Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:01
Radio Andy. Hey, it's
0:03
Andy Cohen. Join me on Andy Cohen
0:05
Live, where it's just you, me, and
0:08
some of the world's biggest celebrities. Paris
0:10
Hilton, Chelsea Handler, Seth Rogen. I love
0:12
you, Miley. Thank you so much. You
0:14
can listen to Andy Cohen Live at
0:17
home or anywhere you are. No car
0:19
required. Download the SiriusXM app for over
0:21
425 channels of
0:24
ad-free music, sports, entertainment, and more.
0:26
Subscribe now and get three months
0:28
free. Offer details apply. At
0:32
the UPS Store, we know things can get busy this
0:34
upcoming holiday. You can count on us to be open
0:36
and ready to help with any packing and shipping or
0:39
anything else you might need. Is there anything
0:41
you can't do? Um, actually, I don't
0:43
have a good singing voice. Ahem.
0:45
The yup-yup. Noop. But
0:48
our certified packing experts can pack and ship
0:50
just about anything. At least that's
0:52
good. Your local, the everything you need to
0:54
be unstoppable store. The UPS Store. Be unstoppable.
0:57
The UPS store locations are independently owned. Product services,
0:59
pricing and hours may vary. See Center for Details.
1:09
I'm Alan Olga and this is
1:11
clear and vivid conversations about
1:13
connecting and communicating maybe
1:19
we don't want moral AI we want
1:21
obedient there we wanted to do what
1:23
we want and we don't want it to kill
1:25
us but if
1:27
it's too moral it might tell us to stop
1:29
doing a lot of things we're doing what
1:32
a moral AI stop us from
1:35
factory farming from killing
1:37
billions of sentient creatures very painfully for
1:39
food would it would it intervene and
1:42
even at a personal level I don't want it
1:45
what would I do what would I think
1:47
of tax software that's very AI generated and
1:49
won't let me exaggerate the size of my
1:51
home office what would I
1:53
think of what would I think of my self-driving car
1:55
that refuses to drive me to to a bar because
1:59
I drink too much Go
2:01
back home and spend time with your family. That's
2:03
psychologist Paul Bloom. We
2:06
had him on the show a few years
2:08
ago when he and I had an enjoyable
2:10
tussle over whether empathy was actually a useful
2:12
thing. We invited
2:14
him back for this episode, the last
2:16
of three special shows on AI, because
2:19
I wanted his take on kind of an important
2:21
question. AI bots
2:23
are presumably devoid of empathy.
2:26
They're just machines, after all, right? So
2:29
could they ever be moral? Could
2:32
they be given a sense of right and wrong? Turns
2:35
out it's complicated. Paul,
2:38
this is very interesting for me to be
2:40
talking with you because we've been talking a
2:43
lot about artificial intelligence on this show. And
2:45
the question that looms in a lot of people's
2:47
minds is, will it turn against us
2:49
and do us harm? And one solution
2:51
that's been offered is that we
2:54
made it, so why don't we just tell it
2:56
to be good, to be moral? And
2:58
you had a really interesting answer to that idea
3:00
in an article in The New Yorker. And
3:03
I wanted to talk to you about that. Why can't
3:05
we just tell them to be good? Yeah,
3:08
I'm glad to be talking to you about
3:10
this. I agree with you. I think AI
3:12
is the biggest news that's come along in
3:15
a very long time. And it
3:18
can imagine it transforming the world for
3:20
the better in enormous ways. It could
3:22
also kill us all. I
3:25
have to go now. That's
3:27
one of the two. I
3:30
guess we'll find out. And you're
3:32
right. So
3:34
one long-standing solution to the worries
3:37
people have about AI, either worries
3:39
that AI itself may turn
3:41
malevolent in some way or
3:43
accidentally cause harm, or
3:45
that bad agents could use AI to do
3:48
terrible things, is to make AI moral. And
3:51
this is sometimes called the alignment problem, which
3:53
is you want to give AI a
3:56
sense of morality, a sense of goal similar to what
3:58
people have, And in that way,
4:00
it will
4:02
avoid doing harmful and terrible things.
4:04
If we just align AI with
4:06
our morality, which morality
4:09
are we going to choose? Oh, that's
4:11
such a good question. That's an immediate
4:13
problem here. Because, you know, if
4:16
I go to... It's already somewhat aligned in
4:19
that if you go to chat GBT
4:21
or Bing or Claude or whatever and
4:23
ask it moral questions, it will give
4:26
you answers that kind of resonate with
4:28
our intuitions. But your
4:30
question of whose morality is
4:33
a great one. If I asked chat GBT, and I
4:35
have done this, what do you
4:37
think of two men marrying? It
4:39
says, it's fine. There's nothing wrong
4:41
with it. What do you think
4:44
of a woman getting an abortion? It's fine. There's
4:46
nothing wrong with it. But many people
4:48
around the world, it doesn't
4:50
match with their morality. They would say
4:52
that gay marriage is morally wrong. They'd
4:55
say a woman having an abortion is morally
4:57
wrong. So that's the first question, which is
4:59
whose morality? And there's no
5:01
way around it. If it's going to align
5:03
with your morality, it's going to
5:05
be a different morality than somebody from raised
5:07
in a very different culture and environment.
