1d • General discussion
Energy, brains and future AI ?
This follows up a discussion, in Challenge Lecture 7(1) of 16July26, about the energy demands of AI in comparison to that of the brain.
A recent IEEE Spectrum article happens to address the same point - that brains are super efficient, by at least 3 orders of magnitude compared with the most optimistic interpretation of AI capabilities, and by many more orders of magnitude compared with pessimistic interpretations.
So brains are astonishing, in what they achieve with so little energy.
The article is techie, about how a future AI might become more like brains in energy efficiency, delving into the semiconductor-physics of MOSFET transistors. It hinges on the presence of a fourth connection in those transistors, compared with three connections in bipolar transistors. I encountered that fourth connection many years ago and, like everybody else, I wired it up in the recognised way that neutralised any actions the connection might cause - actions which were considered spurious and uncontrolled in manufacture.
The article talks of how these ‘spurious’ actions can make the single MOSFET like a brain neuron, consuming far less energy than neuron-simulation in either conventional hardware or software.
Very early days, because no-one has yet interconnected these four-terminal MOSFET ‘neurons’ to do anything brain-like, or be able to process language.
So mega-power consumption remains likely for AI in the foreseeable future - a problem Steve highlighted. The energy inefficiency of AI is a basic reason why I doubt that what it does it copies what our brains do.
0
0 comments
Energy, brains and future AI ?
skool.com/stevekeen-free
Don't study mainstream economics. Learn 50+ years of Real Economics here. Download FREE Book Bundle (3 Books). No email required.
Leaderboard (30-day)
1
+2
2
+1
3
+1
Powered by