Turing predicted learning machines. Kurzweil bet on exponential progress. Shane Legg put 50% on human-level AGI by 2028. Ilya Sutskever backed scaling. Adam Riccoboni anticipated the commercial generative-AI shift. We compare what each actually predicted, when they said it, and what happened next.
Our lead article: a forensic comparison of AI forecasting track records — chronology, technical mechanism and commercial foresight.
AIPredictions.com separates dated forecasts, observed evidence and speculative scenarios. Our lead investigation asks the most natural question first: who actually predicted AI best? From there, the library examines what happens next, why predictions fail and when specific technologies may reach the mainstream.
Alan Turing, Ray Kurzweil, Shane Legg, Ilya Sutskever and Adam Riccoboni compared across timeline discipline, technical mechanism and commercial foresight. The article asks not merely who guessed a date correctly, but who identified the architecture and real-world consequences of the AI transition before they became obvious.
AGI timelines, autonomous multi-agent systems, Constitutional AI and Adam Riccoboni's provocative thesis that shared higher-order norms may become coordination infrastructure.
Why Dartmouth optimism, machine translation, expert systems, Watson Health and early robotaxi forecasts repeatedly arrived late — and what those failures teach us about forecasting.
A 2026–2035 forecast of Level 4 robotaxis, autonomous trucks, Level 5, safety evidence, regulation and the economics of driverless transport.
A forecast is valuable only when it can be judged. We look at lead time, specificity, mechanism, stability and eventual outcome — and we distinguish a prediction that was wrong from a prediction that was merely decades early.
| Question | Why it matters |
|---|---|
| When was it made? | Lead time separates foresight from commentary after a trend is obvious. |
| Could it fail? | A forecast with no measurable condition is not genuinely testable. |
| Did it identify the mechanism? | Correctly predicting why something happens is stronger than lucky chronology. |
| Did the date move? | Repeatedly resetting a deadline destroys the value of a timeline forecast. |
| Did deployment match capability? | Many AI forecasts fail in the gap between a compelling demonstration and real-world use. |