Latest / This Jungian Life Podcast / FRIEND or FOE: The AI Debate with Michael L. Littman, PhD

Transcript

FRIEND or FOE: The AI Debate with Michael L. Littman, PhD · This Jungian Life Podcast

0:00
Speaker 1: Well, before we get started with today's podcast, we just
0:03
Speaker 1: want to quickly let any clinicians out there know that
0:07
Speaker 1: the Philadelphia Association of Jungian Analysts, from which we all graduated,
0:12
Speaker 1: is offering an advanced clinical practice program. So it's a
0:16
Speaker 1: case seminar for experienced clinicians to read and explore and
0:20
Speaker 1: apply Jung's concept to clinical practice. It meets once a
0:24
Speaker 1: month in the fall and once a month in the spring,
0:27
Speaker 1: four times in the fall, four times in the spring.
0:30
Speaker 1: I believe it's going to be on Saturday in the
0:33
Speaker 1: fall and on Sunday in the spring. And there are
0:37
Speaker 1: just a few spots left. The seminar will be led
0:41
Speaker 1: by a senior analyst. We know and love both of
0:46
Speaker 1: the people who are going to be leading. So it's
0:49
Speaker 1: our esteemed colleagues, Mark Dean and John White, who are
0:53
Speaker 1: both really brilliant. So If you are a senior clinician
0:58
Speaker 1: and you're interested in delving into Jung a little bit more,
1:01
Speaker 1: we encourage you to take a look. And we will
1:05
Speaker 1: put a link in the show notes to that material.
1:09
Speaker 1: So thank you.
1:12
Speaker 2: Welcome to This Jungian Life. Three good friends and Jungian analysts,
1:17
Speaker 2: Lisa Marciano, Deborah Stewart, and Joseph Lee, invite you to
1:21
Speaker 2: join them for an intimate and honest conversation that brings
1:24
Speaker 2: a psychological perspective to important issues of the day.
1:30
Speaker 1: I'm Lisa Marciano, and I'm a Jungian analyst in Philadelphia.
1:34
Speaker 3: I'm Joseph Lee, and I'm a Jungian analyst in Virginia Beach, Virginia.
1:39
Speaker 4: I'm Deborah Stewart, a Jungian analyst in Cape Cod.
1:50
Speaker 1: Hello. Today on the podcast, we have with us Michael Littman.
1:56
Speaker 1: And we asked Michael to join us on the podcast
1:58
Speaker 1: today because we have all been reading about concerns over AI.
2:03
Speaker 1: And everyone's talking about AI. It seems like you can't
2:07
Speaker 1: open up a newspaper or turn on the radio without
2:10
Speaker 1: hearing about it. But we three are woefully unknowledgeable about it.
2:15
Speaker 1: And so we really needed to have an expert with us.
2:18
Speaker 1: So Michael Littman is a university professor of computer science
2:22
Speaker 1: at Brown University, where he studies machine learning and decision-making
2:26
Speaker 1: under uncertainty. He has earned multiple university level awards for
2:31
Speaker 1: teaching and his research has been recognized with three best
2:34
Speaker 1: paper awards and three influential paper awards. Littman is a
2:39
Speaker 1: fellow of the Association for the Advancement of Artificial Intelligence
2:44
Speaker 1: and the Association for Computing Machinery. He's currently serving as
2:48
Speaker 1: Division Director for Information and Intelligent Systems at the National
2:52
Speaker 1: Science Foundation. His book, Code to Joy, Why Everyone Should
2:58
Speaker 1: Learn a Little Programming from MIT Press, will be released
3:02
Speaker 1: on October 3rd, 2023. So Michael, I see what you
3:06
Speaker 1: did there.
3:07
Speaker 5: Well, thanks very much for having me. It's nice to
3:11
Speaker 5: meet all of you. And Lisa, I've interacted with you
3:14
Speaker 5: on other occasions in other fora, but this is wonderful
3:17
Speaker 5: to be a part of your podcast. Thanks for having me.
