Which Impact Assessment for AI, Beyond Data Protection
summary
Panel argues that impact assessments must be concrete: regulators model compliance costs with a standard cost model and estimate 5–10% of systems as high risk, deployers must run Article 27 fundamental‑rights assessments before rollout, and authorities run pre‑audits (smart‑camera checks) and set up AI sandboxes and a French public LLM for testing. Businesses map ~70% of legal requirements to data‑science metrics but warn rapid model change, API value chains and liability gaps force recurring reassessments.
AI-written summary · about
chapters (11)
- 0:00AI impact assessments beyond data protection
- 4:21What is impact assessment
- 10:03EU AI Act impact assessments
- 18:34Regulators' AI impact approaches
- 27:11Civil society AI regulation expectations
- 32:31Business AI obligations concerns
- 47:33AI value chain impact integration
- 52:27Building user trust in AI
- 55:42AI impact assessment challenges
- 1:05:32Q&A on AI impact standards
- 1:17:32Closing remarks
related talks
01h 16mBEYOND ETHICS WASHING:IMPACT ASSESSMENTS, AUDITS, AND OVERSIGHT FOR AI
01h 17mASSESSING THE IMPACT ON FUNDAMENTAL RIGHTS IN AI APPLICATIONS
31mPanel: Navigating the AI Policy Landscape
36mResponsible AI: In Theory, Practice, and Policies
01h 15mAssessing Fundamental Rights Risks of AI Opportunities and Challenges Ahead
53mAI: Law, Policy and Common Sense Suggestions on How to Stay Out of Trouble
17mAI Law & Ethics: developing responsible AI systems
01h 06mEMPOWERING THE AI ACT: LIMITS AND OPPORTUNITIES