Hc ai-asu-dc-shneiderman-3-13-2020-v28
•Download as PPT, PDF•
0 likes•925 views
This document summarizes a presentation on human-centered artificial intelligence. The presentation argues that AI should amplify, augment, enhance and empower people rather than replace them. The goal is to develop AI that is reliable, safe and trustworthy, and provides 1000-fold improvements in human capabilities while ensuring human control. It discusses balancing human control and computer automation, with the ideal being high levels of both through strategies like informative feedback, transparency and accountability. The overarching message is that AI should prioritize empowering and serving all of humanity.
Hc ai-asu-dc-shneiderman-3-13-2020-v28
- 1. Human-Centered Artificial Intelligence: Reliable, Safe & Trustworthy Ben Shneiderman @benbendc Founding Director (1983-2000), Human-Computer Interaction Lab Professor, Department of Computer Science Member, National Academy of Engineering Arizona State University, DC, March 13, 2020 Photo: BK Adams
- 2. Interdisciplinary research community - Computer Science & Info Studies - Psych, Socio, Educ, Jour & MITH hcil.umd.edu vimeo.com/72440805
- 3. Designing the User Interface Design Theories Direct manipulation Menus, speech, search Social Media Information Visualization www.cs.umd.edu/hcil/DTUI6 Sixth Edition: 2016
- 4. Web links Tiny touchscreen keyboards Photo tagging
- 7. Rationalism vs Empiricism Rationalism: logical perfectability of human endeavors Aristotle, Descartes, Spinoza & Kant Empiricism: need to continuously refine our beliefs based on our experiences Leonardo, Galileo, Locke & Hume
- 8. Rationalism vs Empiricism Rationalism: logical perfectability of human endeavors Aristotle, Descartes, Spinoza & Kant ... Ronald Fisher Empiricism: need to continuously refine our beliefs based on our experiences Leonardo, Galileo, Locke & Hume ... John Tukey
- 9. Rationalism vs Empiricism Rationalism: logical perfectability of human endeavors Aristotle, Descartes, Spinoza & Kant ... AI Empiricism: need to continuously refine our beliefs based on our experiences Leonardo, Galileo, Locke & Hume ... HCI
- 10. Machines That Think June 30, 1980n Machines Think? March 25, 1996 Can Machines Think? March 25, 1996
- 13. • Privacy • Accountability • Safety & security • Transparency & explainability • Fairness & non-discrimination • Human control of technology • Professional responsibility • Promotion of human values • International human rights Fjeld, Achten, Hilligoss, Nagy, Srikumar (2020) https://cyber.harvard.edu/publication/2020/principled-ai Principled Artificial Intelligence: Governance Summarizes 36 leading reports on AI Governance
- 14. HCI Pride: Serving 6 Billion Users Mobile, desktop, web, cloud Diverse users: include everyone Diverse applications: real problems, users, data Diverse interfaces: theories, principles, guidelines
- 15. Mobile, desktop, web, cloud Diverse users: include everyone Diverse applications: real problems, users, data Diverse interfaces: theories, principles, guidelines HCI Pride: Serving 6 Billion Users User Control … people - not apps - are in control Flexibility … (give) users complete, fine-grained control over their work Apple Design Guidelines
- 16. AI design principles, which I want to change • Humanoid robots • Excessive automation • Insufficient feedback • Hidden features • Restrictions on creativity • Limits on collaboration • Missing audit trails
- 17. Human-Centered AI Amplify, Augment, Enhance & Empower People
- 18. Human-Centered AI Amplify, Augment, Enhance & Empower People ïƒ 1000-fold improvements in capabilities Photography Navigation Business Formation Information Search Email & Text
- 19. Human-Centered AI Amplify, Augment, Enhance & Empower People ïƒ 1000-fold improvements in capabilities ïƒ RST: Reliable, Safe & Trustworthy Photography Navigation Business Formation Information Search Email & Text
- 20. Human-Centered AI Amplify, Augment, Enhance & Empower People ïƒ 1000-fold improvements in capabilities ïƒ RST: Reliable, Safe & Trustworthy ïƒ More people, more creative, more often Photography Navigation Business Formation Information Search Email & Text
- 21. Human-Centered AI Amplify, Augment, Enhance & Empower People ïƒ 1000-fold improvements in capabilities ïƒ RST: Reliable, Safe & Trustworthy ïƒ More people, more creative, more often ïƒ Self-efficacy & societal benefits Photography Navigation Business Formation Information Search Email & Text
- 23. Designing the User Interface First Edition: 1986 Balancing automation & human control
- 24. Designing the User Interface First Edition: 1986 Balancing automation & human control Human control Computer automation
