Keywords

1 Introduction

While incentives can cause behavior modifications [19] automated systems report successfully changing behavior or beliefs [4, 7]. Human energy usage is substantial and is implicated in global warming [12]. Indeed, 30 % of energy used by a building is caused by people’s actions that could easily be modified or reduced in the building [20].

Can persuasion technology change people’s behavior and affect energy usage in a community? We tested this with the SweetFeedback Reward System, a computer- controlled gumball delivery system and environmental sensing platform [10]. It was designed to be easily deployed on people’s desks throughout a building to create an incrementally expandable sensor and feedback network. This paper uses the SweetFeedback platform in public instead of the personal space use it was designed for.

Behavior change requires, among other things, a recognition that it matters. The Fogg Behavior Model [6] predicts that the requirements for a person to perform an identified behavior include (1) enough motivation, (2) the user’s ability to perform the behavior, and (3) the need for a trigger in order to guarantee that people will achieve the target behavior. All three factors should be present for targeted behavior to be performed and learned by a person.

1.1 Related Work

A number of research projects aim to perform persuasion technology with tangible feedback. BinCam installed a camera underneath a garbage bin lid and uploaded the photos taken onto Facebook whenever a user threw garbage into BinCam [22]. The system worked to persuade users to engage in self-reflection of their waste management and recycling habits through social media. The Chocolate Machine applied Ego Depletion theory to show that people can be trained to resist eating by dispensing the chocolate at scheduled intervals [13]. Utmi engaged students on campus to help with exam grading tasks to successfully use a special public vending machine to give tangible non-monetary feedback [9]. Utami attracted many students to resolve expert tasks, and they performed better than a single student expert. Our work focuses on also using tangible feedback, but also uses auditory and visional techniques to attract participants.

Many previous studies have revealed the major challenges: how to entice passers-by to approach the display, how to make them aware of the interactive features, and finally how to motivate them to actively interact with it. The “first click” enticement problem must be solved to achieve any of the other goals [3, 11, 14, 15]. To begin addressing this problem requires understanding the nature of public engagement interactives. A public display must work with a first-time user and a user that has seen it before [11]. Passers-by are typically carrying out other activities [3]. Simply providing attractive appearance, utility, usability, and likability may not be enough to draw people to engage with it. Unlike other computing technologies, interaction with public displays does not start with the interaction itself [15].

To get a person to engage, (1), they must notice the interactive, (2), they must see the interactive as interesting to them, and (3), they must decide that they can interact with it without fear of public embarrassment, intimidation, disrupting their intended activity, or offending someone they are with [14]. They must then understand how to interact and commit to interacting. Finally, they must follow though to get the value of the interaction. Similar to this, Brignull & Rogers [3] defined three activity spaces in respect to the flow of interaction within a crowd: peripheral awareness, where people notice the display; focal awareness, where people engage in social activities associated with the display, such as talking or gesturing, or in other words performing indirect interaction; and direct interaction, where people actively interact with the display.

Several studies about how to improve making a public display noticeable and enticing [14] showed text more effective than icons, color more effective than greyscale, and static imagery more effective than animation. Ju et al. [11] found that physical gestures significantly attract more looks and use of an information kiosk than verbal gestures alone. Oftentimes, the audience doesn’t seek the public displays actively, but discovers it incidentally, which Dix [5] called “incidental interaction”. Spatial configuration indeed matters; a public display is better put near traffic flow in order to ensure a steady stream of people [15] and may trigger more incidental interactions.

Peripheral awareness, regardless of how it is gained, has a significant impact on people’s initial understanding of a public display. Interactive displays often fail to deal appropriately with the social inhibition associated with interaction in public [15]. “Display blindness” is a phenomenon related to audience expectation: if the audience expects that the display shows uninteresting content, they tend to ignore it [8]. Another phenomenon discovered by Kukka et al. [14] is “display avoidance,” when people intentionally avoid looking at the display to avoid a deprioritized engagement.

Those phenomena are highly related to the social context. Akpan et al. [2] explored the role of social context in the success of an interactive public display. The idea is based on the notions of “space” and “place” [8]. A space is a physical, spatial structure which affects humans’ activity and behavior. A place, on the other hand, attaches social meanings, which may include understanding, experience, cultural expectations and norms, and patterns of behavior. SweetFeedback first focused on the desk place. The SweetBuildingGreeter tests it in a vestibule space. Past studies show how individual’s interpretation of “place” instead of “space” to a large degree determined whether they would actively engage with a display [8]. The most successful places to engage passers-by were those where the social context supports “comfort space” or “license to play” with the interactive. This is a challenge that SweetBuildingGreeter must face.

