Paré M, Munoz DP. 1997. Neuronal activity in monkey superior colliculus related
to the initiation of saccadic eye movements. J. Neurosci. 17:8566–79
Duffy CJ, Wurtz RH. 1991. Sensitivity of MST neurons to optic flow stimuli. I. A continuum
of response selectivity of large-field stimuli. J. Neurophysiol. 65:1329–45
Duffy CJ, Wurtz RH. 1995. Response of monkey MST neurons to optic flow stimuli with
shifted centers of motion. J. Neurosci. 15:5192–208
Duffy CJ, Wurtz RH. 1997. Medial superior temporal area neurons respond to speed patterns
in optic flow. J. Neurosci. 17:2839–51
Duhamel JR, Colby CL, Goldberg ME. 1998. Ventral intraparietal area of the macaque: con-
gruent visual and somatic response properties. J. Neurophysiol. 79:126–36
Eskandar EN, Assad JA. 1999. Dissociation of visual, motor and predictive signals in parietal
cortex during visual guidance. Nat. Neurosci. 2:88–93
Everling S, Munoz DP. 2000. Neuronal correlates for preparatory set associated with prosac-
cades and antisaccades in the primate frontal eye field. J. Neurosci. 20:387–400
Freedman DJ, Assad JA. 2006. Experience-dependent representation of visual categories in
parietal cortex. Nature 443:85–88
Freedman DJ, Riesenhuber M, Poggio T, Miller EK. 2002. Visual categorization and the
primate prefrontal cortex: Neurophysiology and behavior. J. Neurophysiol. 88:929–41
Freedman DJ, Riesenhuber M, Poggio T, Miller EK. 2003. A comparison of primate prefrontal
and inferior temporal cortices during visual categorization. J. Neurosci. 23:5235–46
Friedman HR, Goldman-Rakic PS. 1994. Coactivation of prefrontal cortex and inferior pari-
etal cortex in working memory tasks revealed by 2DG functional mapping in the rhesus
monkey. J. Neurosci. 14:2775–88
Fries W. 1984. Cortical projections to the superior colliculus in the macaque monkey: a ret-
rograde study using horseradish peroxidase. J. Comp. Neurol. 230:55–76
Gehring WJ, Willoughby AR. 2002. The medial frontal cortex and the rapid processing of
monetary gains and losses. Science 295:2279–82
Gibson JJ. 1950. Perception of the Visual World. Boston: Houghton-Mifflin
Glimcher PW. 2005. Indeterminacy in brain and behavior. Annu. Rev. Psychol. 56:25–56
566 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
Gold JI, Shadlen MN. 2000. Representation of a perceptual decision in developing oculomotor
commands. Nature 404:390–94
Gold JI, Shadlen MN. 2001. Neural computations that underlie decisions about sensory stimuli.
Trends Cogn. Sci. 5:10–16
Gold JI, Shadlen MN. 2002. Banburismus and the brain: decoding the relationship between
sensory stimuli, decisions, and reward. Neuron 36:299–308
Gold JI, Shadlen MN. 2003. The influence of behavioral context on the representation of a
perceptual decision in developing oculomotor commands. J. Neurosci. 23:632–51
Good IJ. 1979. Studies in the history of probability and statistics. XXXVII A.M. Turing’s
statistical work in World War II. Biometrika 66:393–96
Good IJ. 1983. Good Thinking: The Foundations of Probability and Its Applications. Minneapolis:
Univ. Minn. Press
Goodale M, Milner A. 1992. Separate visual pathways for perception and action. Trends Neurosci.
15:20–25
Graziano MS, Andersen RA, Snowden RJ. 1994. Tuning of MST neurons to spiral motions. J.
Graziano MSA, Hu XT, Gross CG. 1997. Visuospatial properties of ventral premotor cortex.
J. Neurophysiol. 77:2268–92
Green DM, Swets JA. 1966. Signal Detection Theory and Psychophysics. New York: Wiley
Grill-Spector K, Kushnir T, Hendler T, Malach R. 2000. The dynamics of object-selective

