[ExI] Future Powerful AI Systems to Be Contained in Military Bases

[BillK]

On Fri, 24 May 2024 at 14:39, <spike at rainier66.com> wrote:
>
> BillK, if we think that over for a few minutes, we realize it makes perfect sense.  Scary as hell, but I trust the military more than I trust the universities or think tanks.
>
> Eh, on second thought, I don't trust any of them.
>
> spike
>---------------------------------------------


Perhaps Schmidt is mainly trying to get people to pay attention to how
dangerous a powerful AI might become.
And, yes, connecting it to the internet is risky, but probably necessary.

I think it all depends on what level of "powerful" AI we are talking about.
If it is an advanced AGI, then it is too late. It will have already
copied itself into computers around the world and be making its own
plans on what happens next.

If it less than AGI, then the problem is stopping it from misbehaving
(i.e. unaligned to our values). The military have the equipment to
destroy an AI that is endangering them. More destructive than private
companies wondering whether they should press the big red "OFF"
button.

Then there is also the concern of powerful AIs in other countries that
are not friendly to the West. Will their AIs go wild? They may be
thinking that they don't mind bad behaviour of their AIs, provided
that it is directed towards the West countries. There are rumours
about foreign AIs already helping with internet attacks and criminal
behaviour.

Interesting times, Eh?


BillK