Previous studies have also explored various technical and social aspects regarding engagement. The contribution of social context within space and place has shown interactivity often relies on incidental interactions; however, “forcing” passers-by to notice may fail to move people from being spectators to being active users [2]. This limits the explorations as it may disturb the nature of interactivity in the place. Our approach is to extend the goal of interaction in a space to appropriately engage one or a few people [3]. Our work moves beyond screen color or/and content, attempting to entice a passerby to view a display in a space. Can persuasive words and physical feedback entice people to perform important altruistic community service?

How to engage passers-by in a public area, with the goal of helping save energy? Our first model required a user to log in with a QR code on their phone, then close a window to receive a candy reward [10]. It added steps and time to the simple act of closing a window. An alternative might be a simple flag that flies into the hall attracting attention to the open window more concretely. Could SweetFeedback be as concrete and more rewarding?

2 The SweetBuildingGreeter Scenario and System

SweetBuildingGreeter system is divided into several components in Fig. 1. The system utilizes the SweetFeedback reward system [10, 21], which is a gumball dispenser device with an Arduino board. It runs a program to sense and react to the environment. A client on a PC communicates between SweetFeedback and the supervisor server. A server program makes decisions from a user’s behavior, environmental data, and interaction. A web graphical interface presents information to the user (Fig. 2).

Fig. 1.
figure 1

The SweetFeedback reward framework.

Full size image
Fig. 2.
figure 2

The small poster describes how to play with the system in just three steps.

Full size image
Fig. 3.
figure 3

Screensaver comics to make people aware of energy

Full size image

A PC complements the Arduino to add an interactive display, voiced comments, a people-recognizing camera, and Bluetooth network communication. Processing.org is used to write a Java program to manage the communication between the gumball machine and the PC [18]. For example, when the server sends a capital B to the gumball machine, the gumball machine sends calibrated sensor data to the server’s database repository. This program reads sensors and manages the timing to give audio or tangible feedback to users. When a user wins a reward, the program sends a capital A to the serial port to control the dispensing motor. The server display might also show rewards by an animation.

As well as the SweetFeedback sensors, the PC scans nearby Bluetooth-based devices for identifying users and integrating a transportation app. OpenCV [16] is used to interpret webcam data to detect the presence of a person or people. The camera program was further developed to differentiate when to attempt to engage people by applying blob detection. If it notices that there are more than two people, it does not try to intrude or distract them. Without a model of peer pressure, our theory is that people in groups are less likely to be interested or able to take their attention away from the other people to be able or willing to actually focus on a building problem, especially one that is trying to change their behavior.

SweetBuildingGreeter stores user and building models and behavior in a server. The server is implemented in a Python Flask framework, and it provides several restful API to client applications. A MySQL database stores user behavioral data. The feedback mechanism, environment problem detection, and verification is evaluated by this server. Whenever a user fixes an environment problem, such as turning off the light, the server knows it by acquiring the sensor data from the sensor data repository to check if the problem still exists. Once the problem is solved, the server waits for the person to go back to the gumball machine and rewards him or her with visual imagery, sound, or candies. This application server also provides a webpage as a user interface (Figs. 4 and 5). Several iterations made a simpler and clearer interface, with user messages that are easier to understand. The main view of the interface is comprised of a map diagram of the building and a caricatured drawing of a gumball machine. The map diagram shows the location of a problem with a blinking red spot, to let the user identify where to go to solve the problem. Also, it visualizes the sensor data by showing people and light bulbs on the map. The drawing of the gumball machine has speaking bubbles over it, asking the person to do things or answer questions.

Fig. 4.
figure 4

The problem map

Full size image
Fig. 5.
figure 5

The questionnaire

Full size image

The sensor data from the gumball machine is stored in a local database, including light data, temperature, sound level data and window open/close data. Furthermore, we started with a third-party sensor repository from another research group to get all the environmental information for the campus. They provided several kinds of restful APIs for a variety of sensor data queries, especially for the real-time light sensor and motion sensor data. Whenever the server needs to check for environmental problems, the server checks the third-party sensor data repository and gets real-time data.

