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

From Minsky to Musk, confident short-horizon AI forecasts keep outrunning reality

Six dated, falsifiable predictions from four prominent AI researchers and technology leaders, made between 1970 and 2025, each with a deadline that passed and a cited post-deadline outcome. Every prediction below missed its stated deadline and scope — which says nothing about whether the technology arrives later.

claim horizon, statement to deadline stated deadline window wait until outcome was assessed ambiguous metric, lower confidence ✕ deadline passed unmet

What the timeline cannot say alone

The shortest horizon in the set was three to six months — Dario Amodei’s claim that AI would write 90 percent of code; the most generous public figures put AI-generated code at major technology companies at 25–30 percent.sloppish.com

Marvin Minsky’s 1970 three-to-eight-year forecast of average human general intelligence went unmet, and the pattern of unmet expectations eventually fed AI’s second winter by the late 1980s.ai-consciousness.org

Nine years after Geoff Hinton’s 2016 call to stop training radiologists, the 2025 assessment found no radiologist had been replaced — the deadline passed by four years before the check.boredreading.com

Elon Musk accounts for three of the six events; the most precise of them named a day — unsupervised robotaxis in Austin on June 22, 2025 — and the service launched that day with safety drivers instead.forbes.com

All six predictions

Person	Claim, as stated	Stated	Deadline	Outcome at assessment	Source
Marvin MinskyAI researcher / participant	“In from three to eight years we will have a machine with the general intelligence of an average human being.”	~1970	~1973–1978	AI did not achieve general human-level intelligence within that timeframe, leading eventually to AI’s second winter by the late 1980s. Assessed ~1978.	ai-consciousness.org
Geoff Hinton	“We should stop training radiologists now. It’s just completely obvious that within five years, deep learning is going to do better than radiologists.”	~2016	~2021	“Nine years post Hinton, we still haven’t replace a single radiologist, as far as I know.” Assessed ~2025.	boredreading.com
Elon Musk	“I feel very confident predicting that there will be autonomous robotaxis from Tesla next year … next year for sure, we’ll have over a million robotaxis on the road.”	~2019	~2020	“Musk’s robotaxi promise hasn’t fared much better.” Assessed ~2025.	boredreading.com
Elon MuskCEO of Tesla and SpaceX	“AI will probably be smarter than any single human next year.”	Mar 13, 2024	~2025	2025 came and went. No serious observer, including Musk, claims an AI became “smarter than any single human” by the end of 2025 in any general sense. Ambiguous metric — lower confidence.	sloppish.com
Elon Musk	Declared Tesla would have a robotaxi service running unsupervised, with nobody in the car, in Austin on June 22, 2025.	~2025	Jun 22, 2025	They were unable to run the service unsupervised with nobody in the car on June 22, 2025, and instead deployed that day with safety drivers.	forbes.com
Dario AmodeiCEO of Anthropic	“I think we’ll be there in three to six months, where AI is writing 90 percent of the code.”	Mar 10, 2025	~Sep 2025	Neither happened. The most generous public figures put AI-generated code at major technology companies in the range of 25–30%. Assessed ~2026. Ambiguous metric — lower confidence.	sloppish.com

Method and caveats

This is a curated, evidence-screened set, not an exhaustive census. Inclusion required a notable AI researcher or technology leader, a concrete falsifiable claim, a passed deadline, and a cited post-deadline outcome. Opinions without deadlines were not treated as falsified predictions.
“Important” and “controversial” are judgment calls; the web scan was broad but cannot prove exhaustiveness, and source quality varies, especially for retrospective assessments.
Falsification applies to each claim’s stated deadline and scope — not to the broader possibility that the technology arrives later, and not to the individuals’ wider bodies of work. Nothing here establishes motive or deception.
Dates known only to the year are shown as approximate (~). Recent broad claims — “smarter than any single human,” percentages of AI-written code — rest on ambiguous metrics and are marked lower-confidence than concrete deployment counts such as robotaxi launches.
Elon Musk made three separate claims; they are kept as three distinct events, not deduplicated.

Curated timeline of 6 falsified AI predictions by 4 named figures, one row per distinct statement, spanning statements from 1970 to March 2025 and assessments from ~1978 to ~2026. Each bar measures time from statement to stated deadline; dashed tails show the further wait until the cited assessment. Total of 6 events taken as supplied, not recomputed. Full claim wording preserved in the table; timeline labels are shortened for space.