Getting AI to design the robots for us
“Gradient descent” is a method used to train neural nets (skeins of artificial neurons) to evolve in the direction you want. A new paper describes how an AI can use gradient descent to design robots that walk. A nice account in SciAm is here, and the paper itself is here.
If you’re imagining a sort of DALL-E for robots (you say, “please design and 3-D print a robot that will trim the hedge while singing Puccini,” and presto!) well, join the club. That’s far off. But even the first hints of robots that can be built without humans first having to figure out all their details — that’s a big deal.
It’s a small world, after all
Disney’s new robot looks a lot like WALL-E. Among its maybe more consequential traits: It is made of 3-D printed parts and off-the-shelf components. And its programming isn’t bespoke — it can be used in other robot bodies with other forms. One reason you’re seeing more robots these days is because robot-makers are moving away from one-of-a-kind designs and software, built for one machine. The more a robot can be made from hard and soft ware that’s widely available, the more feasible it is as a product that can be made and sold in large quantities.
Large Language Models As Public Utilities?
It takes a lot of resources to develop a Large Language Model like ChatGPT or Bard. Which means there is not going to be a lively market of competitors offering this kind of AI service. So maybe, as these AI monopolies (or at best oligopolies) form, they will need to be regulated as were railroads and telephone companies in the 20th century — as public utilities. So argue these researchers.
Centaurs or Cyborgs?
I’m going to Zoom in to this talk next week by Ethan Mollick, a professor at the Wharton School. It’s about a recent experiment he did with 758 Boston Consulting Group consultants, in which he examined the effects of AI tools on their work. Among his findings: Some users acted like centaurs, choosing which part of their hybrid human-machine worker would do a task. But others “acted more like ‘cyborgs,’ completely integrating their task flow with the AI and continually interacting with the technology.”
An AI Mental Health Tracker?
This paper describes an AI that combines sentiment analysis of your posts on X (formerly Twitter) with analysis of your other online activities to produce a real-time report on your state of mind.
Coming up later this month: A post on the mixed bag experience of using AI in day-to-day life. Later, and some thoughts on all these humanoid robots (supposedly) coming our way. See you next week.