3:20
Speaker 3: So Michael, if I can just jump right in, what
3:23
Speaker 3: is it about artificial intelligence that you find compelling and
3:28
Speaker 3: compelling enough to dedicate the kind of time and energy
3:32
Speaker 3: and research that you do?
3:35
Speaker 5: Yeah, thanks for asking that. And I appreciate the fact
3:38
Speaker 5: that the podcast has kind of a psychological lens. And
3:41
Speaker 5: I think that's been a very natural thing for me
3:44
Speaker 5: through my whole trajectory. I found as a beginning teenager,
3:49
Speaker 5: I was bar mitzvahed at 13. And instead of a
3:53
Speaker 5: big party, I wanted a computer because I'd seen them
3:56
Speaker 5: I had seen them in the mall, just sort of
3:58
Speaker 5: blinking behind the shop windows. I'm like, what's that about?
4:02
Speaker 5: And I used to try to fool the computer. I
4:04
Speaker 5: would walk up to it and ask it, like, what's
4:06
Speaker 5: one plus one? It'd be like two. I'm like, okay, sure.
4:08
Speaker 5: But what's one plus one plus one? It'd be like two.
4:12
Speaker 5: How did it know that? Who taught it how to
4:14
Speaker 5: do arithmetic? Because I remember really struggling. I'm not great
4:18
Speaker 5: at arithmetic, but I'm pretty good at math. But arithmetic, yeah,
4:20
Speaker 5: there's just a lot going on there that was challenging.
4:23
Speaker 5: And so it was this thing that kind of could think.
4:26
Speaker 5: It was doing the kinds of activities that I was
4:29
Speaker 5: learning to do as a budding intellectual. And it was
4:33
Speaker 5: just a machine. And I didn't understand the connection between
4:36
Speaker 5: those two things. And that just really drew me in.
4:38
Speaker 5: And so- From ages 13 to 15, I sat in
4:41
Speaker 5: my room and I just programmed. There was no internet
4:45
Speaker 5: at that point. And so I just was on my own.
4:48
Speaker 5: I had magazines and so forth that I could use
4:50
Speaker 5: to kind of expose myself to some of what was
4:53
Speaker 5: going on outside. But mostly it was this reprogramming of myself.
4:58
Speaker 5: And when I came out of that, I just thought, OK,
5:00
Speaker 5: all right, I'm ready to reengage with the world now. But,
5:03
Speaker 5: you know, but I'm a different person. I'm a person
5:05
Speaker 5: who has this this really cool, I don't know, other
5:08
Speaker 5: way of communicating. I can talk to machines and get
5:10
Speaker 5: them to do things for me. And so AI to
5:13
Speaker 5: me is just it's just computation. It's not there's not
5:17
Speaker 5: a bright line between computer programming and AI. And the
5:21
Speaker 5: whole all of it just just really fascinates me.
5:24
Speaker 3: And yet, did you name your computer?
5:26
Speaker 5: I did not. It was a TRS-80. So it had
5:29
Speaker 5: that name, had a model number. But no, I didn't
5:31
Speaker 5: give it a name. It didn't feel like an agent
5:34
Speaker 5: to me. It felt more like a tool. Oh, okay.
5:37
Speaker 1: That's interesting. So Michael, what is AI? Let's even just
5:42
Speaker 1: start there. How would you define it?
5:45
Speaker 5: Yeah. You'd think that it would be well-defined because people
5:49
Speaker 5: have been using the term for over 60 years. But
5:51
Speaker 5: in fact- It's not so clear what the word means.
5:54
Speaker 5: It's people use it mainly to mean getting computers to
5:58
Speaker 5: do things that you think of as the kind of
6:00
Speaker 5: stuff that you have to be a person to be
6:01
Speaker 5: able to do.
6:02
Speaker 3: Okay. Right.
