AI breakthroughs in robotics won’t change your life any time soon
MIT Technology Review Jamie Condliffe ● Covered by 3 sources
Humanoid robots are getting more AI hype, but they still fumble in the real world. The gap between flashy demos and useful machines is still huge.
Based on reporting by MIT Technology Review, Jamie Condliffe — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Tesla’s Optimus keeps turning up in the kind of clips that make futurists giddy: dancing, serving popcorn, taking out trash, vacuuming, even pressing a microwave button. It also falls over, struggles with a shirt, and mostly shows up handing out food and drinks at Tesla events. Elon Musk says it could become Tesla’s biggest product ever and, eventually, handle nearly all human labor for about $20,000 a robot. He has floated a public sale by the end of 2027. Marc Andreessen and Jensen Huang have made their own grand predictions too, while Morgan Stanley says robots that resemble and act like humans could reach nearly 1 billion by 2050 and form a market worth more than $5 trillion.
The problem is that the pitch assumes robots can ride the same AI wave that powered ChatGPT and Claude. Many roboticists don’t buy it. Yann LeCun says the companies making humanoid robots do not know how to make them useful enough, and Jonathan Hurst of Agility Robotics says it is easy to make a robot look human and much harder to make it move and behave like one. That distinction matters. A humanoid shell is not the same thing as a general-purpose machine.
The more interesting work is happening in smaller, stranger setups. Google DeepMind’s ALOHA 2 is basically a pair of arms, grippers, and cameras, but it can be driven by Gemini Robotics, a vision-language-action model trained on human demonstrations. Give it a lunchbox task and it can manage bread, grapes, a Ziploc bag, a Tupperware container, and a zipper, though not especially elegantly. That’s progress. It is also a reminder that robotics still lives in the land of partial wins, not miracles.
What has changed is how robot software is built. Old systems were full of hand-coded rules; now models are being trained to understand scenes, plan actions, and move. The hope is that more data, better video, and eventually world models will get robots from clumsy demos to real generality. But the source of the optimism is also the source of the bottleneck: there is no giant pile of robot experience sitting around like the internet text that fed language models. So the field keeps inching forward, while the hype sprints ahead.
My take — AI-written commentary, not fact-checked reporting
The humanoid robot crowd is selling a movie trailer and calling it a product roadmap. The real work looks less glamorous: arms on a bench, teleoperation, slow gains, and a lot of failure modes with very expensive shoes. Closed, fully automated fantasies are doing the usual Silicon Valley thing — outrunning the physics by a country mile.
Read more about this at: MIT Technology Review