2.1 Persuasive Application Design

“Turn off the light” was our example scenario to test our persuasive system. The server program checks the status of local sensor repository and third-party sensor network repository periodically and performs simple environmental diagnosis to verify if a light is in correct state. If a motion sensor detects that there is no one in a room but the light is on, the server program will show the problem on the map. We want passers-by to notice the problems on the screen, use it to guide them to solve them, and get a reward. When a user chooses a target problem and clicks the button, “Yes, I will take it”, our server program begins to monitor the target light. When it detects the problem has been solved, the server program sends a reward message. The server program will estimate how long it takes for the problem to be solved (walking to the room to turn off the light and walking back to the machine) after the button “Yes, I will take it” is pressed. It senses with Bluetooth or time that the person is back and deliver the reward when appropriate.

2.2 Questionnaire Application

When there is no pressing environmental issue in the building, the system teaches users about their building, campus environment and energy use. The drawing on the side of graphical user interface shows a question for passers-by (Fig. 5), with four possible answers. If the user does not select the correct answer, we enable them to retry until they get the right one. They then get rewarded with candy.

3 A Persuasive Behavioral Model

Initially, the SweetFeedback reward system was designed to be a personal persuasion process. For use in a public space, it was deployed with a secured laptop, a cloth draped stand, and a sign. OpenKiosk was installed in the laptop [17], which simplified the way users interact with SweetBuildingGreeter’s web interface.

After several days of short experiments, we found that only a few people would use the system. We thought maybe it was because they didn’t know they could use it or how to use it, so we made a small poster describing the flow of the process (Fig. 3). It attempted to attract people and show them that they are welcome to use the machine.

The process was: (1) Arrive - touch the pad to dismiss the energy slideshow screen saver; (2) Do - solve the environmental problem or answer the questionnaire; (3) Joy - if 1 and 2 are done correctly, get the reward.

We then wondered if changing the content of the poster to tell people to try the new system would make people curious about the new system and come to try it out. So we waited for a few days and changed the previous poster to a new poster with a cute gumball machine saying, “Try out the new experiment.”

The system provides motivations for users based on a behavioral model to persuade energy-efficient behaviors; the gumball dispenser device will give candies as the major reward if users follow the interaction flow for SweetBuildingGreeter. The biggest changes in moving from a personal to a public persuasive system were motivation for engagement, enough trigger and higher simplicity for the performance of persuasive behavior. There was also a camera with an OpenCV program that can distinguish between one person approaching or several people approaching. The gumball machine tries to be noticed, to engage a person by triggering the voice on the laptop when it detects them approaching.

In idle time, the kiosk display also showed motivational comics found from 9GAG [1], Google, with the message of energy-saving and green technology, to motivate users to think about their energy behaviors. These comics were more attractive to passers-by than the screen saver that only displayed a simple message like, “Save energy and get reward.” To better attract people, the screen saver was turned into a comic slide show with images about energy saving and candy, and with words inviting them to try it out by touching the trackpad (Fig. 3).

When the camera detects a passerby, the system would attempt to beckon them with two different audio probes tested: “Help me!” or “Try out the new experiment and save energy!” We added a voice to attract the person.

4 Experiments

The effectiveness of SweetBuildingGreeter in engaging people and persuading energy-saving behavior was first tested at a small campus with about 300 people including MS students, PhD students, staff, and visitors. A participant sees a visualization of the building’s environmental condition (Fig. 4); a bubble states how to solve a problem. The supervising server monitors changes depending on the environmental condition, and makes decisions about giving rewards. After the interaction, the participant could fill out a survey to get more candy. The survey asked how participants felt about the system, the least useful feature in the system, what got their attention to interact, the rating of each feature, etc. We also spent a few mornings observing how different interventions (slide show or voice) affected how many passers-by would approach it. Finally, we conducted a user study and interviewed people about their experiences with the system.

5 Results

The system was deployed for 6 weeks. The data shows that in the beginning many people were attracted to our system (Fig. 6). Engagement decreased as people were repeatedly exposed to it, presumably because the novelty had worn off. The rate of correctly answered questions improved over time, shown in Fig. 6. This indicates that people did learn about their campus and building environment.

Fig. 6.
figure 6

New student first tried the system on 8/23. The number of wrong answers decreased because people learned and shared the information. Usage also decreased.

Full size image

As the results show (see Table 1), version 1 and 3 did not attract anyone to the system with a poster. Voice worked better in triggering the behavior, as we see in versions 2 and 4; we succeeded in beckoning 2 out of 15 and 5 out of 20 passers-by respectively to look at our system. Version 4 was the most successful at beckoning, perhaps because it also said that there was a new experiment. Version 4 also had 3 out of 20 passers-by not only look at, but actually interact with our system and complete the process to save energy. These results show that the voice, a broadcast medium, is a more powerful feature to beckon people than the poster.