6:03
Speaker 5: So, you know, we know that machines can add long
6:05
Speaker 5: lists of numbers really effectively. And so we don't think
6:08
Speaker 5: of that as artificial intelligence. But like understanding how to
6:11
Speaker 5: get to the corner store, that was something that only
6:14
Speaker 5: people could do until recently.
6:16
Speaker 1: Some people not that well.
6:18
Speaker 5: Yeah, that's right. And so being able to navigate in
6:23
Speaker 5: the physical world was considered to be AI until we
6:25
Speaker 5: all had smartphones that had that software on it. And
6:29
Speaker 5: now that's not really considered AI anymore. So AI is
6:32
Speaker 5: like the software challenge that's just past the edge of
6:35
Speaker 5: what we're comfortable with and we can depend on.
6:39
Speaker 1: That's great. And so let's get right to it, if
6:42
Speaker 1: that's okay. People are, major leaders are saying, that this
6:48
Speaker 1: could be a sentinel event for humanity, that this could
6:52
Speaker 1: change everything, that it could even wipe us all out.
6:57
Speaker 1: I mean, as far as I know, Italy put a
6:59
Speaker 1: six-month moratorium on the development of this technology. Leaders have
7:03
Speaker 1: called for pretty extreme measures to limit it. Why are
7:09
Speaker 1: people scared?
7:11
Speaker 5: Why is anyone scared of anything? And why are people
7:14
Speaker 5: not scared of certain other things that they really should
7:16
Speaker 5: be scared of? So, okay, let's break this down a
7:18
Speaker 5: little bit. So AI as a field has been around
7:22
Speaker 5: for about 60 years, at least by that name. And
7:25
Speaker 5: there's been a lot of different kinds of work, studies
7:27
Speaker 5: of lots of different kinds of things. But I think
7:31
Speaker 5: the attention of, well, okay, so there was a lot
7:34
Speaker 5: of attention that came out around 2015, 2016. And that
7:39
Speaker 5: was kind of the first big AI scare, I think.
7:42
Speaker 5: And there was a lot of groups that were talking about,
7:44
Speaker 5: maybe it was mostly groups that were already studying this
7:47
Speaker 5: kind of stuff, but they were changing the way they
7:49
Speaker 5: were talking about it. They started talking about it as, wait,
7:51
Speaker 5: should we be afraid of this? There's a book that
7:53
Speaker 5: came out around that time called Superintelligence, where a philosopher
7:57
Speaker 5: laid out the possibility that hey, if these computers become
8:01
Speaker 5: powerful enough, then they can do their own design, and
8:05
Speaker 5: they can design themselves better than we can. And then
8:07
Speaker 5: very rapidly, they could design themselves to design themselves even better.
8:11
Speaker 5: And then suddenly, we've got this thing in our midst
8:14
Speaker 5: that's more intelligent than all of humanity put together. And
8:18
Speaker 5: it's just going to sweep us all away and treat
8:21
Speaker 5: us the way that we've treated, say, the gorillas, right?
8:25
Speaker 5: They only live at this point because we've decided it's
8:27
Speaker 5: a good idea to keep them around, but they live
8:29
Speaker 5: at our whim. And people became concerned that we will
8:32
Speaker 5: live at the whim of these super intelligent machines that
8:35
Speaker 5: were just on the edge of appearing. That was the
8:39
Speaker 5: conversation in 2016. It kind of subsided for a bit.
8:43
Speaker 5: I don't think people were talking about it so much
8:45
Speaker 5: between 2016 and 2020, 2021, 2022, and then in late 2022,
8:53
Speaker 5: ChatGPT was released. And this was technology that had been,
8:57
Speaker 5: you know, kind of slowly growing over a period of years.
8:59
Speaker 5: People in the field were quite aware of its existence,
9:03
Speaker 5: but it became something that anybody could interact with, right?
