Physical Intelligence (PI) is the latest AI paradigm added to the efforts to develop machines smarter than humans. A few days ago DARPA (Defense Projects Research Agency), the people who brought you the Internet and other spooky things you shouldn’t know about, posted a notice announcing informational workshops as a preamble to issuing RFPs (Request for Proposal) for PI research and developments. A more accessible description of PI is given in this Wired post.
You can look at PI as the latest addition to modeling human brain function using the formalisms of mathematics and physics – i.e. these are physical processes described by sets of equations which describe how the brain senses, processes, decides, issues orders, and so on. Maybe PI will even explain and implement how it (some matter and energy) becomes conscious with a sense of self and ability to synthesize aspirations of what it wants to do or become. PI will join such approaches as hierarchical temporal memory to build complex substrates that may become intelligent.
Again, dear Reader, don’t think about any of these steady and accelerating steps toward Singularity as efforts where humans presume to ‘program’ a computer to be smart like a human. ‘Smart like a human’ is most likely beyond human programming capacity – for you techies, it's something to do with self-reference and Gödel’s Theorem (or in this case Gödel’s Constraint). Instead, the more promising approach is to fashion physical brains as very complex and adaptive systems that then can learn like we did as babies and children – but, of course, learn much more rapidly.
So DARPA is going to sponsor this work under its national security charter, which may or not be a bit scary. People like me, who spent a good part of their lives designing and building spooky systems for the government, are always on the lookout for the dark side of success in such areas. The more light that shines on this work in the coming years, the better. And it may all happen very fast, we will never know.