An analysis from MIT Technology Review examines the gap between soaring predictions for humanoid robots and the actual state of the field. The article notes that figures like Elon Musk and Jensen Huang have promised humanoid robots will soon match or exceed human abilities, with Musk suggesting they could be sold to the public by 2027 and Morgan Stanley projecting a $5 trillion market by 2050. These claims are fueled by advances in AI models like ChatGPT, which some believe can be transferred to physical machines.
But many robotics researchers are skeptical. Yann LeCun, a prominent AI researcher, said at the Davos conference that no humanoid robot company has any idea how to make its robots smart enough to be useful. Jonathan Hurst of Agility Robotics adds that making a robot look human is easy, while making one move and behave like a person is dramatically harder. The article stresses that current AI, built on language and images, may not be sufficient to handle the variability of the physical world.
Despite the skepticism, the piece highlights real but modest progress. Google DeepMind's ALOHA 2, a simple two-arm robot, can pack a lunchbox when controlled by the Gemini Robotics system. That task, however, is far from the 'superhuman dexterity' promised by Musk. The article concludes that while a robotics revolution may eventually come, it will not transform daily life any time soon.