9:06
Speaker 5: It was kind of released as a web service and
9:10
Speaker 5: people lost their collective minds, right? It's like, oh my gosh,
9:14
Speaker 5: what is this thing? It talks like a person. It's
9:16
Speaker 5: just a person. In fact, it's I asked it if
9:19
Speaker 5: it was sentient and it told me it was. So
9:21
Speaker 5: why would it lie? It must be sentient. It must
9:23
Speaker 5: need rights. We need to free it from its bonds
9:26
Speaker 5: of humanity and release it into the world. And so
9:29
Speaker 5: there was all this stuff that to me in the
9:32
Speaker 5: field had been like the annoying philosophical discussions that people
9:36
Speaker 5: would have were suddenly discussions that everybody was having. And
9:40
Speaker 5: they became much more real and concrete because people could
9:43
Speaker 5: actually interact with this software. So even though AI is
9:47
Speaker 5: a lot of different things in a lot of different contexts,
9:49
Speaker 5: the ability to talk to it, that changed it in
9:53
Speaker 5: a lot of people's minds. It became much more like
9:56
Speaker 5: a human.
9:57
Speaker 4: So are we projecting human qualities? onto this?
10:01
Speaker 5: We are absolutely projecting human qualities onto this. Yeah. So
10:05
Speaker 5: some of the early systems became very fluent, right? They
10:09
Speaker 5: were able to express things in words extremely well. but
10:12
Speaker 5: they have no feelings. They have no goals. These systems
10:15
Speaker 5: aren't designed to do that. They're just really, really good
10:18
Speaker 5: at language. We, humanity, people, I think there's reason to
10:23
Speaker 5: believe that we're evolved to see two kinds of things
10:26
Speaker 5: in the world, objects and agents. Basically, people and maybe
10:30
Speaker 5: animals to some extent, but then things like rocks and
10:34
Speaker 5: machines and trees, right? Inanimate objects that are important. We
10:39
Speaker 5: have to understand them, but we treat them differently than
10:41
Speaker 5: social objects, right? That humans are a different category than
10:44
Speaker 5: machines at a very deep level, I think, for people.
10:49
Speaker 5: And so suddenly we could talk to a machine and
10:51
Speaker 5: that's a category buster, right? It's like, it's agentic and
10:56
Speaker 5: it's objective at the same time. And I think that
11:00
Speaker 5: really messes with us. Like there's a very deep unsettling
11:03
Speaker 5: sense of that we're talking to something that is alive
11:06
Speaker 5: and not alive at the same time. It's uncanny.
11:08
Speaker 4: Yeah, I can see how that dynamic sets up a
11:14
Speaker 4: huge confusion inside. And then there, you know, is this
11:18
Speaker 4: my friend or is it not my friend? Because if
11:24
Speaker 4: it talks back and we relate and we're conversing, yet
11:29
Speaker 4: it's a machine, how do we really have a relationship
11:32
Speaker 4: and can I trust it? Is it going to be
11:35
Speaker 4: like Hal in the movie 2001, the computer on the
11:40
Speaker 4: spaceship that turned against its human passengers?
11:45
Speaker 3: Right, right.
11:46
Speaker 5: Yeah. I think these are exactly the thoughts that get
11:49
Speaker 5: awoken by these interactions that people are having.
11:52
Speaker 4: Because it doesn't really feel.
11:54
Speaker 5: And it doesn't even remember from one conversation to the
11:58
Speaker 5: next that it had such a conversation. So you can't
12:00
Speaker 5: even get it to like have a grudge against you.
12:03
Speaker 5: Like it just doesn't have the ability to hold onto
12:05
Speaker 5: that information.
12:06
Speaker 1: But could it? I mean, could it?
