AI predictions that missed their deadlines · a curated, evidence-screened timeline

Across 75 years, 23 named forecasters put deadlines on AI that passed unmet — all but one aimed too soon

One representative falsified prediction per person: 23 dated, falsifiable AI and automation claims by notable researchers, founders and commentators, stated between 1950 and 2025, every deadline elapsed by August 28, 2026, each with a cited assessment. This is a curated sample, not a census of every mistaken forecast — people who missed many times appear once. A miss grades the stated deadline and scope, not motives, expertise, or a career; and a later arrival would not make the original deadline correct.

statement made deadline passed unmet later assessment, if different year too pessimistic — the one counterexample dashed span — low confidence: vague metric or debatable assessment

Colour by topic family · exact topics and people counts: AGI 4 Chess 3 Superintelligence 3 AGI / language 2 Autonomous vehicles 2 Jobs / medicine 2 AGI / perception 1 AGI / robotics 1 Jobs 1 Jobs / professional work 1 Language 1 Robotics 1 Software jobs 1

What the timeline cannot say alone

The longest horizon in the set is Alan Turing’s — fifty years, from 1950 to 2000 — and it is also a caveated one: his forecast was conditioned on memory scale, which arrived long before the predicted performance did.multigrid.ai

The shortest horizon belongs to Dario Amodei — three to six months in 2025 for AI writing 90% of code; the later assessment reported major-company public estimates around 25–30%.sloppish.com

Hubert Dreyfus is the single too-pessimistic entry: he argued computers could not achieve competent chess play, then lost to MIT’s MacHack program in 1967 — though parts of his broader critique of symbolic AI endured.multigrid.ai

Some misses were confirmed only well after the deadline: Geoffrey Hinton’s 2016 radiology forecast had a 2021 deadline but the cited assessment dates to 2025, with radiologist numbers and shortages persisting and AI used as an aid.boyswhocriedai.lovable.app

All 23 people

Person	Claim and outcome	Made → deadline	Topic	Confidence	Source

Confidence and caveats

High confidence means a concrete deadline and a clear outcome; Medium means the interpretation, wording or scope requires care; Low means a vague metric or a debatable assessment, and should not be read as equally conclusive.
Alan Turing’s imitation-game forecast was conditioned on memory capacity, not a bare calendar date. Marvin Minsky later disputed the accuracy and context of the magazine quote attributed to him.
Shane Legg’s statement was explicitly probabilistic — a modal estimate, not a promise. Andrew Ng’s “near future” and Sam Altman’s professional-task framing are less sharply falsifiable than dated deployment claims. Mark Zuckerberg framed his forecast as a goal.
Broad terms such as AGI, humanlike intelligence, or “routinely pass the Turing Test” lack universally agreed operational definitions; those rows carry Medium or Low confidence for that reason.
Strict inclusion: a named notable person, a falsifiable AI or automation claim, and a deadline already elapsed by August 28, 2026. One representative prediction per person keeps the picture compact; repeat offenders are intentionally not repeated. Most forecasts here were too optimistic about timing; Hubert Dreyfus is included as an instructive too-pessimistic counterexample.

Curated timeline of 23 falsified AI predictions, exactly one row per named person, statements 1950–2025, deadlines through 2026, assessments through 2026, all elapsed by the 2026-08-28 cutoff. Each span measures statement year to stated deadline year; a dotted ring marks a later assessment year. Topic counts are precomputed, not recalculated. Full claim-and-outcome wording is preserved in the table; timeline labels are shortened for space. Not an exhaustive census of mistaken forecasts.