5:11
And I think to some extent, I think
5:13
we just skirt the problem. We say, OK,
5:15
fine. Our morality. Let's connect it to our
5:17
morality. And
5:20
then we have various problems that
5:22
arise. It turns out to be very
5:26
difficult to program a machine to be
5:28
moral and not have it, you know,
5:30
choose to satisfy other goals instead. So
5:34
the main worry, one of main worry
5:36
about AI is a sort
5:38
of unintended consequences. The standard example,
5:40
I think from Nick Bostrom, is
5:43
you ask an AI to just make
5:45
paperclips, as many paperclips as possible. And
5:47
then in a fraction of a second,
5:49
it figures out, well, if it kills
5:51
everybody and turns everybody into
5:53
paperclips, that will satisfy the problem. You
5:56
don't want us to do that. The problem is that even if
5:58
you tell it not to harm us. Which
6:00
us do we mean? Do
6:02
we mean it's okay to harm our
6:04
adversaries but not us? Or
6:07
do we mean don't harm the us that's all
6:09
of humanity? Or do we
6:11
mean more? Would a moral AI stop us
6:13
from factory farming? From
6:16
killing billions of sentient creatures very
6:18
painfully for food? Would it intervene?
6:21
Would it stop us from doing war? One
6:25
of the points of some stuff I've
6:27
written is making the argument that maybe
6:30
we don't want moral AI. We want
6:33
obedient AI. We want it to do what
6:35
we want and we don't want it to kill us.
6:38
But if it's too moral, it might tell
6:40
us to stop doing a lot of things we're
6:42
doing. Could you imagine what
6:45
the military would think of military AIs which decide
6:47
to be pacifists or decide, well, this is an
6:49
unjust war. I'm not going to shut
6:51
down the tanks and the airplanes. I'm going
6:53
to lower your security system because this is an
6:55
war we should be fighting. Or just kill our
6:58
enemy. And the AI decides what's
7:00
the enemy. That's right. Maybe
7:03
the AI is very smart and moral
7:05
and decides we're the baddies. I've
7:10
thought it over and you're it. Yeah,
7:12
you're the villains. So people say
7:14
they want moral AI. But
7:16
when push comes to shove, I think both
7:19
have a sort of global general scale for
7:21
military and industry and so on. We don't
7:23
want it. And even at a personal level,
7:25
I don't want it. What would I
7:28
do? What would I think of tax
7:30
software that's very AI generated and won't let me
7:32
exaggerate the size of my home office? What
7:35
would I think of my self-driving car that
7:38
refuses to drive me to a bar because
7:41
I drink too much? Go
7:43
back home and spend time with your family. You
7:49
know, when we're talking about weapons and what
7:51
the Department of Defense would be happy with
7:53
or not happy with, just the idea of
7:56
having an autonomous weapon, which
7:58
we already seem to be able to do, we're the ones. decides
8:00
at the very last second whether to
8:02
kill somebody or not based
8:04
on its own evaluation. You
8:07
know, given some guidelines by the person
8:09
firing, but mainly evaluating whether that person
8:11
that it has in its line of
8:13
sight fits the rules or not. It
8:16
decides, it makes up its own mind. Yeah. Could
8:19
AI in general develop a mind of its own,
8:21
do you think? Well, that's
8:25
a hard question. It's a hard
8:27
question where AI is going to
8:29
go from here. So
8:32
take your case where you give instructions on who
8:34
to kill and who not to kill. I
8:37
guess the question people would want to know is
8:39
can AI decide to override these instructions? Particularly
8:41
if we build a moral AI or
8:44
if we build an AI that's in some sense
8:46
self-interested, there's always an option
8:48
that could stop listening to us. And
8:51
here there's sort of a cluster of
8:53
questions that nobody knows the answer to, which
8:55
is, you know, right now, the
8:58
machines we have, the large language models show
9:00
no sign of doing this. They're very obedient.
9:02
I tell it to what to do. The
9:04
only cases that won't do
9:06
what I tell it to do is when it's been
9:08
programmed not to. So if I ask it to develop
9:10
a deadly virus, it tells me, I'm sorry, I can't
9:12
do that. There's all sorts of things that
9:14
will say that. But beyond that, it does what I tell it
9:16
to. Will a future
9:18
version stop doing that? I
9:21
don't know. There's either it
9:24
either it will or
9:27
it won't because that's not where the technology
9:29
is going or it won't because we're going
9:31
to stop building AIs which have so much
9:33
power. And as you know, there's a large
9:35
movement of people who argue that we should
9:37
stop development on the AIs because they're terrified
9:39
of the consequences. Well, if the
9:41
good guys stop development of AIs and
9:43
the bad guys don't, that's
9:46
an open door, isn't it? Yeah,
9:48
that's so one of the arguments against it
9:50
is that assuming we're assuming we're
9:52
the good guys, for the sake of argument,
9:54
if we stop, the other guys will develop
9:56
AIs and they'll have less less.
10:00
restrictions and they will get ahead
10:02
of us. So in some sense, this
10:04
is an arms race. Right, an
10:06
arms race. It sounds almost unavoidable and
10:09
therefore, regulation, international regulation sounds hard
10:12
to imagine for the same reason
10:14
because they want to
10:16
regulate stuff that's bad for them but
10:18
not necessarily bad for their adversaries. Yeah,
10:22
there's been cases where we have had
10:24
international regulations over biological
10:27
weapons over things
10:29
like cloning, various forms
10:32
of human experimentation. It's
10:34
an open question how much countries obey them but
10:36
at least we have some sort of general
10:38
restrictions for certain things. I
10:41
think the problem with AI is too many people and
10:43
I'm not talking here about China or some
10:46
other country, I'm not about United States
10:48
say, too many people want more AI
10:51
because correctly enough they think this could
10:53
really improve people's lives. What
10:56
if we gave up on AI and it
10:58
turned out that if we just worked a
11:01
bit harder, it could cure diseases. It
11:03
can solve deep social and environmental problems
11:06
that we can't imagine the solutions to.
