Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
SlideShare a Scribd company logo

Have you AI'ed today? A Reality Check

•
•

Charles Mok discusses the potentials and limitations of artificial intelligence (AI). While AI shows promise in areas like diagnosing diseases, fighting climate change, and translating languages, it also faces limitations. AI may struggle with tasks requiring "System 2" thinking like abstract reasoning, planning, and understanding cause and effect. It can also replicate and even amplify the biases of its human creators. To realize AI's full potentials, Mok argues we must know its limitations, impose proper oversight and controls, ensure ethics and equality are incorporated, and take a realistic approach focused on specific tasks rather than assuming it can solve all problems.

1 of 18
Download to read offline
"Have You AI'ed Today? A Reality Check" The Potentials, Limits and Pitfalls of AI Charles Mok IET ICT Conference 2020 20112020
2AI or ‘human-assisted’ AI?
The AI promise: Solution for all types of problems? 3
Possibilities Can AI do anything and everything for us?
Artificial intelligence (AI and machine learning (ML changing our world ● Diagnosing diseases with greater accuracy ● Fight climate change and pollution ● Free humans from driving ● Detect online piracy ● Revolutionize art and creativity ● Get help 24/7 with virtual personal assistants ● Translate any language instantly 5
The AI boom is growing, fast ● Ability to spot patterns in tranches of data with ‘superhuman accuracy’ ● Algorithms that can train itself with little or no data ● Businesses rely on AI as catalyst for innovation 6
What can’t AI do well (yet)? Think truly like a human ● Planning ● abstract reasoning ● understanding cause and effect ● open-ended generalization ⟶“System 2” thinking 7
Limitations Where does AI falls short?
Lost in translation? AI translations and language processing is not perfect sometimes… it’s OK 9
Tiny changes, big problems? 10 When AI struggles in handling oddities in self-driving cars, the risk is much higher
Biased data, more prejudices? ● humans are the ones making the algorithms ● humans are the ones feeding those algorithms data ● Are we creating biased algorithms based on biased data? ● Privacy issues from gathering personal data? 11
Pitfalls Can AI be trusted to make important decisions?
Is it fair? Algorithm assigning A-level grades created controversies in the UK 13
Gender and racial bias: the impact is real ● AI systems can unfairly penalize women and minorities ● Researchers find lower accuracy in facial recognition algorithms when detecting non-white facial images ● Training data deficiencies? 14
Where do we go from here? 15 Ethical AI, trusted AI, responsible AI?
AI may be powerful and clever, but it is not immune to mistakes 16
How can AI reach its full potentials? Know what AI can’t do We still need to impose checks and control, and incorporate concepts/values such as ethics, equality Less ambitious, more realistic? Re-calibrating our AI ambitions to achieve specific tasks More policies on AI EU’s model: serve as an example of having policy framework and tools for governments and companies? 17
Thanks! Stay in touch: ● FB @charlesmokoffice ● Twitter: @charlesmok ● LinkedIn: @charlesmok 18

More Related Content

Have you AI'ed today? A Reality Check

  • 1. "Have You AI'ed Today? A Reality Check" The Potentials, Limits and Pitfalls of AI Charles Mok IET ICT Conference 2020 20112020
  • 3. The AI promise: Solution for all types of problems? 3
  • 4. Possibilities Can AI do anything and everything for us?
  • 5. Artificial intelligence (AI and machine learning (ML changing our world ● Diagnosing diseases with greater accuracy ● Fight climate change and pollution ● Free humans from driving ● Detect online piracy ● Revolutionize art and creativity ● Get help 24/7 with virtual personal assistants ● Translate any language instantly 5
  • 6. The AI boom is growing, fast ● Ability to spot patterns in tranches of data with ‘superhuman accuracy’ ● Algorithms that can train itself with little or no data ● Businesses rely on AI as catalyst for innovation 6
  • 7. What can’t AI do well (yet)? Think truly like a human ● Planning ● abstract reasoning ● understanding cause and effect ● open-ended generalization ⟶“System 2” thinking 7
  • 9. Lost in translation? AI translations and language processing is not perfect sometimes… it’s OK 9
  • 10. Tiny changes, big problems? 10 When AI struggles in handling oddities in self-driving cars, the risk is much higher
  • 11. Biased data, more prejudices? ● humans are the ones making the algorithms ● humans are the ones feeding those algorithms data ● Are we creating biased algorithms based on biased data? ● Privacy issues from gathering personal data? 11
  • 12. Pitfalls Can AI be trusted to make important decisions?
  • 13. Is it fair? Algorithm assigning A-level grades created controversies in the UK 13
  • 14. Gender and racial bias: the impact is real ● AI systems can unfairly penalize women and minorities ● Researchers find lower accuracy in facial recognition algorithms when detecting non-white facial images ● Training data deficiencies? 14
  • 15. Where do we go from here? 15 Ethical AI, trusted AI, responsible AI?
  • 16. AI may be powerful and clever, but it is not immune to mistakes 16
  • 17. How can AI reach its full potentials? Know what AI can’t do We still need to impose checks and control, and incorporate concepts/values such as ethics, equality Less ambitious, more realistic? Re-calibrating our AI ambitions to achieve specific tasks More policies on AI EU’s model: serve as an example of having policy framework and tools for governments and companies? 17
  • 18. Thanks! Stay in touch: ● FB @charlesmokoffice ● Twitter: @charlesmok ● LinkedIn: @charlesmok 18