Rodney Brooks has scored his own technology predictions in public for eight years. His forecasts were pessimistic. They were still too optimistic. The failure mode is not technology — it is the assumption that society will rearrange its infrastructure to match capability at the rate capability advances.
Rodney Brooks has been scoring his own predictions in public for eight years. The MIT robotics professor emeritus, iRobot co-founder, and co-founder and CTO of Robust.AI made a set of dated, falsifiable forecasts in 2018 about self-driving cars, robotics, electric vehicles, and artificial intelligence. He committed to reviewing them annually until 2050. On January 1, 2026, he published the eighth review.
The results are not what you would expect from a technology skeptic. His predictions were pessimistic. They were still too optimistic.
The Record
By the fifth year of tracking, fourteen of his original predictions had proved accurate — either events occurring within the projected timeframe or correctly failing to occur before the deadline. Brooks summarized that his forecasts held up well, though overall he had been a little too optimistic. By the eighth year the pattern had only deepened.
Self-driving cars were supposed to be ubiquitous by now — not according to Brooks, who was always skeptical, but according to every major automaker and several well-funded startups that projected availability by 2022. Brooks predicted they would arrive much later. Even that proved generous. Waymo operates in roughly a dozen US cities with remote human operators on standby. Cruise is defunct. The autonomy that exists requires remote human operators. HOV lanes and vehicle-to-infrastructure communication are non-starters without federal investment that never materialized.
Electric vehicles followed the same trajectory. Brooks predicted EV sales would reach thirty percent of the US total no earlier than 2027. He now calls himself an extreme optimist — actual EV sales may not reach even ten percent in 2026.
Robotics revealed the widest gap. Brooks wrote that what is easy for humans is still very hard for robots. Dexterous robot hands remain, in his assessment, pathetic compared to human hands beyond 2036, despite laboratory demonstrations and large language model integration. Lab demos and deployed capability are separated by a gulf that eight years of exponential rhetoric have not narrowed.
The Gap
The failure mode across all three domains is the same, and it is not technology. Self-driving car technology works in controlled settings. EV batteries are cost-competitive on a per-mile basis. Robot arms can pick up objects in laboratory environments. The technology demonstrated capability years ago.
What did not happen was infrastructure adaptation. Roads were not redesigned for autonomous vehicles. Charging networks were not built at the density required for mass adoption. Warehouses were not reconstructed to accommodate the specific limitations of robotic manipulation. Federal standards were not written. Consumer trust was not established. Insurance liability was not resolved.
Capability demonstration operates on quarter timescales. A lab can show a robot hand folding laundry in a press release cycle. Infrastructure adaptation operates on decade timescales. Redesigning a warehouse floor plan, retraining a workforce, passing a federal standard, and rebuilding consumer expectations after a fatal accident each take years, and they must happen in sequence rather than in parallel.
Every prediction that implicitly assumed infrastructure would keep pace with capability was wrong. Including the pessimistic ones.
The New Forecasts
Brooks added five predictions for the 2026-to-2035 window. Each carries the same structural assumption: society adapts more slowly than technology advances.
Quantum computers will function as specialized simulators of physical systems — twenty-first-century analog computers — not general-purpose machines. Self-driving will remain a two-player race between Waymo and Amazon's Zoox, with human intervention rate as the only metric that determines profitability. Humanoid robots will remain too unsafe for proximity to humans without fundamentally new mechanical systems. Neural computation research will produce small academic advances beyond 1960s linear threshold models without clear winners by 2036. And large language models will face an arms race not in capability but in explainability — boxing-in mechanisms for safe deployment that have not yet been invented.
Each prediction bets against the same thing: the assumption that demonstrated capability implies imminent deployment.
What This Changes
The practical implication cuts across domains. Companies whose business models require society-wide infrastructure adoption within five years are structurally disadvantaged — not because the technology will fail, but because the surrounding systems cannot absorb it fast enough. Incremental integration wins: Waymo's city-by-city expansion, warehouse robotics in constrained environments, EV adoption in fleet operations where charging infrastructure can be centralized.
The winners are not the companies with the best demonstrations. They are the companies that correctly estimated how long the world takes to rearrange itself around a new capability.
The meta-lesson is about the only predictions worth making. Brooks's scorecard works because it is public, dated, and scored against reality. Anyone can check the record. The methodology is transparent. The failures are acknowledged in the same document as the successes. Fourteen of thirty-two original predictions accurate after five years is a track record. Everyone else's predictions about technology adoption are marketing dressed as forecasting.
Brooks committed to reviewing his predictions for another twenty-four years. If the pattern holds, the review in 2050 will confirm what the review in 2026 already shows: the bottleneck was never capability. It was everything else.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)