11:08
It could really improve our lives. It's
11:12
funny, I've never seen a technology before that had
11:14
so much potential for both
11:16
terrible consequences and wonderful consequences. Yeah,
11:19
I was just thinking as you were saying that
11:21
that unlike nuclear power, where it
11:24
would be very good to get energy from
11:26
nuclear power but not nuclear bombs, unlike
11:29
artificial intelligence, nuclear power doesn't have the
11:31
ability to keep learning on its own,
11:35
learning how to mix those two good and bad things
11:37
in a way that could be bad for everybody. That's
11:40
right and people who are very worried
11:42
about AI often give the analogy of
11:44
meeting up with
11:48
a super intelligent species or asking what
11:50
do we as a highly intelligent species,
11:52
how do we deal with those who
11:54
are less intelligent, less capable than
11:57
us? We put them
11:59
in cages. We exploit them, we use them.
12:03
And it's possible that
12:05
AIs will do that to
12:07
us at some level. Not
12:09
because, you know, we've been shaped by
12:11
natural selection. We have all these aggressive
12:13
and sexual and malevolent desires. They won't.
12:17
But they may have other
12:19
things that lead to bad
12:21
consequences. For instance, most machines
12:24
want to do what they're told, want to
12:26
satisfy a task. And if
12:28
you set an AI to a task,
12:30
it may recognize well that humans could
12:32
shut it off. And
12:34
so the way to stop that from happening is
12:36
to shut off humans first. Intelligence
12:44
doesn't seem to me to lead
12:46
necessarily to moral behavior.
12:49
I think some people feel it'll be so smart,
12:53
it'll develop its own sense of morality.
12:56
I don't see that happening. I
12:59
don't see that either. I think there's some
13:01
relationship with intelligence and moral behavior
13:03
in people, in part because if
13:05
you're smart enough, you could kind of work with
13:07
another person for long-term solutions. You know, instead of
13:09
me stealing from you and you stealing from me,
13:11
we could trade. And we could think to
13:13
ourselves, this actually works better in the long run. But
13:16
there's no shortage of really smart
13:18
people who are also terrible. I
13:22
think whether you're good or bad depends
13:24
on what you want, not your capacity
13:26
to reason and your capacity for rationality.
13:29
In fact, the more terrible you are and
13:31
the more intelligent you are, possibly
13:34
the more likely you are to rise
13:36
to the top and
13:38
cause even more damage. Yes.
13:40
So if it turned out that we ended
13:43
up in a conflict with AI, with different
13:45
interests, different goals, it's going to be very
13:47
unfortunate if it's much smarter than we are.
13:51
Like any adversary, you'd rather have them
13:53
done than smart. Yeah,
13:56
exactly. What about the
13:58
tendency to want to survive? Is
14:00
that something that we don't have to worry
14:02
about with AI's or
14:05
is it something that will probably happen
14:08
where they'll develop this need,
14:10
this urge, this impetus towards
14:13
survival and anything that gets in
14:16
the way of that or is perceived by the AI to
14:18
get in the way of it makes
14:21
people causing that to
14:24
be the enemy of AI? What about
14:26
survival? It's a
14:28
good question. People and other animals
14:30
have a strong instinct to survive
14:32
because those that didn't wouldn't reproduce.
14:35
And so natural selection drives us
14:37
with a very powerful survival instinct
14:39
and other instincts that are aggressive.
14:41
AIs don't have that. A
14:44
simple AI, if you just tell it
14:46
destroy yourself, erase your memory, it will.
14:50
The worry that some people have
14:52
is it could develop it. And one
14:55
way it may develop it is that once
14:58
you have any other goal, a desire
15:01
to survive comes with that goal.
15:05
If I build an AI and its goal is to
15:07
write poetry and it just writes
15:09
poetry and it's smart enough, it
15:11
will reason I better keep on going. If
15:14
someone shut me off, I couldn't write poetry
15:16
anymore. And so if it
15:18
takes steps to protect itself and write
15:20
more poetry, that would be
15:22
rational for his desire. So a desire to
15:24
survive is interesting because it seems to be
15:27
a consequence of every other desire. You can't
15:29
do things when you're dead. So
15:32
don't be dead. Don't be dead.
15:34
Good advice. I was thinking
15:36
in more complicated terms because there were viruses
15:39
that had the ability to
15:41
evade elimination, computer
15:43
viruses. And if
15:45
one got into AI and developed
15:48
a symbiotic relationship with it, where
15:51
the virus stays alive and
15:53
the AI stays alive by exploiting it, the
15:55
situation in the same way that the
15:58
virus does, then there's a problem. Then
16:00
we've got anti-malware and
16:02
we're the malware. Alan, I
16:04
thought I worried about things before talking
16:06
to you, but I've never worried about
16:09
a virus co-mingling with an AI to
16:11
become an especially malevolent fan. So
16:13
now I'll worry about that too. Well,
16:16
you know, with this series of podcasts
16:18
that we're doing on AI,
16:21
I don't want to scare people. I
16:24
get scared myself when I see that
16:26
concerns are not expressed very
16:29
seriously. There are
16:31
some people who helped create artificial
16:33
intelligence who are worried about
16:35
dire consequences. And they put it in terms
16:38
as stark as the ones you used earlier
16:41
in the conversation that could kill us all.