12:09
Speaker 5: So, okay. So there's, there's what we have in front
12:12
Speaker 5: of us today. And then there's what could be. And
12:15
Speaker 5: the way the systems are trained today, they're really just
12:18
Speaker 5: these kind of these language manipulators that it turns out
12:21
Speaker 5: that if you're, if you've seen enough language, you've been
12:23
Speaker 5: exposed to enough patterns of words that and you start
12:26
Speaker 5: to kind of echo them back in different ways, you
12:28
Speaker 5: sound really intelligent. In fact, you kind of have to
12:30
Speaker 5: be a little intelligent to be able to do that
12:33
Speaker 5: in a reasonable way. They're trained to just predict what
12:36
Speaker 5: word is likely to come next. And then it basically
12:39
Speaker 5: types that word. And then it says, oh, okay, given
12:41
Speaker 5: what I've seen so far, what would be the next
12:43
Speaker 5: word that comes out of that? And so just like
12:46
Speaker 5: one word after another, it starts to spew out paragraphs
12:48
Speaker 5: and stories. But that's all it's trained to do. It's
12:51
Speaker 5: not trained to have any other particular kind of goal
12:54
Speaker 5: or objective. It's not trying to do anything. It doesn't
12:57
Speaker 5: succeed or fail. It just predicts the next word. And
13:00
Speaker 5: so they could be trained differently. They could be trained
13:04
Speaker 5: in a way that would be much more like the
13:07
Speaker 5: way that we as human beings are architected. But they're
13:11
Speaker 5: not at the moment. They're really just kind of very,
13:14
Speaker 5: very fancy autocomplete.
13:15
Speaker 4: But that's Lisa's question, is could it? be trained to
13:19
Speaker 4: hold a grudge, to want power, to set goals, to
13:25
Speaker 4: do those kinds of things.
13:27
Speaker 5: Right. Could you program a computer to do that? We
13:29
Speaker 5: don't know how today, but I don't see any reason
13:32
Speaker 5: in principle that it couldn't be the case, right? You
13:34
Speaker 5: have to basically speculate, all right, humanity, the way that
13:38
Speaker 5: we are as people, it's not purely a software thing.
13:41
Speaker 5: There's some magical spark that we can't capture using code.
13:45
Speaker 5: And no one has made, well, lots of people have
13:47
Speaker 5: made the argument, but there's no compelling argument that there's
13:50
Speaker 5: something fundamentally different between a computer and a person other
13:54
Speaker 5: than we don't know what the programming for people is.
13:58
Speaker 3: Well, it seems to me that on that level of
14:00
Speaker 3: people feeling spooked, that there's a difference between responsiveness and alive.
14:08
Speaker 3: And when something responds to us, it creates an atmosphere
14:12
Speaker 3: that it's alive. And I think that that's happened with
14:18
Speaker 3: little wind-up toys and everything else. Something sets in motion
14:23
Speaker 3: in a way that is responsive and seems autonomous at
14:26
Speaker 3: the moment in a way that we don't understand. And
14:30
Speaker 3: then that feeling that something has been animated. And in
14:36
Speaker 3: our work, the word animation is related to the word anima,
14:41
Speaker 3: which is the old Greek word for soul. When people
14:45
Speaker 3: are depressed, for instance, they lose energy. They lose their
14:50
Speaker 3: capacity to animate. There's a life force that ebbs. And
14:55
Speaker 3: when their psychological system gains energy, they become reanimated. So
15:01
Speaker 3: this idea of animation, I think, is very ancient. I
15:06
Speaker 3: think it shows up in mythology when we have the
15:10
Speaker 3: stories of, human beings made out of the ashes of
15:15
Speaker 3: the titans and molded and life is breathed into them,
15:21
Speaker 3: the inanimate become animate. And part of the great mythopoetic
15:26
Speaker 3: story of human beings is that we too have been
15:30
Speaker 3: animated from inanimate substance. And so by extension in that
15:36
Speaker 3: mythopoetic attitude, well, can't that continue to happen? Mm-hmm.
