Rodney Brooks argues AI hype overstates what generative AI and humanoid robots can do

Rodney Brooks argues AI hype overstates what generative AI and humanoid robots can do

Rodney Brooks published this essay on May 5, 2025. He argues that generative AI and humanoid robots attract similar hype because human-like language and appearance invite people to assume abilities beyond those demonstrated. Fluent language, he says, leads people to expect large language models to reason and take on white-collar work; a human-shaped robot suggests it could do any human task and provide cheap blue-collar labor. Brooks says both fields rely on “big learning,” using large amounts of data while the learned weights remain poorly understood. Their data costs put academics at a disadvantage, though the data differs: generative AI uses existing digital material, while humanoid-robot training data comes from people demonstrating tasks. He challenges predictions of extremely cheap robot labor and rapid deployment at scale, arguing that mechanical systems cannot follow computation’s cost reductions and physical systems take decades to deploy. Brooks concludes that both technologies may improve human lives, but not at the physical scale or on the short timescale proponents imagine. He predicts regret over the capital spent pursuing them. His footnote questions whether current demonstrations capture the touch information needed for dexterous manipulation.

Publication: e73ba001-352f-9742-0866-764672c7772c

attribution — pass: The source is a single essay by Rodney Brooks on his own blog, and the reading attributes its arguments, laws and predictions to him throughout. It does not present them as established fact.

This short blog post is to point out some strong similarities between these two “application areas”

summary — pass: The summary accurately conveys Brooks's thesis: human-like language and appearance drive hype. It also covers big learning, academic exclusion, exponentialism, deployment timescales, his conclusion and the touch-data footnote. LLM is defined beside the reading, and big learning is explained in plain words. The reading does not name his performance-vs-competence and magic sins, or his claim that LLMs only predict next words, but it keeps his core argument and force. The question follows from the evidence.

It is the apparent human-ness of these two technologies that both lure people in

date — pass: originalPublished is 2025-05-05T21:13:02Z, so May 5, 2025 is the matching UTC day. The kind is publisher-page, so no archive disclosure is needed. The source's own timeline (thirteen years after a 2012 ride) is consistent with 2025. Brooks's regret claim is reported as a prediction, and the reading adds no hindsight.

Thirteen years after my first ride I can still only take a ride in a Waymo

support — pass: Each claim matches the cited passage. The limit on mechanical cost reduction reflects Brooks's argument that circuits could shrink about 4-million-fold while mechanical systems might gain perhaps 4x. The limitations fairly describe the unquantified regret prediction and the footnote's scope. No benefits or costs are invented.

But you can’t do that with mechanical systems that are doing real work with real payloads.

media — pass: No media was selected and no verified media candidates were supplied, so nothing is misidentified.