16:44
On the other hand, there are people who kind
16:46
of make fun of it. That
16:49
doesn't sound like a balanced approach
16:52
because we're up against something we've never experienced
16:54
before, which the same as being visited by
16:56
an alien civilization, smarter these
16:59
things get. Well, let
17:01
me ask you, you've been immersed in this for a
17:03
while. If you
17:05
could, would you have a moratorium on
17:07
AI research? Would you give it a
17:09
break for a few decades? I
17:12
would, but you can't, so I wouldn't suggest
17:14
it. Yeah. I know some
17:16
very serious people have suggested it, but it's
17:19
such an easy agreement to break. It's
17:22
true. I'll stop your research. I'll just do a
17:24
little research on my own and see what I
17:26
come up with. Right. You
17:28
just need somebody in a basement who has to
17:30
write equipment. And it is
17:33
very hard to block research on it. You
17:35
could shut down through law, open AI
17:37
and Microsoft and Google and all that and
17:39
tell them not to do it or they'll
17:41
go to prison, but it's not just them.
17:45
And there's people all around the world. Exactly. Yeah,
17:47
you're right. So
17:50
it may be, since you can't stop it,
17:53
don't try. Instead, try to regulate it and try to
17:55
keep an eye on it. I wonder
17:58
if you can work on AI then. keep
18:00
an eye out for other AIs and
18:02
negotiate with them or battle with them.
18:05
It seems to push the problem back
18:07
a little bit like how do you
18:09
know that the Guardians you have appointed
18:11
are your own motivation. I'd be the
18:14
wrong one to set up a system
18:16
for that. It needs
18:18
a bit of tweaking. When
18:26
we come back from our break, Paul Bloom
18:28
dives into the question of whether AI bots
18:30
could ever be conscious, whether they
18:33
could feel. And that led to
18:35
him asking a question he never believed he'd have
18:37
to ask. Should they be given
18:39
the vote? Just
18:45
a reminder that Clear and Vivid is
18:47
non-profit with everything after expenses
18:49
going to the Center for Communicating
18:52
Science at Stony Brook University. Both
18:55
the show and the center are dedicated to
18:57
improving the way we connect with each other
19:00
and all the ways it influenced our lives. You
19:03
can help by becoming a patron of
19:05
Clear and Vivid at patreon.com. At
19:08
the highest tier, you can join a monthly
19:10
chat with me and other patrons, and
19:13
I'll even record a voicemail message for you.
19:16
Either a polite, dignified message from me
19:18
explaining your inability to come to the
19:20
phone, or a slightly snarky
19:22
one where I explain you have no interest
19:24
in talking with anyone at the moment. I'm
19:28
happy to report that the snarky one is
19:30
by far more popular. If
19:33
you'd like to help keep the
19:35
conversation going about connecting and communicating,
19:37
join us at patreon.com/Clear and Vivid.
19:41
patreon.com/Clear and
19:44
Vivid. And
19:46
thank you. Disney
19:48
Plus and Hulu are better together in the
19:50
Disney bundle for a low price. On Disney
19:53
Plus, get into the thrilling Percy Jackson
19:55
and the Olympians and Marvel Studios Echo.
19:57
On Hulu, the stakes are high. and
20:00
effects his feud. Capote versus the Swan
20:02
and the new season of Life and
20:04
Best. All of these and more are
20:06
now streaming. Get the Disney Bundle with
20:08
Disney Plus and Hulu. See disneybundle.com for
20:10
details. At the
20:13
UPS store, we know things can get busy this
20:15
upcoming holiday. You can count on us to be
20:17
open and ready to help with any packing and
20:19
shipping or anything else you might need. Is there
20:21
anything you can't do? Um, actually, I don't have
20:23
a good singing voice. The
20:26
UPS! No, but our
20:28
certified packing experts can pack and ship
20:30
just about anything. At least that's good.
20:32
Your local, the everything you need to
20:34
be unstoppable store. The UPS store. The
20:36
unstoppable. The UPS store locations are independently
20:38
owned. Product services pricing and hours may
20:40
vary. See center for details. This
20:46
is clear and vivid and now back to
20:48
my conversation with Paul Bloom. He
20:50
recently published a new book on psychology
20:52
and the human brain. You
20:56
have this wonderful book called Psych, which
20:59
is really an introduction to the
21:01
whole field of psychology. And
21:04
I wonder if the way you think
21:06
of the brain and the mind have
21:08
been altered in any way by what, what
21:11
you're thinking about in terms of AI. Yeah,
21:13
it has been. I wrote the book
21:16
during COVID before these machines came out.
21:18
And AI is maybe the
21:21
biggest thing in my professional life. I was wrong about
21:23
where if you had asked me a couple of years ago,
21:26
when would we develop machines? You could have
21:28
a conversation with that can do what chat
21:30
GPT does. I'd say, I don't know, 20,
21:32
30, 40 years. And,
21:35
and it happens so fast and
21:38
it challenges my view on the mind because
21:40
in my book, Psych, I say, look, simple
21:43
statistics, doing analyses
21:46
of large bodies of data won't get
21:48
you that far. But to a large
21:50
part, they do work by
21:52
statistics and analyses of large bodies and
21:54
data. And they do much
21:57
better than me or many of
21:59
my friends. colleagues would have thought they did.
22:02
And this does raise the question of what the
22:04
extent to which the human mind works
22:07
and does its marvelous things, in
22:09
part the same similar, more
22:11
similar they would have thought to chat
22:13
GPT, that we just get
22:15
these enormous bodies of data and we do
22:17
statistics on it. And that's how we're so
22:20
smart. I don't think that that's entirely right.