activation correlate with recognition performance in humans. Nat. Neurosci. 3:837–43
Gu Y, Watkins PV, Angelaki DE, DeAngelis GC. 2006. Visual and nonvisual contributions
to three-dimensional heading selectivity in the medial superior temporal area. J. Neurosci.
26:73–85
Gurney K, Bogacz R. 2006. The basal ganglia and cortex implement optimal decision making
between alternative actions. Neural Comput. 19:442–77
Hanes DP, Carpenter RH. 1999. Countermanding saccades in humans. Vis. Res. 39:2777–91
Hanes DP, Patterson WF, Schall JD. 1998. Role of frontal eye fields in countermanding
saccades: visual, movement, and fixation activity. J. Neurophys. 79:817–34
Hanes DP, Schall JD. 1996. Neural control of voluntary movement initiation. Science 274:427–
30
Hanks TD, Ditterich J, Shadlen MN. 2006. Microstimulation of macaque area LIP affects
decision-making in a motion discrimination task. Nat. Neurosci. 9:682–89
Haxby JV, Horwitz B, Ungerleider LG, Maisog JM, Pietrini P, Grady CL. 1994. The functional
organization of human extrastriate cortex: a PET-rCBF study of selective attention to faces
and locations. J. Neurosci. 14:6336–53
Heekeren HR, Marrett S, Bandettini PA, Ungerleider LG. 2004. A general mechanism for
perceptual decision-making in the human brain. Nature 431:859–62
Henning GB, Hertz BG, Broadbent DE. 1975. Some experiments bearing on the hypothesis
that the visual system analyzes patterns in independent bands of spatial frequency. Vis. Res.
15:887–89
Hernandez A, Zainos A, Romo R. 2000. Neuronal correlates of sensory discrimination in the
somatosensory cortex. Proc. Natl. Acad. Sci. USA 97:6191–96
Hernandez A, Zainos A, Romo R. 2002. Temporal evolution of a decision-making process in
medial premotor cortex. Neuron 33:959–72
Herrnstein RJ. 1961. Relative and absolute strength of response as a function of frequency of
reinforcement. J. Exp. Anal. Behav. 4:267–72
Herrnstein RJ, Vaughan W. 1980. Melioration and behavioral allocation. In Limits to Action:
The Allocation of Individual Behavior, ed. J Staddon, pp. 143–76. New York: Academic

ANRV314-NE30-21 ARI 21 May 2007 13:44
Heuer HW, Britten KH. 2004. Optic flow signals in extrastriate area MST: comparison of
perceptual and neuronal sensitivity. J. Neurophysiol. 91:1314–26
Horwitz GD, Batista AP, Newsome WT. 2004. Representation of an abstract perceptual dec-
cision in macaque superior colliculus. J. Neurophysiol. 91:2281–96
Horwitz GD, Newsome WT. 1999. Separate signals for target selection and movement spec-
ification in the superior colliculus. Science 284:1158–61
Horwitz GD, Newsome WT. 2001. Target selection for saccadic eye movements: prelude
activity in the superior colliculus during a direction-discrimination task. J. Neurophysiol.
86:2543–58
Huk AC, Shadlen MN. 2005. Neural activity in macaque parietal cortex reflects temporal inte-
gration of visual motion signals during perceptual decision making. J. Neurosci. 25:10420–
36
Ito S, Stuphorn V, Brown JW, Schall JD. 2003. Performance monitoring by the anterior
cingulate cortex during saccade countermanding. Science 302:120–22
Janssen P, Shadlen MN. 2005. A representation of the hazard rate of elapsed time in macaque
area LIP. Nat. Neurosci. 8:234–41
Johnson KO. 1980a. Sensory discrimination: decision process. J. Neurophysiol. 43:1771–92
Johnson KO. 1980b. Sensory discrimination: neural processes preceding discrimination deci-
sion. J. Neurophysiol. 43:1793–815
Kaelbling LP, Littman ML, Moore AW. 1996. Reinforcement learning: a survey. J. Artif. Intel.
Res. 4:237–85
Kahneman D. 2002. Nobel prize lecture: Maps of Bounded Rationality: a perspective on intu-
itive judgment and choice. In Nobel Prizes 2002: Nobel Prizes, Presentations, Biographies, &
Lectures, ed. T Frangsmyr, pp. 416–99. Stockholm: Almqvist & Wiksell Int.
Kanwisher N, McDermott J, Chun MM. 1997. The fusiform face area: a module in human
extrastriate cortex specialized for face perception. J. Neurosci. 17:4302–11
Kawagoe R, Takikawa Y, Hikosaka O. 2004. Reward-predicting activity of dopamine and
caudate neurons—a possible mechanism of motivational control of saccadic eye movement.
J. Neurophysiol. 91:1013–24
Kepecs A, Uchida N, Mainen ZF. 2006. The sniff as a unit of olfactory processing. Chem. Senses
31:167–79
Kersten D, Mamassian P, Yuille A. 2004. Object perception as Bayesian inference. Annu. Rev.
Psychol. 55:271–304
Kiani R, Hanks TD, Shadlen MN. 2006. Improvement in sensitivity with viewing time is
limited by an integration-to-bound mechanism in area LIP. Soc. Neurosci. Abstr. 605.7
Kim JN, Shadlen MN. 1999. Neural correlates of a decision in the dorsolateral prefrontal
cortex of the macaque. Nat. Neurosci. 2:176–85
Kohn A, Smith MA. 2005. Stimulus dependence of neuronal correlation in primary visual
cortex of the macaque. J. Neurosci. 25:3661–73
Knutson B, Taylor J, Kaufman M, Peterson R, Glover G. 2005. Distributed neural represen-
tation of expected value. J. Neurosci. 25:4806–12
Kobayashi S, Lauwereyns J, Koizumi M, Sakagami M, Hikosaka O. 2002. Influence of reward
expectation on visuospatial processing in macaque lateral prefrontal cortex. J. Neurophysiol.
87:1488–98
Krug K, Cumming BG, Parker AJ. 2004. Comparing perceptual signals of single V5/MT
neurons in two binocular depth tasks. J. Neurophysiol. 92:1586–96
LaBerge DA. 1962. A recruitment theory of simple behavior. Psychometrika 27:375–96
Lagae L, Maes H, Raiguel S, Xiao DK, Orban GA. 1994. Responses of macaque STS neurons to
optic flow components: a comparison of areas MT and MST. J. Neurophysiol. 71:1597–626
568 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
Laming DRJ. 1968. Information Theory of Choice-Reaction Times. London: Academic

