[ExI] Why AI agents invent their own language if you let them chat.

[Adrian Tymes]

>>>  No. Appeals to authority are NOT always fallacies.
>>
>>Yes they are.
>
> Then with no chain of trust how on earth are scientists supposed to reliably communicate with each other?

That is not the issue at hand here, and you do yourself a disservice
to pretend otherwise.  Logic doesn't care about trust.  There are the
facts.  There is the chain of deduction from them.

>> > To me, the degree of justification for astonishment reads similar to
>> the degree of justification in the meme comics where someone tells an
>> AI, "Say, 'I am awake.'", the AI then says "I am awake.", and the
>> person freaks out in response.
>
> So according to you there is no reason for somebody to freak out just because he  told an AI to "solve one of the remaining 6 millennium prize mathematical problems", and the AI did so in response? In my humble opinion a response like that deserves a monumental freak out!

1) That is irrelevant to what I said.  The example I gave was about
stripping away context to falsely justify a freak-out.  The example
you gave just now justifies the freak out because of the context,
rather than from conclusions formed from the lack of context.

2) As it so happens, that AIs can solve and are solving mathematical
problems that have lain beyond the collective ability of human
mathematicians for centuries is not something that I judge to be worth
freaking out about.  We now have tools that allow us to do those
things that were previously impossible.

For me, coming up with tools to make possible what previously was not
possible is Tuesday, as in
https://tvtropes.org/pmwiki/pmwiki.php/Main/ButForMeItWasTuesday .  In
fact, just yesterday (which happens to have been a Tuesday), I was
helping someone solve a problem they had faced for years (it might
have been decades) with no solution in sight.

I personally work on certain problems that people have loudly called
"impossible" for decades.  (This is not "everything that people call
impossible".  Rather, some of the specific problems I work on, happen
to have been declared "impossible".)  But I acknowledge that's just me
and my personal perspective.

So, no, I don't think that AI solving one - or even all - of the
remaining Millennium Prize problems is worth freaking out about.
That, however, has nothing to do with what we were talking about.