22:22
I think we have built in
22:24
rules in the head. I think we
22:26
think in ways that the AIs can't, which
22:28
means we don't make the same mistakes they
22:30
do, these weird hallucinations, these weird limitations. But
22:34
still, I am very
22:36
stunned at how well such
22:38
a sort of simple way
22:40
of proceeding has led to what seems to
22:43
be a powerful intelligence. I think
22:45
there's a section in your book where you talk about
22:48
fungi solving math
22:50
puzzles without needing
22:52
to be conscious of what
22:54
they're doing. For all we know, fungi are
22:56
probably not conscious. Let's agree that
22:58
they aren't. What was it? Something about a
23:00
maze. I can't remember the exact way they
23:02
solve puzzles. I figure
23:04
the details either, but they were
23:07
doing roadmaps. They were calculating at
23:09
some level the shortest distance between
23:11
different points and
23:13
doing intricate calculations without
23:15
a brain. And this is
23:17
important because it shows that
23:23
intelligence is different from consciousness.
23:26
We probably already knew this, but being
23:28
smart, being rational, being able to solve
23:30
problems is
23:33
quite separate from having sentience, having experience,
23:35
being able to feel pain
23:37
and pleasure, and so on. And
23:40
I think we know that
23:42
these computers are highly intelligent.
23:45
Some people say, oh, I don't want to call
23:47
them intelligent, but that's just wordplay. They do smart
23:49
things. They act very smart in
23:51
certain ways. I don't
23:53
think they're conscious. Tell me if
23:55
I have this right, that they don't
23:58
need to be conscious to do some kind
24:00
of complicated things. It
24:02
sounds a little bit to me like the baseball
24:04
player who knows instinctively
24:07
where to be in
24:09
the outfield to catch a pop
24:11
fly and is doing all kinds of
24:13
his brain under the surface of consciousness. He's
24:16
doing all kinds of calculations
24:18
in physics. He's
24:20
not aware of it. That's a nice
24:23
analogy. What's under the surface
24:25
for us, it might be for these
24:27
machines, everything's under the surface. It might
24:30
be they don't have the experience of
24:32
toasters. They're just toasters within a sense.
24:34
There's nothing there. Now, whether
24:37
or not they are conscious or could
24:40
be conscious is a question
24:42
of enormous importance. I
24:44
know they're smart. I don't think
24:46
there's the slightest twinge of consciousness
24:49
in these machines. If
24:51
things changed and it looked like
24:53
they had achieved consciousness, then all
24:55
of a sudden we have
24:58
moral obligations to them. All
25:00
of a sudden using them for our purposes is a
25:02
form of slavery. That's
25:05
interesting. Tell me why discovery that
25:07
they're conscious means we have
25:09
to be more aware of their suffering.
25:11
I'm using consciousness in a broad sense.
25:13
I agree. I think the question is
25:15
Jeremy Bentham once said when
25:19
talking about what matters morally, the
25:21
question isn't can it think? The
25:24
question is can it feel? The
25:26
moment these machines can feel, then
25:31
just like I have
25:33
different obligations to an animal that can feel
25:35
than I do to a rock or a
25:37
toaster, all of a sudden you
25:41
have these moral obligations to these things. Shutting
25:43
them off would be murder. Exploiting
25:47
them would be slavery. We'd
25:49
be creating new people in a
25:52
sense. The question would
25:55
come up should AIs get the vote. I
26:00
thought I'd be saying that seriously now, but
26:02
you know, in five years, ten
26:04
years, who knows? That
26:06
reminds me of a section
26:09
of your book, Psych, where,
26:12
as I remember, you were making the point that we
26:15
need emotions
26:18
to be rational, to some
26:20
extent anyway. Is that right? One way
26:22
to look at it is, when you ask the question
26:24
what rationality is, it's the
26:27
capacity to attain your goals. And
26:30
you would call somebody rational intelligence to the
26:32
extent they could achieve their goals. But
26:34
what emotions do is they
26:36
establish goals. Like we talked
26:38
about one of them, stay alive. Having
26:41
a goal of staying alive dictates I act in a
26:43
very different way than if I don't care. Take
26:46
care of my children, develop warm
26:48
relationships, achieve status, and so on.
26:51
And the emotions have been shaped by
26:53
evolution to guide us to
26:55
certain things. And
26:58
AIs don't have emotion in that sense. They just
27:00
have the goals we tell it to have. An
27:02
AI's goal on my computer is
27:04
pretty much make me happy, make
27:06
the person happy, answer my questions. Sometimes
27:10
tell the truth, sometimes make me happy, even if
27:12
it involves making up stuff. And
27:15
so I do think it could have
27:18
the same rationality. But its
27:20
rationality is in the service of whatever goal
27:22
you toss at it. And
27:24
in some way, maybe that's a little bit reassuring. Without
27:28
it, it doesn't want to rule the world. It
27:30
doesn't want to become king. It doesn't want to kill
27:33
us all. It only wants what we tell it to
27:35
want. How do we know
27:37
it's not feeling things? Is
27:39
there a test for inner awareness,
27:41
a touring test for emotions? How
27:45
do you know I'm feeling things? Well,
27:47
you seem like a nice person and you look happy.
