How AI Systems Are Driving Wall Street Decisions (And the Risks Leaders Ignore)

0:00
You’ve built AI systems that move
billions on Wall Street. What did those
0:02
years teach you about risk? Um, well,
0:08
the final systems era and automation
0:11
decision-m and just looking at your
0:14
early algorithms powered top performing
0:17
financial systems.
0:19
So, yes, first let me tell you just
0:21
something I learned from Wall Street.
0:23
So, I didn’t come from a Wall Street
0:24
background or whatever. I was always
0:26
interested in intelligent systems and AI
0:29
systems and collective intelligence
0:31
systems. So the idea of collective
0:33
intelligence is many minds are better
0:35
than one. Two heads are better than one.
0:37
Imagine what a million brains could do.
0:39
Right? That was the idea. And I’ve run
0:41
IQ company for many years since 1993.
0:44
And we’ve been working on these systems
0:46
for various clients. And then we
0:48
thought, you know, we really want to
0:50
prove this would be around 2000. We kind
0:52
of had this thought. We want to prove
0:53
the power of collective intelligence. It
0:55
was a new idea. People hadn’t talked
0:57
about crowdsourcing yet. We were
0:59
actually building crowdsourcing systems
1:01
before people came up with that name.
1:02
Interesting.
1:03
And we wanted to show that this kind of
1:06
approach of collective intelligence of
1:08
millions of people, even if we’re the
1:10
people weren’t geniuses, just average
1:12
people. When you put them together, if
1:13
you did it in the right way, you could
1:14
get really intelligent performance,
1:16
better than the best of the best. And so
1:19
we had an intern. Actually, we were
1:20
having pizza and beer, and an intern
1:22
said, “Well, you know, if we could do it
1:24
in the stock market, that would show
1:25
everybody.” And that was the beginning
1:27
of what turned out to be a 14-year
1:30
company called Predict Wall Street,
1:32
which I found it. And the idea was we
1:35
would get input from millions of just
1:37
average everyday investors, you know,
1:39
maybe they have a share or two of Apple
1:41
or something and they would go on their
1:42
TDM air trade account or Schwab account
1:45
and um we would ask them what they
1:47
thought about stocks and we’d get input
1:49
from millions of people and if we
1:50
combined their input in the right way,
1:52
the idea was we could beat the best pros
1:54
on Wall Street. And at the at the
1:56
beginning pretty much everybody said,
1:58
“This is crazy. This is never going to
1:59
work. You’re wasting your time. You
2:01
know, you’re a smart guy. why don’t you
2:03
do something else? I mean, I heard that
2:04
from a lot of people, but I just really
2:06
felt like this could work. It took a
2:08
long time and in the end it did. We were
2:10
able to power a hedge fund that ranked
2:13
in the top 10. So, we beat most of the
2:14
Wall Street pros, but it was powered by
2:17
millions of everyday people. It was 100%
2:19
powered by the millions of everyday
2:21
people.
2:21
So, that was collective intelligence.
2:23
But probably the most useful thing that
2:26
I learned in the process of doing this,
2:28
I had to learn about Wall Street and
2:29
everything is the way that the Wall
2:31
Street traders think about the market. I
2:34
kind of naively thought, well, you know,
2:36
you’re sure this stock is going to go up
2:38
and so you put the money on that stock.
2:40
That’s not at all how the sophisticated
2:42
people look at. It’s all probabilistic.
2:44
They only need to be like 51% sure, like
2:47
a little bit more than chance that this
2:49
will go up. And if they can do that
2:51
consistently, they can make billions of
2:53
dollars. And so it’s this notion of just
2:56
shifting the odds a little bit in your
2:57
favor more is how the money is made.
3:00
That’s a very powerful idea. And I’ve
3:02
applied that to AI safety because in the
3:05
AI world, there’s this notion, it sounds
3:08
a little science fictiony and scary,
3:10
poom. Poom, probability of doom. What is
3:14
the probability that AI kills us all?
3:17
Right? Which some people worry about. It
3:19
sounds like Terminator or something. But
3:21
actually, I mentioned Jeff Hinton
3:23
earlier, the Nobel Prize winner,
3:24
godfather of AI, quit Google so he could
3:27
talk openly about this.
3:28
He is openly said he thinks the chance P
3:32
Doom his probability is 10 to 20%. Which
3:35
the good news is 80% it’s all great, but
3:38
20% chance that it killed us all, I
3:40
mean, that’s way too high, right? So
3:42
this notion
3:43
in Wall Street of shifting the odds in
3:45
your favor is how I now look at things.
3:48
And I say for AI safety, we don’t have
3:50
to solve it and guarantee that it’s, you
3:52