[ExI] A Paranormal Prediction

[Kelly Anderson]

On Sat, Jan 6, 2024 at 12:41 PM Jason Resch via extropy-chat
<extropy-chat at lists.extropy.org> wrote:
>
>
>
> On Sat, Jan 6, 2024, 2:27 PM Kelly Anderson via extropy-chat <extropy-chat at lists.extropy.org> wrote:
> The advantage of an AI-tocracy is that it can be open source and everyone can run a local copy on their own computer to verify the public pronouncements of executives or the rulings of judges.
>
> The disadvantage is that if they are public they can also be games by adversaries that run closed systems. If international affairs are like a poker game, then there are advantages to keeping ones cards close to the chest. Perhaps domestic government can be run by open source AI while international government uses closed source AIs.
>

Very interesting ideas Jason. Because of the large size of the data
that would be required for an AI smart enough to run a government, I
think that technologically it would not be feasible for everyone to
have their own copy for verification. Perhaps if they ran out of the
same cloud, you could duplicate basic opinions. Much of the data that
such an AI would be using to making decisions would indeed be secret,
as you point out. Can't have copies of that running around, now can
we. Bradley Manning and Edward Snowden showing how that can be
disruptive to the flow of international relations. And very shortly,
we MAY find globalism itself to be threatened as a war starts to
disrupt low friction international trade. Again, a story for another
day.

Another problem with your idea is that if you ask AI the same question
twice, you normally don't get the same answer. You can try this with
any of the popular LLMs out there. As they grow to be more than just
LLMs, I think you can take this point further forward. In fact, many
AIs are now being trained on "simulated" data, rather than data
generated by humans. It's like AlphaGo playing itself a million times,
but with conversations. As AI talks to itself, it gets smarter,
without direct human involvement. This is what you would predict. And
this seems to be a big part of what many AI companies are doing to
train their next generation models.

-Kelly