Lasley DJ, Cohn T. 1981. Detection of a luminance increment: effect of temporal uncertainty.
J. Opt. Soc. Am. 71:845–50
Lau B, Glimcher PW. 2005. Dynamic response-by-response models of matching behavior in
rhesus monkeys. J. Exp. Anal. Behav. 84:555–79
Lauwereyns J, Takikawa Y, Kawagoe R, Kobayashi S, Koizumi M, et al. 2002. Feature-based
anticipation of cues that predict reward in monkey caudate nucleus. Neuron 33:463–73
Law C, Gold JI. 2005. Physiological correlates of perceptual learning in monkey areas MT and
LIP. Soc. Neurosci. Abstr. 621.15
Lecas JC, Requin J, Anger C, Vitton N. 1986. Changes in neuronal activity of the monkey
precentral cortex during preparation for movement. J. Neurophysiol. 56:1680–702
Lee D, McGreevy BP, Barraclough DJ. 2005. Learning and decision making in monkeys during
a rock-paper-scissors game. Brain. Res. Cogn. Brain. Res. 25:416–30
Leon MI, Shadlen MN. 1998. Modulation of dorsolateral prefrontal cortex neurons by varying
expectations of reward magnitude. Soc. Neurosci. Abstr. 24:1425
Leon MI, Shadlen MN. 2003. Representation of time by neurons in the posterior parietal
cortex of the macaque. Neuron 38:317–27
Lewis JW, Van Essen DC. 2000a. Corticocortical connections of visual, sensorimotor, and
multimodal processing areas in the parietal lobe of the macaque monkey. J. Comp. Neurol.
428:112–37
Lewis JW, Van Essen DC. 2000b. Mapping of architectonic subdivisions in the macaque mon-
key, with emphasis on parieto-occipital cortex. J. Comp. Neurol. 428:79–111
Link SW. 1992. The Wave Theory of Difference and Similarity. Hillsdale, NJ: Lawrence Erlbaum
Link SW, Heath RA. 1975. A sequential theory of psychological discrimination. Psychometrika
40:77–105
Lo CC, Wang XJ. 2006. Cortico-basal ganglia circuit mechanism for a decision threshold in
reaction time tasks. Nat. Neurosci. 9:956–63
Logan GD. 2002. An instance theory of attention and memory. Psychol. Rev. 109:376–400
Logan GD, Cowan WB, Davis KA. 1984. On the ability to inhibit simple and choice reaction
time responses: a model and a method. J. Exp. Psychol. Hum. Percept. Perform. 10:276–91
Luce RD. 1986. Response Times: Their Role in Inferring Elementary Mental Organization. New
York: Oxford Univ. Press
Ludwig CJ, Gilchrist ID, McSorley E, Baddeley RJ. 2005. The temporal impulse response
underlying saccadic decisions. J. Neurosci. 25:9907–12
Luna R, Hernandez A, Brody CD, Romo R. 2005. Neural codes for perceptual discrimination
in primary somatosensory cortex. Nat. Neurosci. 8:1210–19
Lynch JC, Graybiel AM, Lobeck LJ. 1985. The differential projection of two cytoarchitectonic
subregions of the inferior parietal lobule of macaque upon the deep layers of the superior
colliculus. J. Comp. Neurol. 235:241–54
Machens CK, Romo R, Brody CD. 2005. Flexible control of mutual inhibition: a neural model
of two-interval discrimination. Science 307:1121–24
Major G, Tank D. 2004. Persistent neural activity: prevalence and mechanisms. Curr. Opin.
Neurobiol. 14:675–84
Maunsell JHR, Van Essen DC. 1983. Functional properties of neurons in the middle temporal
visual area (MT) of the macaque monkey: II. Binocular interactions and the sensitivity to
binocular disparity. J. Neurophysiol. 49:1148–67
Mazurek ME, Roitman JD, Ditterich J, Shadlen MN. 2003. A role for neural integrators in
perceptual decision making. Cereb. Cortex 13:1257–69