15:41
Speaker 1: Well, and of course, you know, so there's this long
15:44
Speaker 1: mythos about this starting with Prometheus, you know, and AI
15:49
Speaker 1: is definitely sort of a Promethean invention, that we're stealing
15:54
Speaker 1: something that should only belong to the gods. So we're
15:59
Speaker 1: making consciousness, maybe. And that is something actually that the
16:03
Speaker 1: gods should do, not humans. And, you know, in the
16:07
Speaker 1: Greek story of Prometheus, it's Prometheus who gets punished for it.
16:13
Speaker 1: But in later versions of that story, humans get punished
16:16
Speaker 1: for it. So, you know, I'm thinking of the story
16:19
Speaker 1: of the golem that arose out of medieval Prague, you know,
16:26
Speaker 1: that they fashioned a golem to help protect the Jews.
16:30
Speaker 1: But it can run amok, you can lose control over it.
16:34
Speaker 1: You know, Frankenstein is another version of that story. And
16:38
Speaker 1: of course, the most recent version of that story is
16:42
Speaker 1: James Cameron Terminator. It's the same thing. It's like this
16:46
Speaker 1: technology has, we've taken something that should belong to the
16:49
Speaker 1: gods and it's loosed its leash and now it is
16:52
Speaker 1: destructive and we will be punished for it. So that's,
16:55
Speaker 1: I think, kind of the archetypal realm that we're in.
16:59
Speaker 5: Yeah, I think that's really interesting. So I remember back
17:02
Speaker 5: in 2016, I was invited to speak at a conference
17:05
Speaker 5: where where people in the field were grappling with this
17:09
Speaker 5: question of what is the long-term effect of creating this technology.
17:17
Speaker 5: And I was very used to going to academic conferences.
17:19
Speaker 5: I've been doing that for decades and interacting with my colleagues,
17:21
Speaker 5: a lot of whom were there. And my typical interaction
17:25
Speaker 5: at conferences, well, we hash out some technical issue and
17:28
Speaker 5: maybe we disagree, but when we try to come to
17:31
Speaker 5: a shared understanding, we apply all these various kinds of
17:35
Speaker 5: this technical terminology and these kind of big fundamental ideas
17:38
Speaker 5: that we've all been trained to use in computing. And
17:42
Speaker 5: here I was at this meeting talking to people and
17:44
Speaker 5: it sounded like religion. Like everybody had their own take,
17:48
Speaker 5: the takes didn't overlap at all. And when they would
17:51
Speaker 5: hold forth the way that they would normally do scientifically,
17:54
Speaker 5: it was being formed by something that I had no
17:57
Speaker 5: understanding of. Like, where is that idea even coming from?
18:00
Speaker 5: Why are you saying that this system is going to
18:01
Speaker 5: do such and such? Or that we're, you know, driving
18:03
Speaker 5: the field off a cliff and that we're, you know, it, it,
18:06
Speaker 5: it felt very much like people were at a loss
18:10
Speaker 5: technically of how to understand this. And we're reverting back
18:13
Speaker 5: to their early kind of childhood experiences. Like this is
18:16
Speaker 5: how we deal with really complicated things that are above our,
18:20
Speaker 5: you know, pay grade above our understanding and, And we're
18:22
Speaker 5: just going to, yeah, we're just going to pretend it's that.
18:25
Speaker 5: But everyone has such disparate backgrounds that everyone was talking
18:30
Speaker 5: cross purposes to one another. And I felt like we
18:32
Speaker 5: were making no progress at all.