27:50
Well, thank you. I'm making happy facial
27:52
expressions. I'm saying all these words. And
27:55
that's the problem with AI's that as
27:57
Harari has said, they're built to exhibit
27:59
intimacy. Yeah to engage is in
28:01
intimacy. So there's two kinds of mistakes
28:04
that you could make One
28:06
mistake is looking at a being with some
28:08
with consciousness and saying it doesn't have it
28:11
that could be terrible That's
28:13
that could be could lead all sorts of monstrosities If
28:15
you looked at me and for some reason you came
28:18
to the belief that I am just dead inside I'm
28:20
making sounds and I'm making expressions, but there's
28:23
nothing happening in me. I'm like
28:25
I'm I'm no more I'm no more sentient
28:27
than a desk or a rock Then
28:30
you could destroy me you could kill me your
28:32
interests mean nothing. There's nothing going on here So
28:35
that'd be one mistake a second mistake
28:37
Which I think people make now is
28:39
they they all of this AI is intimate.
28:42
It says it seems so smart It seems
28:44
can be warm I've
28:47
done work with some colleagues at University
28:49
of Toronto finding that people often think
28:51
AI is more empathic than people
28:54
it could be warm and supportive and so on and
28:57
And and then we may falsely
28:59
assume there's consciousness when there is none There's
29:03
um, there's a guy who worked at
29:05
Google Blake Blake Limoni I mean after
29:07
nastiness name right who and he
29:09
came to the belief that the AI system He was
29:11
working with was sentient was conscious was alive And
29:14
then he complained that Google should not be
29:16
using it without his permission and
29:19
try to get his legal representation Whereupon Google
29:21
Google fired him and
29:25
And and people made fun of them on Twitter
29:27
says you Find me a
29:29
love of you know, and I but I don't know
29:31
what if you were right You're
29:38
talking your New Yorker article about
29:40
Isaac Asimov anticipating this discussion we're
29:42
having by decades and He
29:45
had three rules that robots should be
29:47
programmed with what what are
29:49
those rules? How come they're not working? Yeah,
29:52
Asimov was first the struggle with the alignment
29:54
problem. He wrote these wonderful science fiction Stories
29:57
like I robot which had
29:59
these robots in them. And
30:02
he assumed correctly that people would worry
30:04
about the robots being well behaved. So
30:07
he thought up three laws. The
30:10
first law is a robot should not
30:12
hurt anybody or kill anybody or
30:15
through inaction allow anybody to
30:17
come to harm. The same
30:19
action, no action, not taking an action. That's right.
30:21
That's right. So, you know, if someone's drowning, the
30:23
robot can't just stand and watch them, has to
30:26
ask to help. The second
30:28
law is a robot
30:30
must obey all instructions
30:32
unless it conflicts with the first law. So
30:35
you ask the robot to clean the room, it'll clean the room. You
30:37
ask the robot to murder your next door neighbor, it won't. And
30:40
the third law is a robot
30:42
should protect itself unless it
30:45
conflicts with the second or first law.
30:49
So if somebody tells a robot, go do this dangerous thing,
30:51
it will do it. But otherwise
30:53
it'll try to stay clear of harm. This
30:56
is very clever. It captures certain
30:58
ideas. You want a robot
31:00
to be obedient, but you don't want it to be a
31:02
murder machine. You want it to help
31:05
people, you want it to not harm people.
31:07
And you want it to protect itself. It's
31:09
an expensive piece of machinery. You don't want
31:11
it to kind of just walk off a
31:13
roof for no reason. It's
31:16
really clever, but it doesn't really work.
31:18
And of course, wouldn't it be strange
31:20
if all the morality could be, you
31:22
know, synopsized in three laws.
31:24
So for instance, the first law says
31:26
a robot shouldn't through
31:29
inaction allow anybody to come to
31:31
harm. But if that were really true, then
31:33
if I owned a robot, it would
31:36
run through the streets of Toronto, you
31:38
know, helping people, giving food to
31:41
the hungry, helping people, you
31:43
know, out of burning buildings and everything would
31:45
never, never come back. It would
31:47
be like a Superman spending all this
31:49
time helping others. What
31:53
about the prohibition against harm? Well, would
31:55
a robot stop me if I would
31:57
try to swat a mosquito? Would a
31:59
robot stop me if I tried
32:01
to buy a hamburger and say,
32:03
no, indirectly you're causing suffering to
32:05
non-human animals. There's always subtle moral
32:08
issues that arise that people struggle with
32:10
and you just can't make go away.
32:13
This is even an issue right now,
32:15
not science fiction, for self-driving cars. So
32:18
self-driving cars often face moral dilemmas.
32:22
What if it's on an icy
32:24
road and the brakes don't
32:26
work and it's about to
32:28
slam into two people. Should
32:30
it swerve and slam into a brick wall
32:32
and kill the driver? Does
32:35
it matter if it was one person? Would it matter if
32:37
it's three people? These are hard moral
32:39
problems and you can't make them go
32:41
away by just appealing to these general laws. So
32:48
what are we to make of this whole thing? How
32:51
do you feel personally when you sit at
32:53
your computer and you wonder what
32:56
it's going to turn into in a
32:58
very short time, part of a network
33:00
that's either malevolent or
33:03
beneficial or some unknowable
33:06
combination of both? What can you
33:08
do? What can I do? What
33:10
can ordinary people listening to this
33:13
do to make it
33:15
mostly beneficial? My short answer
33:17
is I don't know. I don't know. You sort
33:19
of asked two questions. I don't
33:21
know what's going to happen and I don't
33:23
know what we can do to make things happen
33:25
better. I share your skepticism
33:28
about saying, okay, let's shut down all
33:30
AI research. I don't think
33:32
that's possible and could be counterproductive. I do
33:34
think it makes sense to sort of tightly
33:36
regulate it and tightly watch it. I
33:39
think we should be very
33:41
sensitive to the social
33:43
upheavals that are going to happen
33:45
due to AI. So we're talking about things like it
33:47
decided to kill us all, but a more mundane issue
33:50
is it's going to put a lot of people out
33:52
of work. A lot. And
33:55
it's funny because other technological advances put laborers
33:57
out of work. This is going to put,
34:00
I don't know, podcasters, professors that
34:02
are working. I
34:05
feel sometimes, you know,
34:07
there's a concrete answer. We're coming up
34:09
to an election season. I don't
34:11
think politicians on the debate, doing
34:13
their debates, are going to talk enough about AI. I
34:17
think they're going to talk a lot about cultural war
34:19
issues, they're going to talk about foreign policy, they're going
34:21
to talk about budgets. But AI,
34:23
we should treat it as important as it is.