ANRV314-NE30-21 ARI 21 May 2007 13:44
McCoy AN, Crowley JC, Haghighian G, Dean HL, Platt ML. 2003. Saccade reward signals
in posterior cingulate cortex. Neuron 40:1031–40
McMillen T, Holmes P. 2006. The dynamics of choice among multiple alternatives. J. Math.
Psychol. 50:30–57
Meister M, Bonhoeffer T. 2001. Tuning and topography in an odor map on the rat olfactory
bulb. J. Neurosci. 21:1351–60
Merleau-Ponty M. 1962. Phenomenology of Perception. London: Routledge & Kegan Paul
Mombaerts P, Wang F, Dulac C, Chao SK, Nemes A, et al. 1996. Visualizing an olfactory
sensory map. Cell 87:675–86
Mountcastle VB, Steinmetz MA, Romo R. 1990. Frequency discrimination in the sense of flut-
ter: psychophysical measurements correlated with postcentral events in behaving monkeys.
J. Neurosci. 10:3032–44
Munoz DP, Wurtz RH. 1993. Fixation cells in monkey superior colliculus. I. Characteristics
of cell discharge. J. Neurophys. 70:559–75

Nachmias J, Rogowitz BE. 1983. Masking by spatially-modulated gratings. Vis. Res. 23:1621–29
Newsome WT, Paré EB. 1988. A selective impairment of motion perception following lesions
of the middle temporal visual area (MT). J. Neurosci. 8:2201–11
Op de Beeck H, Wagemans J, Vogels R. 2001. Inferotemporal neurons represent low-
dimensional configurations of parameterized shapes. Nat. Neurosci. 4:1244–52

