Learning the language of life, the universe and everything – LLMs go metaphorical

More than two years after the advent of ChatGPT, we all know a little bit about LLMs or ‘large language models’ (although we, that is ordinary people, don’t really know what that means, especially ‘model’).

So, just to recap, and using a definition provided by Claude (as the phrase is not yet in the Oxford English Dictionary!): “In essence, a large language model is like an incredibly sophisticated text prediction and generation tool that learns by analyzing enormous amounts of human-written text.” (Claude’s ‘definition’ was much longer, but that’s basically it) And, as far as I get it, the ’tool’ is a complex computational model or artificial neural network.

As this year’s Nobels have shown, AI and computational modelling have gone beyond human language and things have become quite metaphorical (you can never get away from human language in that respect!). We hear or read a lot of phrases that metaphorically suggest that AI is developing a deep understanding of complex systems, much like learning a nuanced language.

People now say: “AI is learning the language of proteins” …. – that was the alpha-folding Nobel stuff

But AI is learning a lot more.. Here are just a few of many examples

“AI learns the language of life”

“AI is learning the language of cells”

“AI is learning DNA’s hidden language”

“AI is learning the language of chemistry”

“AI is learning the language of climate” – that phrase hasn’t been used yet, but could

And now: “AI is learning the language of atoms”!!

Similarly, we hear about AI “cracking the code of life’ or “cracking the code for proteins’ amazing structures” and also…. “Cracking the code of cancer“, “Cracking the code of bacterial communication” and much more. Or even, to throw in another rather stale metaphor: “Evo: The AI That’s Decoding Life’s Genetic Blueprint“.

What does this mean? Is all this just metaphor or rather hyperbole? Or are we on the verge of AI speaking the language of life, the universe and everything??

PS 19 December 2024

For some answers, basically saying ‘no’, you can now read Philip Ball’s little article where he links these questions to Searle’s Chinese room argument.

Image: pxhere