Futures and foresight

 
drew-beamer-xU5Mqq0Chck-unsplash.jpg
 

AI is progressing rapidly, both in terms of fundamental research and application to scientific and societal challenges. Performance benchmarks are rapidly getting saturated, and frontier models are showing marked improvement in general performance across an increasingly wide range of tasks. Newer frontier models demonstrate improved performance on tasks requiring agency, autonomy, and longer time horizons. Recent evaluations have even demonstrated that large language models can be fruitfully applied to relevant AI research tasks – further progress in this domain will have profound implications for capability trajectories. These advances have profound economic, geopolitical, and risk implications.

Future possibilities may become pressing challenges quite quickly. Norms and governance established now will have a crucial happing role. There is an ongoing need for work focusing on better characterising AI trajectories and their societal implications, incorporating perspectives from across science and society.

Research within this theme includes:

  • Technology forecasting: Exploring the implications of different avenues of AI progress

  • Trends analysis: Identifying and exploring broader technological and social trends likely to influence AI impacts and risks.

  • Methodological innovation: Developing foresight and scenario analysis methodologies to improve our ability to do research on the above

  • Conceptual analysis: Clarifying conceptual frameworks to improve our ability to think clearly and communicate about impacts and risks.

Projects include:

  • AI strategy role play: A scenario tool to explore the societal and geopolitical impacts of AI, focusing on a range of technological developments and the interplay of different actors (technology, governance, civil society)

  • Mid-term Impacts of AI: exploring the possible impacts on society of advances in AI falling short of human-level intelligence

Relevant papers include:

Gruetzemacher, R., Avin, S., Fox, J., & Saeri, A. K. (2024). Strategic Insights from Simulation Gaming of AI Race Dynamics. arXiv preprint arXiv:2410.03092.

Schellaert, W., Martınez-Plumed, F., Vold, K., Burden, J., Casares, P. A., Loe, B. S., ...ÓhÉigeartaigh, S. S. & Hernández-Orallo, J. (2023). Your Prompt is My Command: On Assessing the Human-Centred Generality of Multimodal Models. Journal of Artificial Intelligence Research, 77, 377-394.

Zhou, L., Moreno-Casares, P.A., Martínez-Plumed, F., Burden, J., Burnell, R., Cheke, L., Ferri, C., Marcoci, A., Mehrbakhsh, B., Moros-Daval, Y. and Ó hÉigeartaigh, S.S., 2023. Predictable Artificial Intelligence. arXiv preprint arXiv:2310.06167.

Clarke, S., & Whittlestone, J. (2022). A Survey of the Potential Long-term Impacts of AI. AIES 2022

Gruetzemacher, R., & Whittlestone, J. (2022) The transformative potential of artificial intelligence. Futures, 135.

Hernández-Orallo, J., Loe, B. S., Cheke, L., Martínez-Plumed, F., & Ó hÉigeartaigh, S. (2021). General intelligence disentangled via a generality metric for natural and artificial intelligence. Scientific reports, 11(1), 22822.

Martínez-Plumed, F., Barredo, P., Heigeartaigh, S. O., & Hernandez-Orallo, J. (2021). Research community dynamics behind popular AI benchmarks. Nature Machine Intelligence, 3(7), 581-589.

Whittlestone, J., Arulkumaran, K., & Crosby, M. (2021). The Societal Implications of Deep Reinforcement Learning. Journal of Artificial Intelligence Research, 70, 1003-1030. https://jair.org/index.php/jair/article/view/12360/26667 

Cremer, C. Z., & Whittlestone, J. (2021). Artificial Canaries: Early Warning Signs for Anticipatory and Democratic Governance of AI. International Journal of Interactive Multimedia & Artificial Intelligence, 6(5).