O’Regan JK, Noë A. 2001. A sensorimotor account of vision and visual consciousness. Behav.
Brain Sci. 24:939–73
Padoa-Schioppa C, Assad JA. 2006. Neurons in the orbitofrontal cortex encode economic
value. Nature 441:223–26
Palmer J, Huk AC, Shadlen MN. 2005. The effect of stimulus strength on the speed and
accuracy of a perceptual decision. J. Vis. 5:376–404
Paré M, Hanes DP. 2003. Controlled movement processing: superior colliculus activity asso-
ciated with countermanded saccades. J. Neurosci. 23:6480–89
Paré M, Wurtz RH. 1997. Monkey posterior parietal cortex neurons antidromically activated
from superior colliculus. J. Neurophysiol. 78:3493–98
Parker AJ, Newsome WT. 1998. Sense and the single neuron: probing the physiology of
perception. Annu. Rev. Neurosci. 21:227–77
Philiastides MG, Ratcliff R, Sajda P. 2006. Neural representation of task difficulty and decision
making during perceptual categorization: a timing diagram. J. Neurosci. 26:8965–75
Philiastides MG, Sajda P. 2006. Temporal characterization of the neural correlates of perceptual
decision making in the human brain. Cereb. Cortex 16:509–18
Platt ML, Glimcher PW. 1999. Neural correlates of decision variables in parietal cortex. Nature
400:233–38
Pouget P, Emeric EE, Stuphorn V, Reis K, Schall JD. 2005. Chronometry of visual responses
in frontal eye field, supplementary eye field, and anterior cingulate cortex. J. Neurophysiol.
94:2086–92
Rainer G, Lee H, Logothetis NK. 2004. The effect of learning on the function of monkey
extrastriate visual cortex. PLoS Biol. 2:E44
Rao RPN. 1999. An optimal estimation approach to visual perception and learning. Vis. Res.
39:1963–89
Ratcliff R. 1978. A theory of memory retrieval. Psychol. Rev. 85:59–108
Ratcliff R, Cherian A, Segraves M. 2003. A comparison of macaque behavior and superior
colliculus neuronal activity to predictions from models of two-choice decisions. J. Neuro-
physiol. 90:1392–407
570 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
Ratcliff R, Rouder JN. 1998. Modeling response times for two-choice decisions. Psychol. Sci.
9:347–56
Ratcliff R, Smith PL. 2004. A comparison of sequential sampling models for two-choice reac-
tion time. Psychol. Rev. 111:333–67
Reddi BA, Asrress KN, Carpenter RH. 2003. Accuracy, information, and response time in a
saccadic decision task. J. Neurophysiol. 90:3538–46
Reddi BA, Carpenter RH. 2000. The influence of urgency on decision time. Nat. Neurosci.
3:827–30
Ress D, Backus BT, Heeger DJ. 2000. Activity in primary visual cortex predicts performance
in a visual detection task. Nat. Neurosci. 3:940–45
Ress D, Heeger DJ. 2003. Neuronal correlates of perception in early visual cortex. Nat. Neurosci.
6:414–20
Ridderinkhof KR, Ullsperger M, Crone EA, Nieuwenhuis S. 2004. The role of the medial
frontal cortex in cognitive control. Science 306:443–47
Riehle A, Requin J. 1993. The predictive value for performance speed of preparatory changes
in neuronal activity of the monkey motor and premotor cortex. Behav. Brain Res. 53:35–49
Rieke F, Warland D, de Ruyter van Steveninck RR, Bialek W. 1997. Spikes: Exploring the Neural
Code. Cambridge, MA: MIT Press
Rinberg D, Koulakov A, Gelperin A. 2006. Speed-accuracy tradeoff in olfaction. Neuron
51:351–58
Robinson DA. 1972. Eye movements evoked by collicular stimulation in the alert monkey. Vis.
Res. 12:1795–807
Robinson DA, Fuchs AF. 1969. Eye movements evoked by stimulation of frontal eye fields. J.
Neurophysiol. 32:637–48
Roe RM, Busemeyer JR, Townsend JT. 2001. Multialternative decision field theory: a dynamic
connectionist model of decision making. Psychol. Rev. 108:370–92
Roitman JD, Shadlen MN. 2002. Response of neurons in the lateral intraparietal area during
a combined visual discrimination reaction time task. J. Neurosci. 22:9475–89
Romo R, Brody C, Hernandez A, Lemus L. 1999. Neuronal correlates of parametric working
memory in the prefrontal cortex. Nature 399:470–73
Romo R, Hernandez A, Zainos A. 2004. Neuronal correlates of a perceptual decision in ventral
premotor cortex. Neuron 41:165–73
Romo R, Hernandez A, Zainos A, Brody CD, Lemus L. 2000. Sensing without touching:
psychophysical performance based on cortical microstimulation. Neuron 26:273–78
Romo R, Hernandez A, Zainos A, Lemus L, Brody CD. 2002. Neuronal correlates of decision-
making in secondary somatosensory cortex. Nat. Neurosci. 5:1217–25
Romo R, Hernandez A, Zainos A, Salinas E. 1998. Somatosensory discrimination based on
cortical microstimulation. Nature 392:387–90
Rubin BD, Katz LC. 1999. Optical imaging of odorant representations in the mammalian
olfactory bulb. Neuron 23:499–511
Saito H, Yukie M, Tanaka K, Hikosaka K, Fukada Y, Iwai E. 1986. Integration of direction
signals of image motion in the superior temporal sulcus of the macaque monkey. J. Neurosci.
6:145–57
Salinas E, Hernandez A, Zainos A, Romo R. 2000. Periodicity and firing rate as candidate
neural codes for the frequency of vibrotactile stimuli. J. Neurosci. 20:5503–15
Salzman CD, Britten KH, Newsome WT. 1990. Cortical microstimulation influences percep-
tual judgements of motion direction. Nature 346:174–77
Salzman CD, Murasugi CM, Britten KH, Newsome WT. 1992. Microstimulation in visual
area MT: effects on direction discrimination performance. J. Neurosci. 12:2331–55

