Noam Chomsky criticizes large language models
as useless —
fundamentally, by the way they are designed — for advancing
understanding of cognition.
This is because cognition plays no part in them.
They don't try to understand anything.
Machine-learning systems can do many practical tasks well — those
tasks that don't inherently involve conceptual understanding of
anything. They can be useful, for good purposes and bad purposes. At
the same time, they can treat their users justly (if released as free
software) or unjustly (if released as
nonfree software
or
only as a "service")
But we should distinguish this from intelligence, and reserve the
term artificial intelligence for artificial systems that can understand.