8.2 Prediction of Future Developments and Impacts
Original Problem in the Paper
Paper motivation/open problem: “Anticipating the trajectory and potential impact of AI systems may allow policymakers to proactively set governance priorities, determine the urgency of addressing specific issues, and allocate resources accordingly”; foresight supports “adaptive and anticipatory” governance. Open problems: extend empirical measurement of trends such as training compute and algorithmic progress to other trends like industry usage; assess prediction accuracy; estimate impacts before deployment, including economic impacts; develop technical tools to safely/ethically experiment and simulate outcomes without causing harm; understand AI-specific hardware developments and compute governability.
July 2026 Update & Trajectory
Forecasting is significantly more data-rich by July 2026: AI Index 2026, OECD live data, Epoch-style compute/algorithmic tracking, METR time-horizon scaling, IEA Energy & AI, and International AI Safety Report syntheses. But policy-grade prediction remains open: forecasts are sensitive to external validity, capability discontinuities, private data opacity, deployment/adoption bottlenecks, and strategic behavior by labs/states. Hardware-governability forecasts are especially unstable because export-control evasion, cloud access, inference chips, and algorithmic efficiency can substitute for simple chip-count metrics.
Deployed / Operationalized
- Stanford AI Index 2026 tracks technical progress, adoption, investment, policy, and preparedness gaps for policymakers.
- OECD.AI live data tracks AI research, investment, jobs/skills, software development, search/news trends, compute, models/datasets, patents; page flags evolving methodology.
- METR time-horizon benchmark operationalizes one forecastable capability dimension: human-time length of tasks AI agents can complete; updated Time Horizon 1.1 in 2026.
- OpenAI/Google/Anthropic frontier-safety frameworks operationalize risk forecasting through scorecards, early-warning evaluations, capability thresholds, and pre/post-mitigation assessments.
- IEA Energy & AI Observatory/report operationalizes electricity-demand and AI-energy projections for policymakers.
New Tractable Vectors
- Forecast autonomous-agent capability using task time horizons rather than benchmark percent-correct scores.
- Backtest capability forecasts against model release data, benchmark saturation, time-to-complete tasks, and AI Index/OECD trend datasets.
- Estimate energy/grid impacts of AI data centers using IEA regional modelling and data-center datasets.
- Combine incident trends with capability forecasts to prioritize new governance interventions.
- Scenario-test policy thresholds against algorithmic progress and hardware efficiency improvements.
Key Open Questions
- Predicting economic/labor displacement before deployment when adoption, workflow redesign, and organizational learning lag model capability.
- Forecasting tail risks and emergent capabilities from private frontier models with sparse external access.
- Estimating hardware governability under chip smuggling, cloud resale, inference optimization, distributed training, and export-control adaptation.
- Validating simulations of social/political impacts without causing the very harms being studied.
- Forecasting multi-agent and tool-use failures when benchmarks underrepresent long-horizon real-world tasks.
Evidence & Primary Sources
- Stanford 2026 AI Index says AI capabilities, investment, and adoption are accelerating while governance/evaluation/preparedness frameworks lag; it provides globally sourced data for policymakers. (2026 AI Index Report page): https://hai.stanford.edu/ai-index
- OECD.AI live data provides timely indicators across AI research, demographics, investment, jobs/skills, software development, search/news, compute, knowledge flows, models/datasets, and patents, while noting methods remain evolving. (Copyright 2026; page notes data section being updated): https://oecd.ai/en/data
- METR reports frontier agent task-completion time horizon has doubled about every 7 months and Time Horizon 1.1 is its current 2026 measurement. (19 March 2025; page includes 2026 Time Horizon 1.1 update): https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/
- International AI Safety Report 2026 summarizes what general-purpose AI can do, emerging risks, and mitigation approaches for policymakers. (3 February 2026): https://internationalaisafetyreport.org/
- IEA Energy and AI report aims to fill data gaps with global/regional modelling and datasets for AI electricity consumption, energy supply, emissions, security, innovation, and affordability. (Published 10 April 2025): https://www.iea.org/reports/energy-and-ai
- OpenAI Preparedness Framework explicitly tracks, evaluates, forecasts, and protects against catastrophic risks via scorecards and tracked risk categories. (18 December 2023): https://cdn.openai.com/openai-preparedness-framework-beta.pdf