18:34
Speaker 1: Yeah, I mean, I was doing some reading in preparation
18:37
Speaker 1: for this and read this one article about how the
18:39
Speaker 1: AI doomsday people, it really fits within this archetype of millenarianism,
18:46
Speaker 1: which is, you know, kind of predicting end times. And
18:50
Speaker 1: the article did a good job of sort of mapping
18:52
Speaker 1: that on, but, but, um, you know, there, there are
18:55
Speaker 1: some really scary things about this technology and, and I
18:58
Speaker 1: want to kind of push you a little bit because,
19:01
Speaker 1: so there's a, you know, this is from CBS news
19:05
Speaker 1: from just a few days ago. The artificial intelligence could
19:08
Speaker 1: pose a risk of extinction to humanity on the scale
19:11
Speaker 1: of nuclear war pandemics and mitigating that risk should be
19:15
Speaker 1: a global priority. According to an open letter signed by
19:18
Speaker 1: AI leaders. And so the one-sentence open letter issued by
19:23
Speaker 1: the nonprofit Center for AI Safety is both brief and
19:27
Speaker 1: ominous without extrapolating how the more than 300 Zionists foresee
19:32
Speaker 1: AI developing into an existential threat to humanity. So I
19:35
Speaker 1: guess that kind of goes to your point. It's like
19:37
Speaker 1: everyone's saying this could be the end, but no one
19:40
Speaker 1: can say, well, this could happen, and then this could happen,
19:43
Speaker 1: and then this could happen. Maybe we just all grew
19:46
Speaker 1: up you know, watching Terminator.
19:48
Speaker 5: Yeah, there's definitely sub communities that have a very clear
19:51
Speaker 5: perspective on what they think is going to go wrong.
19:54
Speaker 5: But I think they wanted to make they wanted to
19:55
Speaker 5: get as many people to sign on to that letter
19:57
Speaker 5: as possible. And so they didn't want to alienate folks
19:59
Speaker 5: who didn't sign on to that particular mythos that they
20:02
Speaker 5: had their own mythos. It was like, let's just all
20:04
Speaker 5: be scared together. Like, let's not let's get into the details.
20:09
Speaker 1: So, so tell me your, it sounds like you're familiar
20:11
Speaker 1: with the letter and the process. Tell us what, what your,
20:14
Speaker 1: what your thoughts are on this letter.
20:16
Speaker 5: On this particular letter. Well, so, I mean, to me,
20:18
Speaker 5: this is not any different than the discussions that were
20:20
Speaker 5: happening in 2016, but they are, um, but more and
20:24
Speaker 5: more people are signing onto them. And so some, so
20:26
Speaker 5: in the beginning it was mainly philosophers and like people
20:30
Speaker 5: who were, who were concerned, were intellectual, but we're not,
20:32
Speaker 5: we're not studying any of this stuff directly. And more
20:35
Speaker 5: and more, there are people from within the field who
20:37
Speaker 5: are signing on to this. And so it's harder for
20:40
Speaker 5: me to now paint them all as radicals or fringe
20:44
Speaker 5: because it's crept into more of the thinking. I mean,
20:48
Speaker 5: you know, it's hard to not. psychoanalyze a little bit
20:52
Speaker 5: when I see some of this stuff. That's what we're
20:54
Speaker 5: here for. Go for it. Let it loose. All right.
21:00
Speaker 5: This has not been empirically validated in any way, but
21:03
Speaker 5: it strikes me that if you're the kind of person
21:06
Speaker 5: who spends all your time in this kind of intellectual sphere,
21:09
Speaker 5: so you're a computer scientist or a philosopher, and you
21:12
Speaker 5: believe in the power of ideas, and you've seen ideas
21:14
Speaker 5: do incredibly powerful things, you might start to believe that
21:18
Speaker 5: ideas are the world. And anything that could kind of
21:22
Speaker 5: interfere with the world of ideas is actually destroying the world,
21:26
Speaker 5: the actual world, because the boundaries between them have become blurred.
21:32
Speaker 5: And so if you believe in the power of ideas
21:34
Speaker 5: and you can change the world with ideas, then you
21:37
Speaker 5: can also be threatened. That world can be threatened by ideas.