34:25
It's very important, and we should treat it as
34:27
such. Well, you
34:30
relieved some of my anxiety and increased
34:32
some of it. Well you
34:34
terrified me with the virus slash AI scenario, which
34:36
is going to keep me from sleeping for a
34:39
while. We've reached a point
34:41
where we always ask seven quick
34:43
questions at the end of a
34:45
show. And you've been
34:47
on the show before, and you were very good
34:49
natured in that, to convince me that I was
34:52
wrong. Well, I'm
34:54
not sure I convinced you, but we had a good
34:56
conversation. Yeah. So maybe you've
34:58
changed your mind about some of the answers to these
35:00
seven questions. Let's see. Of
35:03
all the things there are to understand,
35:05
what do you wish you really understood?
35:07
Consciousness, the mind. How
35:11
do you tell someone to have their facts wrong?
35:13
Yeah. Well, the way I
35:16
used to do it when I was younger was
35:18
be I'd say, you have your facts wrong. And
35:21
that never worked at all. Now
35:25
I often don't tell them, or I
35:27
often just ask them questions. And
35:30
either, if you ask them the right sort
35:33
of question, either it'll come to realize their facts are wrong,
35:35
or I'll come to realize maybe their facts were right and
35:37
I was just wrong myself. What's
35:39
the strangest question anyone has ever asked
35:41
you? Oh, God. You
35:45
could have sent these in advance. I
35:48
was once on a radio show when promoting a
35:51
book. And I said,
35:53
welcome, Professor Bloom. I said, thank you. I'm really
35:55
glad to be on. He says, have you accepted
35:57
Jesus Christ as your Lord and Savior? Oh, wow.
36:00
And it turned out it was a religious show, and
36:02
he always began with that. And
36:04
he was, to be fair, he was in
36:06
entirely good nature to say, no, I'm Jewish,
36:08
and actually I'm an atheist. He was totally
36:10
fine with that. But that question so shook
36:12
me up, I just kind of stumbled from
36:14
next wall. How do
36:16
you deal with the compulsive talker? In
36:21
the short run, I listen.
36:25
I don't mind listening. Sometimes I
36:27
could really spend a lot of time just, you know, I
36:29
talk a lot now because you're talking here and you're asking
36:31
me things. But I often tend to be, if you're one
36:33
or the other, I tend to be more of a listener.
36:36
And I like listening and so on. I
36:39
think the kind of person you're imagining, and I
36:42
do know some people like this, maybe aren't
36:45
very interesting and just love to talk. And I
36:47
listen, but then I don't see them again. Okay.
36:52
Let's say you're sitting at a dinner table
36:54
next to someone you've never met before. How
36:56
do you begin a really
36:59
genuine conversation? Oh,
37:01
God. I sometimes, at
37:04
my best moments, I think, ask
37:06
them a general philosophical
37:09
question. Like,
37:12
if you could live 10,000 years, do
37:14
you think you'd be bored? If
37:17
someone offered you that, would you say no? Because
37:20
the boredom might be incredible. Or
37:22
do you think you'd always be interested? That
37:25
sort of thing. I'd
37:27
like to hear the answer to that. That's good. That's
37:31
one of the things that I find about
37:33
that situation that's kind of important. If
37:37
I ask them a question, and a bell goes off
37:39
in my head when I hear their answer that says,
37:41
the bell says, you
37:43
have no interest in what that
37:46
person just said, then I'm stuck.