ANRV314-NE30-21 ARI 21 May 2007 13:44
Sanfey AG, Loewenstein G, McClure SM, Cohen JD. 2006. Neuroeconomics: cross-currents
in research on decision-making. Trends Cogn. Sci. 10:108–16
Schall JD, Stuphorn V, Brown JW. 2002. Monitoring and control of action by the frontal lobes.
Neuron 36:309–22
Schiller PH, Stryker M. 1972. Single-unit recording and stimulation in superior colliculus of
the alert rhesus monkey. J. Neurophysiol. 35:915–24
Schmolesky MT, Wang Y, Hanes DP, Thompson KG, Leutgeb S, et al. 1998. Signal timing
across the macaque visual system. J. Neurosci. 79:3272–78
Schultz W. 1992. Activity of dopamine neurons in the behaving primate. Semin. Neurosci.
4:129–38
Schultz W. 1998. Predictive reward signal of dopamine neurons. J. Neurophysiol. 80:1–27
Schultz W. 2004. Neural coding of basic reward terms of animal learning theory, game theory,
microeconomics and behavioural ecology. Curr. Opin. Neurbiol. 14:139–47
Schultz W, Dayan P, Montague PR. 1997. A neural substrate of prediction and reward. Science
275:1593–99
Schrater PR, Knill DC, Simoncelli EP. 2000. Mechanisms of visual motion detection. Nat.
Neurosci. 3:64–68
Shadlen MN, Britten KH, Newsome WT, Movshon JA. 1996. A computational analysis of
the relationship between neuronal and behavioral responses to visual motion. J. Neurosci.
16:1486–510
Shadlen MN, Hanks TD, Churchland AK, Kiani R, Yang T. 2006. The speed and accuracy
of a simple perceptual decision: a mathematical primer. In Bayesian Brain: Probabilistic
Approaches to Neural Coding, ed. K Doya, S Ishii, R Rao, A Pouget, pp. 209–37. Cambridge,
MA: MIT Press
Shadlen MN, Newsome WT. 1996. Motion perception: Seeing and deciding. Proc. Natl. Acad.
Sci. USA 93:628–33
Shadlen MN, Newsome WT. 2001. Neural basis of a perceptual decision in the parietal cortex
(area LIP) of the rhesus monkey. J. Neurophysiol. 86:1916–36
Sheinberg DL, Logothetis NK. 1997. The role of temporal cortical areas in perceptual orga-
nization. Proc. Natl. Acad. Sci. USA 94:3408–13
Sheinberg DL, Logothetis NK. 2001. Noticing familiar objects in real world scenes: the role
of temporal cortical neurons in natural vision. J. Neurosci. 21:1340–50
Sinha N, Brown JT, Carpenter RH. 2006. Task switching as a two-stage decision process. J.
Neurophysiol. 95:3146–53
Smith PL. 2000. Stochastic dynamic models of response time and accuracy: a foundational
primer. J. Math. Psychol. 44:408–63
Smith PL, Ratcliff R. 2004. Psychology and neurobiology of simple decisions. Trends Neurosci.
27:161–68
Smith PL. 1995. Psychophysically principled models of visual simple reaction time. Psychol.
Rev. 102:567–93
Smith PL. 1998. Bloch’s law predictions from diffusion process models of detection. Aust. J.
Psychol. 50:139–47
Sparks DL. 2002. The brainstem control of saccadic eye movements. Nat. Rev. Neurosci. 3:952–
64
Spivey MJ, Grosjean M, Knoblich G. 2005. Continuous attraction toward phonological com-
petitors. Proc. Natl. Acad. Sci. USA 102:10393–98
Stuphorn V, Taylor TL, Schall JD. 2000. Performance monitoring by the supplementary eye
field. Nature 408:857–60
572 Gold · Shadlen

ANRV314-NE30-21 ARI 21 May 2007 13:44
Sugrue LP, Corrado GS, Newsome WT. 2004. Matching behavior and the representation of
value in the parietal cortex. Science 304:1782–87
Sugrue LP, Corrado GS, Newsome WT. 2005. Choosing the greater of two goods: neural
currencies for valuation and decision making. Nat. Rev. Neurosci. 6:363–75
Tanaka K, Hikosaka H, Saito H, Yukie Y, Fukada Y, Iwai E. 1986. Analysis of local and wide-
field movements in the superior temporal visual areas of the macaque monkey. J. Neurosci.
6:134–44
Tanaka K, Saito H. 1989. Analysis of motion of the visual field by direction, expansion/
contraction and rotation cells clustered in the dorsal part of the medial superior tem-
poral area of the macaque monkey. J. Neurophysiol. 62:626–41
Tehovnik EJ. 1996. Electrical stimulation of neural tissue to evoke behavioral responses. J.
Neurosci. Methods 65:1–17
Tenenbaum JB, Griffiths TL. 2001. Generalization, similarity, and Bayesian inference. Behav.
Brain Sci. 24:629–40
Tremblay L, Schultz W. 1999. Relative reward preference in primate orbitofrontal cortex.

Nature 398:704–8
Tremblay L, Schultz W. 2000. Reward-related neuronal activity during Go-nogo task perfor-
mance in primate orbitofrontal cortex. J. Neurophysiol. 83:1864–76
Uchida N, Kepecs A, Mainen ZF. 2006. Seeing at a glance, smelling in a whiff: rapid forms of
perceptual decision making. Nat. Rev. Neurosci. 7:485–91

