Some AI-related links, which I hope will be of use:
✨Final text of AI Pact pledges, promulgated by the European Commission to get tech companies to comply voluntarily with (at least some of) the AI Act before its formal applicable date. So, the text is not dissimilar from that of the EU AI Act, and indeed the G7 Principles from the Hiroshima Process
✨Human oversight of ADM / automated decision-making: videos from the EDPS's workshop, 3 Sept 24
✨AI Act briefing for European Parliament, 2 Sept 24
✨BSA (The Software Alliance) Best Practices for Information Sharing Along the General Purpose AI Value Chain. There's some overlap with the EU AI Act's GPAI requirements, 3 Sept 24
✨Computer & Communications Industry Association's recommendations on GPAI code of practice, Aug 24
✨Note that the Council of Europe's AI Treaty, signed by the UK, EU and others, will come into force only on the first day of the month following the 3-month period after 5 signatories, including at least 3 Council of Europe member states, have ratified it. (On treaties/conventions, see the differences between signing versus ratification versus accession).
To support the Treaty's implementation, the COE's HUDERIA is a "legally non-binding methodology" for Risk & Impact Assessment of AI Systems for Human Rights, Democracy and Rule of Law (good summary). The UK's Alan Turing Institute is assisting on HUDERIA. The COE Committee on AI is considering HUDERIA soon. The European Commission & Council of the European Union are also involved. What's the betting as to how much HUDERIA will influence what is going to be required in Fundamental Rights Assessments for certain high-risk AI systems under the AI Act?
(Compare the US Department of State's risk management profile for AI and human rights, July 2024)
✨Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025. Certainly, genai risk disclosures in US SEC filings are increasing.
✨A data scientist has written a great outline of a practical approach in the face of the pressure to to AI-ify everything ASAP. Consider, is using AI always the best solution? Especially given that AI systems will increasingly be subject to more onerous obligations than non-AI systems (e.g. under the AI Act), is it always best to use AI when non-AI approaches/methods could work equally well or perhaps better?
✨What's up with the planned AI Liability Directive? It won't automatically be brought into force, but a June Parliamentary Committee report stated that "the legislative work under the leadership of the Committee on Legal Affairs... will continue under the new Parliament. In the meantime, the
Committee... requested an additional impact assessment from the European
Parliament Research Service (EPRS). It is to primarily deal with the compatibility of the three legal acts mentioned [AIAct, product liability etc] and the risk-based concept and is expected to be completed by the end of 2024". Note the date for the additional impact assessment - end of 2024. So it's very unlikely this Directive will be passed this year, perhaps ever.
(I had tried to post the above on LinkedIn last Thursday, please see this AI developments post, but LinkedIn saw fit to demote it and not show it in people's feeds, so I think it's best to blog here instead, and at least there's space for me to flesh things out a bit more here.)