A couple of years ago my mathematician friend Juliette, of An almanac of curious thoughts, mentioned an interesting thought for how AI might acquire generality: an integration between deep learning and standard programming, by analogy with the human right and left brain hemispheres. I wondered at the time whether the Wolfram Language could bridge this gap, providing both the ‘standard programming’ and its own representation of the physical world.
Pleasingly I noticed yesterday that Stephen Wolfram is working on integrating Wolfram Language with ChatGPT, which might amount to exactly that. Deutsch argues that general intelligence needs chains of reasoning that construct explanations.
Perhaps the Wolfram Language could provide this, as it already functionally is a kind of low resolution calculus for the physical world. Meanwhile the deep learning side of the Large Language Model ('LLM') could provide the data upon with Wolfram Language code operates, rather like the human brain unconsciously processes input data and delivers up the processed results to the conscious mind for deliberation, often what we call 'intuition'.
This might be the key that unlocks generality.