Table 1. Attracting people U.S. Version 1: a poster only. Version 2: “Help me.” Version 3: New poster without the voice. Version 4: “Try the experiment and save energy.”
Full size table

All participants were asked to help us fill out a survey to provide their feelings and suggestions about SweetBuildingGreeter; 16 were collected. Participants rated the poster as the least useful feature. The blue bar charts show what did get their attention (see Fig. 7). Surprisingly, most of the participants did not state that the voice had attracted them. They stated that the slide show or the promise of candies had attracted them, when our data shows that it was the voice promising something new that brought them to the system.

Fig. 7.
figure 7

Blue bars: “What got their attention.” Red bars: “The least useful feature.” (Color figure online)

Full size image

Our final set of experiments were performed with students in Taiwan. It included versions 1, 2, and 4 (Table 2). All Taiwan students encountered a building problem (a window that needed to be closed). 7 out of 23 passers-by took a look at our system in version 1. More than half the passers-by looked at our system in versions 2 and 4. 4 out of 20 passers-by in version 2 and 2 out of 30 in version 4 used the system to close the window (Fig. 8).

Table 2. Attracting people Taiwan. Version 1: a poster only. Version 2: “Help me.” Version 4: “Try the experiment and save energy.”
Full size table
Fig. 8.
figure 8

“Why did you use the Sweet Building Greeter today?” People said they wanted candy but the beckoning sound making a request got them to come.

Full size image

Still, because they had previously not associated any opportunity they could do for the building with the previous system, they may have thought they had already used the gumball machine and didn’t imagine they would get anything new by taking time with it again.

We then tried a new auditory message, “Try out the new experiment and save energy.” Of 20 people, 3 actually came over and completed the process of turning off the light in the other room. In this case, people who had not responded to earlier systems responded to the attractor to change behavior and used it to solve a building problem. One might see 3 of 20 as a low number, but it is not, for the goal of the system is to solve simple problems that sometimes occur when people don’t take responsibility for the building. We realized that in such things, the light or thermostat only needs to be dealt with episodically.

6 Discussion

We have described a platform for increasing awareness of energy use and promoting behavior to save energy in a public space. The system could use audio feedback that sounds like candy being dispensed… or some other sound or stated affirmation instead of candy, so as not to overwhelm users with too much candy in their experience.

Adding a camera that made something happen when someone arrived made a huge improvement in people’s willingness to attend to the kiosk. The system evolved to include a continually animating attractor that shows where the problems are in the building graphically.

The addition of a quiz about energy use in the building and on campus succeeded in attracting people.

Adding a slide-show of comics that would run between interactions made the system more noticeable, but an audio attractor was the most effective beckoner to get them to interact.

Our camera is designed to welcome people that are alone, but not people that are together. We anticipate that such groups don’t want to wait around for one member to go turn off the lights for others.

The system did make people consider the environment and learn about their campus. A student saw a question asking about the motto of the campus. He turned to his friends and discussed the meaning of the motto, even though he already knew the motto.

The last experiment tested a subtle change to the audio beckon, from a request, “Come over and help me” to an opportunity, “Try out the new experiment and save energy.” This caused people who had already seen the system to stop and work with it.

7 Conclusion

The overall purpose of SweetBuildingGreeter is solving building problems, not attracting every user. If the system is working correctly, the lights might be turned off a few times a day by the system’s intervention. Statements of opportunity instead of requests were crucial. The gumball machine was shown to be useful in many ways; the installation was a community focus that people used to learn about their campus and become more aware of energy saving.

As we contemplate the many persuasive applications that have been built and tested, we come to important questions of place and space. Which interactivities are useful for a private desk experience and which are useful for a public experience? For the desk environment, SweetFeedback includes a number of personal performance monitoring and feedback applications. If the feedback occurs too often or requires a person to engage too often it will be distracting. The public interface would most naturally present different and social experiences. The SweetBuilding interface may succeed by making everyone aware of its ability to remind them of group commitments and a culture of energy savings. This experiment showed that audio beckoning was a necessary part of this peripheral computing system’s mode of communication to attract participants to interact with the system to save energy. Public interfaces for green interaction can succeed at episodically engage while staying a focus of interest and energy savings. We further show that even when tangible rewards are present, active graphics and motivational sound are helpful at attracting users.