⚡ Quick Summary
Published by the OECD as No. 27 in its Artificial Intelligence Papers series in November 2024, this report reviews research and expert perspectives on potential future AI benefits, risks and policy actions. It draws on contributions from the OECD Expert Group on AI Futures, a multidisciplinary group of 70 experts co-chaired by Stuart Russell, Francesca Rossi and Michael Schönstein, and on a members' survey conducted August-September 2023, discussions and scenario exploration.
The report sets out characteristics of desirable AI futures, then presents ten priority benefits, ten priority risks and ten priority policy actions, selected from 21 potential benefits, 38 potential risks and 66 potential policy approaches. Benefits range from accelerated scientific progress and productivity gains to empowered citizens and improved institutional transparency; priority risks include malicious cyber activity, disinformation and fraud, race dynamics, misalignment, power concentration and surveillance; policy actions include clearer liability rules, "red line" restrictions, disclosure requirements and lifecycle risk management. It also reviews existing efforts, including the EU AI Act, the Council of Europe Framework Convention on AI, the US AI Executive Order and the G7 Hiroshima Process.
🧩 What’s Covered
The report moves from an executive summary through four chapters to two annexes, and closes with references and notes.
- Framing and foresight (Chapter 1): characteristics of desirable AI futures, including widely distributed benefits, respect for human rights, privacy and intellectual property rights, resilience of physical, digital and societal systems, mechanisms to maximise AI security, prevention of excessive power concentration, strong risk management and international co-operation. Box 1.1 lists strategic foresight methods: expert surveys and consultations, scenario planning and road mapping, trend and data analysis, and horizon scanning.
- Priority benefits (Chapter 2): ten benefits, from accelerated scientific progress, economic growth, reduced inequality and progress on the SDGs, through better decision-making, information production and healthcare and education services, to improved job quality, empowered citizens and improved institutional transparency. Each benefit covers current evidence and examples of policy efforts.
- Priority risks (Chapter 3): ten risks, including increasingly sophisticated malicious cyber activity, manipulation, disinformation and fraud, race dynamics, inadequate alignment methods, power concentration, incidents and disasters in critical systems, invasive surveillance, governance unable to keep pace, weak explainability and interpretability, and exacerbated inequality or poverty.
- Priority policy actions (Chapter 4): ten actions, from clearer rules including on liability and "red line" restrictions, through disclosure, lifecycle risk management and mitigation of race dynamics and power concentration, to investment in AI safety research, education and reskilling, and empowerment of stakeholders, each with recent and emerging policy efforts.
- Methodology (Annex A): a literature review of approximately 250 sources produced an initial set of 17 benefits, 36 risks and 68 policy actions; a prioritisation survey rated importance and actionability on a 0-10 scale, with 53 of 61 Expert Group members responding (87%).
- Rankings (Annex B): Figures B.1 to B.3 and Table B.1, which numbers all 66 policy actions, from liability rules for AI-caused harm and AI safety R&D to an advanced AI R&D moratorium and a ban on advanced AI.
- References and notes: a bibliography of several hundred entries and 43 endnotes with further links.
💡 Why it matters?
The report gives governance, risk and compliance teams a ranked view of which future AI impacts an expert group considered most important and most actionable, and pairs each with instruments already in force or emerging. Those connections are drawn in the document itself: the EU AI Act's prohibitions, high-risk system requirements and general-purpose AI obligations; the Council of Europe Framework Convention on AI; the US AI Executive Order; the G7 Hiroshima Process Code of Conduct; the NIST AI Risk Management Framework; and ISO/IEC 42001:2023 as a framework for managing risk and opportunities. That makes it usable for aligning foresight work with concrete regulatory and standards obligations rather than treating them separately.
❓ What’s Missing
The foreword states that many future-oriented aspects are "necessarily speculative" given the lack of rigorous study on some topics, and Annex A notes the survey was "not designed to be a scientific instrument" but a gauge of subjective expert opinion. The report offers no timelines, cost estimates or step-by-step implementation guidance, and it leaves the contested question of humans losing control of AGI unresolved, recording diverging views rather than consensus. Coverage of individual benefits is uneven, with job quality described as under-recognised relative to job quantity, and public investment in AI in education assessed as low relative to its potential.
👥 Best For
Policy advisers and foresight teams building national or institutional AI agendas; risk and compliance leads who need a ranked list of future risks to map against the EU AI Act, the CoE Framework Convention and the NIST AI RMF; standards, audit and safety teams scoping research and assurance priorities; and researchers tracking expert opinion on AI futures.
📄 Source Details
Assessing Potential Future Artificial Intelligence Risks, Benefits and Policy Imperatives, OECD Artificial Intelligence Papers No. 27, published by the OECD in November 2024, 69 pages, in English. The report was drafted by Jamie Berryhill, Hamish Hobbs and Dexter Docherty, discussed and reviewed by the OECD Expert Group on AI Futures, and approved and declassified by the Digital Policy Committee on 30 October 2024. Its reference code is DSTI/CDEP/AIGO(2023)13/FINAL. No URL or DOI for the report itself is printed. The extracted text covered all 69 pages, but Figures B.1 to B.3 appear only as captions.