Uchida N, Mainen ZF. 2003. Speed and accuracy of olfactory discrimination in the rat. Nat.
Uka T, DeAngelis GC. 2003. Contribution of middle temporal area to coarse depth discrimi-
nation: comparison of neuronal and psychophysical sensitivity. J. Neurosci. 23:3515–30
Uka T, DeAngelis GC. 2006. Linking neural representation to function in stereoscopic depth
perception: roles of the middle temporal area in coarse versus fine disparity discrimination.
J. Neurosci. 26:6791–802
Ungerleider LG, Mishkin M. 1982. Two cortical visual systems. In Analysis of Visual Behavior,
ed. DJ Ingle, MA Goodale, RJW Mansfield, pp. 549–80. Cambridge, MA: MIT Press
Ullman JD. 1979. The Interpretation of Visual Motion. Cambridge, MA: MIT Press
Usher M, McClelland JL. 2001. The time course of perceptual choice: the leaky, competing
accumulator model. Psychol. Rev. 108:550–92
VanRullen R, Thorpe SJ. 2001. The time course of visual processing: from early perception to
decision-making. J. Cogn. Neurosci. 13:454–61
Verghese P, Watamaniuk SNJ, McKee SP, Grzywacz NM. 1999. Local motion detectors cannot
account for the detectability of an extended trajectory in noise. Vis. Res. 39:19–30
Vickers D. 1970. Evidence for an accumulator model of psychophysical discrimination. In
Contemporary Problems in Perception: Ergonomics, ed. AT Welford, L Houssiadas, pp. 37–
58. London: Taylor & Francis
von Helmholtz HLF. 1925. Treatise on Physiological Optics. New York: Dover
von Neumann J, Morgenstern O. 1944. The Theory of Games and Economic Behavior. Princeton:
Princeton Univ. Press
Wagner AD, Schacter DL, Rotte M, Koutstaal W, Maril A, et al. 1998. Building memories:
remembering and forgetting of verbal experiences as predicted by brain activity. Science
281:1188–91
Wald A, Wolfowitz J. 1947. Optimum character of the sequential probability ratio test. Ann.
Math. Statist. 19:326–39
Wallis JD, Anderson KC, Miller EK. 2001. Single neurons in prefrontal cortex encode abstract
rules. Nature 411:953–56

ANRV314-NE30-21 ARI 21 May 2007 13:44
Wang XJ. 2002. Probabilistic decision making by slow reverberation in cortical circuits. Neuron
36:955–68
Watanabe K, Lauwereyns J, Hikosaka O. 2003. Neural correlates of rewarded and unrewarded
eye movements in the primate caudate nucleus. J. Neurosci. 23:10052–57
Watanabe M. 1996. Reward expectancy in primate prefrontal neurons. Nature 382:629–32
Watson AB. 1986. Temporal sensitivity. In Handbook of Perception and Human Performance. ed.
K Boff, J Thomas, pp. 6.1–6.43. New York: Wiley
Westheimer G. 1954. Mechanism of saccadic eye movements. AMA Arch. Ophthalmol. 52:710–
24
Wong KF, Huk A, Shadlen MN, Wang XJ. 2005. Time integration in a perceptual decision
task: adding and subtracting brief pulses of evidence in a recurrent cortical network model.
Soc. Neurosci. Abstr. 621.5
Wong KF, Wang XJ. 2006. A recurrent network mechanism of time integration in perceptual
decisions. J. Neurosci. 26:1314–28

Yeung N, Sanfey AG. 2004. Independent coding of reward magnitude and valence in the human
brain. J. Neurosci. 24:6258–64
Zhang T, Britten KH. 2004. Clustering of selectivity for optic flow in the ventral intraparietal
area. NeuroReport 15:1941–45
Zhang T, Heuer HW, Britten KH. 2004. Parietal area VIP neuronal responses to heading
stimuli are encoded in head-centered coordinates. Neuron 42:993–1001
Zohary E, Shadlen MN, Newsome WT. 1994. Correlated neuronal discharge rate and its
implications for psychophysical performance. Nature 370:140–43
574 Gold · Shadlen

AR314-FM ARI 20 May 2007 16:26
Annual Review of
Neuroscience
Contents Volume 30, 2007
Information Processing in the Primate Retina: Circuitry and Coding

G.D. Field and E.J. Chichilnisky p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p1

Orbitofrontal Cortex and Its Contribution to Decision-Making
Jonathan D. Wallis p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 31
Fundamental Components of Attention
Eric I. Knudsen p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 57
Anatomical and Physiological Plasticity of Dendritic Spines
Veronica A. Alvarez and Bernardo L. Sabatini p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p 79
Visual Perception and Memory: A New View of Medial Temporal
Lobe Function in Primates and Rodents
Elisabeth A. Murray, Timothy J. Bussey, and Lisa M. Saksida p p p p p p p p p p p p p p p p p p p p p p p p 99
The Medial Temporal Lobe and Recognition Memory
H. Eichenbaum, A.P. Yonelinas, and C. Ranganath p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p123
Why Is Wallerian Degeneration in the CNS So Slow?
Mauricio E. Vargas and Ben A. Barres p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p153
The Head Direction Signal: Origins and Sensory-Motor Integration
Jeffrey S. Taube p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p181
Peripheral Regeneration
Zu-Lin Chen, Wei-Ming Yu, and Sidney Strickland p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p209
Neuron-Glial Interactions in Blood-Brain Barrier Formation
Swati Banerjee and Manzoor A. Bhat p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p235
Multiple Dopamine Functions at Different Time Courses
Wolfram Schultz p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p259
Ventral Tegmental Area Neurons in Learned Appetitive Behavior and
Positive Reinforcement
Howard L. Fields, Gregory O. Hjelmstad, Elyssa B. Margolis,
and Saleem M. Nicola p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p289
v
AR314-FM ARI 20 May 2007 16:26
Copper and Iron Disorders of the Brain

