A team of AI scientists has lately made the claim,
They've built a pair of AIs that, combined, can do the same.
Intrigued, we ask ourselves, what does this product from Deep Mind
Portend for mathematics that's produced by humankind?
It might well be supposed that an AI that medals hath
A comparable talent for sophisticated math.
But as the ancient proverb warns, our lives have many slips.
The vintage in the cup may spill and never reach the lips.
A human doing math draws on their own experience
Of the external world, profound, immersive, and intense.
The symbols used by AIs are not comparably grounded,
So claims that AIs "understand" their content are ill-founded.
A human has a repertoire of spatial concepts which
Is large and multifarious, wide-ranging, broad, and rich.
The concepts in the scope of AlphaGéométry2
Pervade the IMO, but they are limited and few.
To read math is a difficult but necessary art.
To read a problem statement as it's written is a start.
These AIs do not work without a manual translation
That turns the problem statement into their preferred notation.
Moreover, it is common that a human math'matician
Will tell us that their thoughts reflect some kind of "intuition".
We don't have a good theory to explain what that might mean
But there's no sign it's been attained by any such machine.
So AI mathematics programs, we can safely state,
Are not a patch on human beings at the present date.
How long this will continue is unknown, I must confess.
Forever? For a century? A decade? Maybe less?
This is part of the collection Verses for the Information Age by Ernest Davis