Food handling is an area of work that still relies heavily on human activity. Fruit, vegetables, meat and other food products should be handled quickly but gently. This is arduous to automate because no two pieces of fruit, vegetables or chicken nuggets look exactly the same.
Eka’s demos suggest the company could be onto something large. I found myself mentally comparing their robots to GPT-1, OpenAI’s first major language model, developed four years before ChatGPT. GPT-1 was often incoherent, but showed glimpses of general linguistic intelligence.
The robots I’ve seen seem to have a similar kind of emergent physical intelligence. When I watched a video of someone reaching for a set of keys in snail-paced motion, I noticed that he did something that seemed remarkably human: he touched the ends of the grippers to the table and moved them across the surface, then touched the keys and secured them between the digits. Eki’s algorithms seem to instinctively know how to recover from failure. It’s tough for other robots to learn these kinds of things unless the humans training them intentionally make a series of mistakes.
Unlike every other robot I can think of, you can almost imagine what the world of a robot is like. Its sensors appear to sense the arm’s weight and inertia as it moves toward the keys and slows down. Once he has the keys in his hand, he seems to feel their weight hanging from his talons.
I don’t know if Eka’s approach is really a path to a breakthrough in robotics like ChatGPT. Some very brilliant experts believe that combining human demonstration with simulation will produce better results than simulation alone. Maybe it will ultimately be necessary to combine these two approaches? But it seems clear that robots will eventually need to have the kind of tactile, physical intelligence Eka is working on if they are to achieve human dexterity.
Agrawal tells me that the same general approach should work for more gentle manipulation. For example, the strange dexterity required to build an iPhone can be achieved by building various actuators and sensors and practicing the task in a simulation.
After a few hours in Eka, I decide to drop by the restaurant downstairs. I watch at the counter as the staff prepares food and brews coffee. The descendant of the machine at the top may be able to do these things just as well, if not better. But considering how much I enjoy talking to the people who work there, I think I’d pay extra to keep people around. Unless my hands get automated too.
What do you say?
Let us know what you think about this article in the comments below. Alternatively, you can send a letter to the editor at [email protected].