Erik Madsen and Jonathan D. Gitlin p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p317
The Micromachinery of Mechanotransduction in Hair Cells
Melissa A. Vollrath, Kelvin Y. Kwan, and David P. Corey p p p p p p p p p p p p p p p p p p p p p p p p p p p p p339
Neurobiology of Feeding and Energy Expenditure
Qian Gao and Tamas L. Horvath p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p367
Mechanisms that Regulate Establishment, Maintenance, and
Remodeling of Dendritic Fields
Jay Z. Parrish, Kazuo Emoto, Michael D. Kim, and Yuh Nung Jan p p p p p p p p p p p p p p p p p399
Dynamic Aspects of CNS Synapse Formation
A. Kimberley McAllister p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p425

Adhesion Molecules in the Nervous System: Structural Insights into
Function and Diversity
Lawrence Shapiro, James Love, and David R. Colman p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p451
Development of Neural Systems for Reading

Bradley L. Schlaggar and Bruce D. McCandliss p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p475
Molecular Architecture of Smell and Taste in Drosophila
Leslie B. Vosshall and Reinhard F. Stocker p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p505
The Neural Basis of Decision Making
Joshua I. Gold and Michael N. Shadlen p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p535
Trinucleotide Repeat Disorders
Harry T. Orr and Huda Y. Zoghbi p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p575
Indexes
Cumulative Index of Contributing Authors, Volumes 21–30 p p p p p p p p p p p p p p p p p p p p p p p p623

Cumulative Index of Chapter Titles, Volumes 21–30 p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p p627
Errata
An online log of corrections to Annual Review of Neuroscience chapters (if any, 1997
to the present) may be found at http://neuro.annualreviews.org/
vi Contents

Will Bases de La Toma de Desiciones

Uploaded by

Copyright:

Available Formats

Will Bases de La Toma de Desiciones

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Will Bases de La Toma de Desiciones

Uploaded by

Copyright:

Available Formats

ANRV314-NE30-21 ARI 21 May 2007 13:44

The Neural Basis of

Annu. Rev. Neurosci. 2007. 30:535–74 Key Words

of this approach for inferring principles of

(often n = 2) that each represent a state of

536 Gold · Shadlen

In the world In the brain

(4 possible) Decision variable

Apply decision rule Establish

Evaluation Experience payoff or cost

www.annualreviews.org • Decision Making 537

538 Gold · Shadlen

www.annualreviews.org • Decision Making 539

a Sequential analysis framework

Symmetric random walk

540 Gold · Shadlen

positive or negative criterion (the bounds in

able tool for psychophysical analysis, particu- ⎪

www.annualreviews.org • Decision Making 541

rule. Analogous mechanisms in the brain are Perceptual Tasks

542 Gold · Shadlen

www.annualreviews.org • Decision Making

544 Gold · Shadlen

the delay period. on other perceptual tasks, including even a

www.annualreviews.org • Decision Making 545

100 200 400 800 -5 0 5 10

Viewing duration (ms) X position (º)

d 1.5 51.2% coh

546 Gold · Shadlen

www.annualreviews.org • Decision Making 547

548 Gold · Shadlen

www.annualreviews.org • Decision Making 549

a Stimulate rightward MT neurons

Relatively large effect on choice and RT:

Reaction time rightward choices

b Stimulate right choice LIP neurons

Momentary evidence in MT DV in LIP

550 Gold · Shadlen

www.annualreviews.org • Decision Making 551

microstimulation intended to bias judgments a RDM stimulus viewed stereoscopically

MST represents only the momentary evi- effects on performance. Microstimulation at

552 Gold · Shadlen

www.annualreviews.org • Decision Making 553

554 Gold · Shadlen

Firing rate (sp s–1)

Firing rate (sp s–1)

Missed False alarms

www.annualreviews.org • Decision Making 555

556 Gold · Shadlen

www.annualreviews.org • Decision Making

a mere trigger, perhaps necessarily including

tor latencies (Figure 9). Latency distributions

distribution of reciprocal latencies plotted on

558 Gold · Shadlen

www.annualreviews.org • Decision Making 559

Value-Based Decisions LIP that represent the DV on perceptual tasks

though some studies suggest that quantities

Padoa-Schioppa & Assad 2006, Platt & Glim-

560 Gold · Shadlen

www.annualreviews.org • Decision Making 561