AI and Working Memory

A plausible theory; certainly at least part of the explanation.

When an AI system solves a difficult mathematical problem, the usual explanation is that it has become more intelligent.... [but also] AI has access to a vastly larger working memory than the human brain.

Or, more precisely, it has access to an enormous external symbolic workspace that performs many of the functions that working memory performs in humans.

This difference may be especially important in mathematics.

Emphasis in the original. 

One AI recently informed me that I have "used 50x as many tokens as The Hobbit" on that platform. That's a silly thing to say. Tolkien didn't use tokens to write anything, so it's incommensurable; but his experience of authoring the book also involved his decades of study, wide reading in several languages, and so forth before he even sat down to write. (The process for  composing LotR was far more intricate.) If you asked one to 'compose The Hobbit' it would not use anything like the number of tokens it would take to replicate in training cycles what Tolkien did in study; it doesn't have to do so.

The AI experience is that there have been vast cycles of training that have gone into building this 'working memory' (or, as noted, its analog). It can practically draw on much more than even Professor Tolkien, or than Thomas Jefferson dining alone. 

That in itself makes AI extremely useful partners to humans in thinking. We should take care to continue to train our own memories, even to recover the benefits of the discipline of wide reading and study. Yet neither ought we to disregard the clear advantages of such a practical increase in working memory.