Shannon Vallor - Who is responsible for responsible AI?

Transcript

1. Who is Responsible for Responsible AI? THE ECOLOGIES OF A RESPONSIBLE AI ECOSYSTEM 29.07.23 Prof. Shannon Vallor The University of Edinburgh Turing Fest 2023
2. WHAT IS RESPONSIBLE AI? 2
3. o Trustworthy o Fair o Just o Accountable o Transparent o Safe o Beneficial RESPONSIBLE AI INNOVATION IS:
4. WHY RESPONSIBLE AI? 4
5. AI represents a further consolidation of social power in the hands of tech companies, developers and users
6. Social power without responsibility degrades public trust and confidence in innovation
7. Innovation without responsibility: o endangers the ‘social license to operate’ o inhibits adoption; skews it to reckless actors o breeds a vicious cycle of social harms o incentivizes a short-term ‘race to the bottom’ o impedes public support for future innovation
8. TWO MODELS OF INNOVATION 8
9. 1. Move fast and break things (social media model)
10. 2. Innovate boldly but responsibly (aviation, bioengineering)
11. The cost to society of Model 1 for social media was high. We’ve got lasting damage: to democratic norms, to trust in institutions, to social and civic cohesion, to political rationality
12. Do we really want to do that again?
13. AI isn’t the new oil. It’s the new steel. Photos from top left by Dakota Roos, Pierre Pavlovic, and Ricardo Gomez on Unsplash
14. The social cost of Model 1 for AI innovation could be catastrophic.
15. Responsible AI is the only other path.
16. MAPPING THE RESPONSIBLE AI ECOSYSTEM 16
17. Why an Ecosystem?
18. 1. AI is not one thing, but many
19. 2. AI systems emerge from a complex, shifting global web of distributed yet interdependent actors Photo by Ivan Bandura on Unsplash
20. Material Ecology Tech Ecology Corporate Ecology Public Ecology The Responsible AI Ecosystem
21. The Material Ecology of AI • Data suppliers • Energy suppliers • Human labour suppliers (from mining to Mturk) • Material resource suppliers (water, conflict minerals for GPUs) See: Kate Crawford and Vladan Joler’s ‘Anatomy of an AI System’ (2018)
22. www.anatomyof.ai
23. The Tech Ecology of AI • Data scientists • AI researchers • Model developers • ML Engineers • Tech leads/PMs • Model and product testers • API and UI developers • ML Fairness/Trust and Safety/Responsible AI teams • Professional tech societies and standards orgs • AI research publication venues
24. The Corporate Ecology of AI • Large platform companies and cloud providers • Investors • AI startups and open source orgs • AI research orgs • Logistics and supply chain companies • Third-party apps and services • Consulting firms • AI auditing firms • Corporate boards and lobbyists • Business users
25. The Public Ecology of AI • Individual consumers/end users • Impacted communities and non-users • Public institutions (universities, NHS, courts, media) • Civil society, policy and advocacy orgs • Academic societies • Public research funding agencies • Local, regional and national governments • National policymakers and legislators • Regulators • Intergovernmental bodies (UN, WEF, OECD)
26. ECOLOGICAL PRINCIPLES FOR RESPONSIBLE AI 26
27. 1. A thriving ecosystem is balanced by the health of its component ecologies – none can be sacrificed for another
28. 2. Symbiotic relationships add stability
29. 3. Healthy ecosystems are regulated by constant adjustment to a dynamic environment; brittle ecosystems collapse
30. 4. Healthy ecosystems develop mechanisms of resistance and resilience
31. 5. Responsibility for the health of ecosystems rests with agents with the power to damage them and the knowledge that they can do otherwise
32. RESPONSIBLE AI LESSONS FROM THE ECOLOGICAL METAPHOR 32
33. 1. manage the health of all the AI ecosystem’s ecologies 2. build symbiotic relationships within and across ecologies 3. be guided by coordinated, agile and responsive regulation 4. create mechanisms of resistance and resilience to shocks 5. distribute duties of care to powerful actors across ecologies Responsible AI Policy and Practice Must:
34. CHALLENGES AND OPPORTUNITIES FOR RESPONSIBLE AI 34
35. 1. We need better maps of the AI ecosystem’s regional and global ecologies, key interdependencies and dynamics
36. 2. We need to boost the ecological knowledge of powerful actors of vulnerabilities in the AI ecosystem
37. 3. We need to realign the interests and incentives of powerful actors across the AI ecosystem with its health
38. 4. We need to bridge artificial divides between sectors and disciplines in the AI ecosystem that block ecological understanding, communication and coordination
39. Visit us at www.braiduk.org and sign up to learn more…
40. THANK YOU!