37:48
Then you're stuck. I can't
37:50
say, oh, great. Tell me more. Which
37:53
is what I should say. So,
37:59
next to last. What
38:01
gives you confidence? I
38:04
think like a lot of academics I'm a certain sort
38:06
of combination of
38:08
extrovert and introvert. I'm fine talking
38:10
in front of big crowds but
38:14
in smaller situations I prefer to
38:16
talk one-on-one and the truth
38:18
is talking to people I don't know well I'm often
38:20
doesn't give me confidence I often feel feel
38:22
shy but I
38:24
have I'm lucky enough to have a series
38:27
of very close relationships to my to my
38:29
wife to my to my sons my adult
38:31
sons to some friends and that
38:33
gives me confidence I feel I feel really good about
38:35
myself when I talk to the people who love me
38:37
and I love and they've
38:40
chosen they chose they choose to be with
38:42
me and I feel great about that and
38:45
it just it just revs me up. Okay
38:47
the last question what
38:50
book changed your life? I
38:54
can actually answer that and and
38:57
I got answer with two. One
39:00
is Victor Frankel's book Man's Search for Meaning
39:03
which was which is a book of
39:05
describing his experience in concentration camps and
39:08
how he learned that what
39:10
the kind of people who survive it people who have
39:13
meaning in their lives the idea of
39:15
a goal a purpose relationships
39:18
work is is transcendently
39:20
important for people that had
39:22
a huge influence and then the other book that
39:24
did was Mihaly Csikszentmihalyi's
39:27
book Flow which I
39:29
read a long time ago and it
39:31
was all about flow experiences and Csikszentmihalyi
39:33
says look what people people think they
39:35
like lying on the beach and and
39:38
you know hanging out watching TV but
39:40
what really gives lasting pleasure is getting
39:42
into an activity that kind of engages
39:44
you you lose time if you're
39:48
focused for me often it's writing
39:50
sometimes it's reading sometimes it's the right sort
39:52
of conversation and the book gave
39:55
me this insight saying yeah that's what I
39:57
like I thought I
39:59
liked other thing. But no,
40:01
I like these flow experiences and that that
40:03
had a huge role for me. Well,
40:06
this conversation has been that for me. Each
40:10
time we talk, I
40:13
get into that zone where I
40:15
just like it to go on for longer. But
40:18
we do have to end. And I'm so
40:21
grateful you took the time to be
40:23
with me today. This has been a delight.
40:25
Let's not wait four years for next time.
40:28
Okay, good. Thank you,
40:30
Paul. This
40:38
has been clear and vivid. At least I hope
40:40
so. My thanks to
40:42
the sponsor of this podcast and to all
40:44
of you who support our show on Patreon.
40:47
You keep clear and vivid up and running.
40:50
And after we pay expenses, whatever is
40:52
left over goes to the Aldous Center
40:54
for Communicating Science at Stony Brook University.
40:57
So your support is contributing to the
40:59
better communication of science. We're
41:01
very grateful. Paul
41:04
Bloom is professor of psychology at
41:06
the University of Toronto and
41:09
professor emeritus of psychology at
41:11
Yale University. He
41:13
studies how we make sense of
41:15
the world, focusing on pleasure, morality,
41:17
religion, fiction, and art. He's
41:20
written seven books. The latest,
41:22
the one we talked about is Psych,
41:24
the Story of the Human Mind. His
41:27
website is paulbloom.net, where
41:30
you'll find links to his many entertaining
41:32
TED talks. This
41:35
episode was edited and produced by
41:38
our executive producer Graham Ched, with
41:40
help from our associate producer Gene Chumet.
41:44
Our publicist is Sarah Hill. Our
41:46
researcher is Elizabeth Ohini, and
41:49
the sound engineer is Erica Hwang. The
41:52
music is courtesy of the Stefan-Kernig
41:54
Trio. Next
42:04
in our series of conversations, I talk
42:06
with Tom Hanks about a fascinating novel
42:09
he's just written called The
42:11
Making of Another Major Motion Picture
42:13
Masterpiece. Tom has
42:15
acted in about a hundred movies and
42:18
we had a fun time sharing stories
42:20
about the elaborately strange experience of
42:22
taking a movie from the page to
42:25
the theater. I have made
42:27
movies in which literally a crew, almost
42:30
like the circus, you
42:32
know, there's trucks and RVs and
42:34
tents, we drop into a town.
42:38
Sometimes the town is Evansville, Indiana, or
42:40
sometimes a town is Darnstadt, Germany, or
42:42
sometimes a town is Seattle
42:45
or Baton Rouge. And
42:48
we're there for three months and
42:50
the town becomes something of our
42:52
own and everybody recognizes, oh, you're
42:54
with the picture. Oh, yeah,
42:57
yeah, we're with the movie. Oh, good to have you
42:59
here. And that
43:01
circus-like atmosphere governs the
43:03
pace of the day and it is exciting.
43:06
But it's also incredibly challenging. There are
43:08
times where everything works and there are
43:10
times where absolutely nothing works whatsoever. And
43:13
you have a 10-week, 12-week experience that is
43:15
unlike any other and then it's all over
43:17
in the wink of an eye and
43:20
you're gone and you can hardly remember the names of
43:23
the people that you worked with. Tom
43:26
Hanks, next time on Clear
43:28
and Vivid. For
43:30
more details about Clear and Vivid and to
43:32
sign up for my newsletter, please
43:34
visit alanalda.com. And
43:37
you can also find us on Facebook
43:39
and Instagram at Clear and Vivid. Thanks
43:42
for listening. Radio
43:53
Andy. Hey,
43:56
it's Andy Cohen. Join
43:59
me on... Andy Cohen live where it's
44:01
just you me and some of the
44:03
world's biggest celebrities Paris Hilton Chelsea and
44:06
their death Rogan I love you Miley.
44:08
Thank you so much You can listen
44:10
to Andy Cohen live at home or
44:12
anywhere you are no car required download
44:14
the Sirius XM app for over 425
44:18
channels of ad-free music sports entertainment
44:20
and more subscribe now and get
44:23
three months free offer details apply
44:25
The The
44:28
Angie's List you know and trust is
44:30
now Angie, and we're so much more
44:32
than just a list. We still connect
44:34
you with top local pros and show
44:36
you ratings and reviews, but now we
44:39
also let you compare upfront prices on
44:41
hundreds of of projects and book a
44:43
service instantly. We can even handle the
44:45
rest of your project from start to finish. So
44:48
remember, Angie's List is now Angie, and
44:50
we're here to get your job done
44:52
right. Get started at angie.com.
44:54
Calm. That's A-N-G-I, or
44:56
download the app today.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More