00:00:00
We've been working on AI safety for two
00:00:01
decades at least.
00:00:02
Yeah. I was convinced we can make safe
00:00:04
AI, but the more I looked at it, the
00:00:06
more I realized is not something we can
00:00:08
actually do.
00:00:09
You have made a series of predictions
00:00:11
about variety of different states. So,
00:00:13
what is your prediction for 2027?
00:00:19
In 2 years, the capability to replace
00:00:22
most humans in most occupations will
00:00:24
come very quickly. I mean, in 5 years,
00:00:26
we're looking at a world where we have
00:00:28
levels of unemployment we never seen
00:00:30
before. Not talking about 10% but 99%.
00:00:34
And that's without super intelligence, a
00:00:36
system smarter than all humans in all
00:00:38
domains. So, it would be better than us
00:00:40
at making new AI. But it's worse than
00:00:43
that. We don't know how to make them
00:00:44
safe. And yet, we still have the
00:00:46
smartest people in the world competing
00:00:47
to win the race to super intelligence.
00:00:49
But
00:00:50
what do you make of people like Sam's
00:00:51
journey with AI? So decade ago, we
00:00:54
published guard rails for how to do AI,
00:00:56
right? They violated every single one
00:00:59
and he's gambling 8 billion lives on
00:01:01
getting richer and more powerful. So I
00:01:03
guess some people want to go to Mars,
00:01:05
others want to control the universe. But
00:01:08
it doesn't matter who builds it. The
00:01:09
moment you switch to super intelligence,
00:01:11
we will most likely regret it terribly.
00:01:14
And then by 2045,
00:01:16
now this is where it gets interesting.
00:01:19
Dr. Roman Gimpolski, let's talk about
00:01:22
simulation theory.
00:01:22
I think we are in one and there is a lot
00:01:24
of agreement on this and this is what
00:01:26
you should be doing in it so we don't
00:01:28
shut it down first.