- 25. Designing the User Interface Sixth Edition: 2016 Ensuring human control while increasing automation
- 26. Designing the User Interface Sixth Edition: 2016 Ensuring human control while increasing automation Computer Automation High Human Control HighLow Low
- 27. Computer Automation HighLow Reliable, Safe & Trustworthy Human-Centered AI High Low Human Control
- 28. HighLow Elevator Camera Human-Centered AI Reliable, Safe & Trustworthy High Low Human Control Computer Automation
- 29. HighLow Airbag Elevator Camera Pacemaker Computer Control Human-Centered AI Reliable, Safe & Trustworthy High Low Human Control Computer Automation
- 30. HighLow Computer Control Human Mastery Bicycle Piano Elevator Camera Airbag Pacemaker Human-Centered AI High Low Human Control Computer Automation Reliable, Safe & Trustworthy
- 31. HighLow Human Mastery Bicycle Piano Excessive Human Control ExcessiveAutomation Airbag Pacemaker Computer Control Elevator Camera Human-Centered AI High Low Human Control Computer Automation Reliable, Safe & Trustworthy
- 32. Excessive Human Control ExcessiveAutomation HighLow Pain control designs Human Mastery Computer Control Morphine drip bag High Low Human Control Computer Automation Reliable, Safe & Trustworthy
- 33. Excessive Human Control ExcessiveAutomation HighLow Pain control designs Human Mastery Computer Control Morphine drip bag Automatic dispenser Reliable, Safe & Trustworthy High Low Human Control Computer Automation
- 34. Excessive Human Control ExcessiveAutomation HighLow Pain control designs Human Mastery Patient- controlled Computer Control Morphine drip bag Automatic dispenser Reliable, Safe & Trustworthy High Low Human Control Computer Automation
- 35. Excessive Human Control ExcessiveAutomation HighLow Pain control designs Human Mastery Patient- controlled Patient-guided Clinician-monitored Computer Control Automatic dispenser Morphine drip bag Reliable, Safe & Trustworthy High Low Human Control Computer Automation
- 36. Excessive Human Control ExcessiveAutomation HighLow Human Mastery 1980 Car Computer Control Car control designs Reliable, Safe & Trustworthy High Low Human Control Computer Automation
- 37. Excessive Human Control ExcessiveAutomation HighLow 2020 Car Human Mastery 1980 Car Computer Control Car control designs Reliable, Safe & Trustworthy High Low Human Control Computer Automation
- 38. Excessive Human Control ExcessiveAutomation HighLow Car control designs Human Mastery 1980 Car 2040 Car Computer Control 2020 Car Reliable, Safe & Trustworthy High Low Human Control Computer Automation
- 39. Summary of Design Strategies Goal: High levels of human control & computer automation ïƒ RST: Reliable, Safe & Trustworthy ïƒ Comprehensible, predictable & controllable ïƒ Continuous display of status ïƒ Informative feedback
- 40. Predictive Search Keeps Users in Control
- 41. Mercedes-Benz Parking Assist Contextually Rich Informative Feedback https://youtu.be/NXIMjL39c2Q
- 42. February 5, 2019 https://www.pnas.org/content/116/6/1844 to productively employ AI methods while ensuring that people remain in control: unconstrained in pursuing complex goals and exercising domain expertise
- 45. Accountability, Liability, Fairness, Transparency, Explainability & Interpretability
- 46. Design systems using reliable strategies Audit trails to review failures & near misses Benchmark tests for verification & validation Continuous improvement & process monitoring Data quality & bias testing Explainable user interfaces Accountability, Liability, Fairness, Transparency, Explainability & Interpretability
- 47. Seek culture of safety by open management strategies: Leadership commitment to safety Open discussion of failures & near misses Internal oversight boards for problems & future plans Public reports of failures & future plans Accountability, Liability, Fairness, Transparency, Explainability & Interpretability
- 48. Develop trustworthy processes for independent oversight Government regulation Insurance companies require standards Accounting firms audit systems Voluntary industry plans: Partnership on AI Professional societies promote education & ethics Accountability, Liability, Fairness, Transparency, Explainability & Interpretability
- 49. Human-Centered AI Amplify, Augment, Enhance & Empower People ïƒ 1000-fold improvements in capabilities ïƒ RST: Reliable, Safe & Trustworthy ïƒ More people, more creative, more often ïƒ Self-efficacy & societal benefits Photography Navigation Business Formation Information Search Email & Text
- 50. Excessive Human Control ExcessiveAutomation HighLow Car control designs Human Mastery 1980 Car 2040 Car Computer Control 2020 Car Reliable, Safe & Trustworthy High Low Human Control Computer Automation
- 51. hcil.umd.edu HCIL 37th Annual Symposium: May 28, 2020