"Broad Timelines" is a guest article by the philosopher Toby Ord, author of The Precipice, published in Forethought on 2026-03-19. Ord argues that the correct epistemic response to lasting expert disagreement on AI timelines is neither the "short" nor the "long" camp but a broad probability distribution, and he draws planning implications from that stance.
Summary of argument
Ord's central claim is that "the correct epistemic response to the lasting expert disagreement is to have a broad distribution over AI timelines." He supports this by noting that even forecasters identified with one camp hold wide distributions: the short-timelines figure Daniel Kokotajlo's own 80% interval spans 2027 to 2050+ for AI "at least as good as top human experts at virtually all cognitive tasks," the Metaculus community forecast has a similarly broad distribution, and a survey of AI capability researchers (Grace et al.) shows what Ord calls "wild variation," with 80% intervals spanning years to centuries.
Ord states his own distribution for transformative AI — defined as AI powerful enough to take over the world if misaligned, or to double the rate of technological progress — with a median of 2038 and an 80% interval of 3 to 100 years from the time of writing. He summarizes the stance with the line: "We're in a race against AI timelines. We just don't know if it's a sprint or a marathon."
Key claims
Ord draws several implications for how individuals and organizations should plan under timeline uncertainty:
- Uncertainty is not a license to believe whichever timeline one prefers; he compares this to a mayor who cannot ignore a 20% chance of a volcanic eruption.
- The uncertainty will not resolve until events are upon us, so waiting for resolution amounts to choosing the bluntest available responses.
- Different timelines imply materially different worlds. By 2035, Ord suggests, there could be a different president, possibly different leading AI companies, export controls that may have backfired, a Chinese invasion of Taiwan, double-digit unemployment, and an AI risk community that may be seen as discredited.
- Long-term projects can carry very high multipliers but require time to pay off. Founding a field, an organization, a book, or a movement typically takes 5+ years; he notes that AI governance went from an unknown area to hundreds of practitioners over roughly 10 years.
- The community working on AI safety may be over-hedging toward short timelines and underrating work that needs 5+ years to mature.
Provenance
The article appeared as a guest contribution by Toby Ord in Forethought on 2026-03-19 (Source: Raw Sources/Broad Timelines.md). It is a position piece advancing an epistemic and planning argument rather than presenting new empirical results; its forecasts are presented as Ord's own attributed estimates.
Relationships
- related: AI 2027 — Ord directly engages the AI 2027 scenario and Kokotajlo's forecasts, arguing that 2027 should be read as "an important illustrative scenario" rather than a sharp prediction.
- related: Compressed 21st Century — Ord's framework accommodates both Amodei's compressed timeline and decades-long alternatives within a single broad distribution.
- related: AI as Normal Technology — Ord positions broad timelines as a middle ground between the normal-technology and civilizational-challenge framings.
- related: Export Controls (AI) — Ord notes that in long-timeline worlds export controls may backfire by giving China 13+ years to build a domestic chip industry.