I wrote this today to a Reform UK councillor, who had
offered an AI-generated response to help work out our strategy to resolve a
problem in local government.
<diatribe>
How we use AI is a very serious matter.
AI can maybe be cute, cuddly and helpful, as its proponents
make out. But it is seriously fallible. It doesn't understand context. It has
no connection to reality. It cannot make an observation to determine whether
something is true or false. It cannot distinguish facts from opinions, and it
cannot tell truth from advocacy. And if it has to resolve a contradiction, it
can only do so in an arbitrary way, such as sampling opinions and taking the
loudest. So, it tends to amplify rumours. And that's a big problem when people
accept its mistakes and re-broadcast them. That makes it dangerous. And then
there are the legal responsibility issues...
I do think that we as a party need to take a sane, objective
view on AI. By all means use AI if you want. But when you communicate with
others, and you have used AI in constructing what you tell them, you really
ought to check before you send it that what AI told you is actually so. There
are far too many lies and half-truths in the world already, and AI in the hands
of those who don't understand its weaknesses can only make the problem worse.
</diatribe>
No comments:
Post a Comment