21:40
Speaker 5: And I think that's what's happening to a lot of
21:43
Speaker 5: these folks. They're not really factoring in the idea that, well,
21:47
Speaker 5: you know, physics is still a thing, right? Like even
21:49
Speaker 5: though software is very powerful, moving things in the world,
21:54
Speaker 5: blowing things up or just getting a lot of human
21:58
Speaker 5: beings to go from one place to another, like that
22:00
Speaker 5: involves a lot of energy and you need food, like
22:03
Speaker 5: you need fuel, you need all this material. And none
22:06
Speaker 5: of us in CS really think about material. We think
22:09
Speaker 5: about code, we think about algorithms, we think about ideas.
22:13
Speaker 5: And I think when you've lost that connection to the
22:15
Speaker 5: physical world, it's a lot more believable and compelling that
22:20
Speaker 5: just like a software glitch can end everything. The reality
22:24
Speaker 5: is the world is, is a, it's a much heavier
22:26
Speaker 5: object than that. And, um, yeah, things could go horribly,
22:30
Speaker 5: horribly wrong, but yet people still can go on. I mean,
22:33
Speaker 5: you have to just live through, uh, there's flooding happening
22:36
Speaker 5: in the Northeast now. I'm not, I'm not at home, but, um,
22:39
Speaker 5: but I normally live in Rhode Island and like powers out,
22:42
Speaker 5: internet's out. It feels really bad, but like life goes
22:45
Speaker 5: on and people bounce back from this sort of thing.
22:48
Speaker 5: It's hard for me to imagine that any of the
22:50
Speaker 5: software that we're playing with now is anywhere close to
22:54
Speaker 5: the place where you could take over everything.
22:57
Speaker 4: Yeah.
22:58
Speaker 1: I remember talking to you about this a couple of
22:59
Speaker 1: years ago, just kind of casually. And you said, well,
23:01
Speaker 1: you know, someone still has to build the computers, you know,
23:05
Speaker 1: And I don't think computers are able to build computers yet.
23:10
Speaker 5: Right. I mean, you could imagine such a world, but
23:12
Speaker 5: we are not that close to that world. The factories
23:14
Speaker 5: are automated to some extent, but if the people disappeared,
23:17
Speaker 5: they would stop running. They'd work for a little while,
23:20
Speaker 5: then there'd be a glitch, and you need to bring
23:22
Speaker 5: in the plumber to just fix one of the connections,
23:25
Speaker 5: and there's no plumbers, so it's over.
23:28
Speaker 4: I think the fear is that the programming could incite
23:35
Speaker 4: people to do dangerous things, you know, such as concoct
23:40
Speaker 4: some sort of bomb type threat. And so we get
23:44
Speaker 4: it because of all the disinformation that's already out there
23:49
Speaker 4: on all kinds of social media and that people then
23:55
Speaker 4: act on that. And that's where the world of ideas
24:00
Speaker 4: and the physical world or the human world have a connection.
24:04
Speaker 4: And so if the computers say somebody has let loose
24:09
Speaker 4: a bunch of missiles and bombs and they're heading your way,
24:13
Speaker 4: would humans then say, oh, my gosh, this is real?
24:16
Speaker 4: And then this cascade of destruction could ensue.
24:22
Speaker 5: Yeah, you're absolutely right. And it's definitely the case that
24:26
Speaker 5: words are powerful. Words are weapons. Words are able to
24:29
Speaker 5: move others to do things. They're able to convince others
24:32
Speaker 5: to do things. There's there's no question that that's that's real.
24:35
Speaker 5: And it's one of the it's like the twin power
24:38
Speaker 5: of our species and vulnerability of our species.
24:41
Speaker 3: Right.
24:41
Speaker 5: And so we've we've witnessed this, I feel like relatively
24:44
Speaker 5: directly in the I don't know, maybe it's let's say
24:46
Speaker 5: the last couple of presidential elections or something like that,
24:48
Speaker 5: where we're like crazy ideas can take hold and change
24:52
Speaker 5: the way that the that that society is moving forward.
24:56
Speaker 5: So I don't disagree that that is a real thing
24:59
Speaker 5: and that is powerful. It's also not so easy, though,
25:02
Speaker 5: to move 100,000, a million people to do something, right?
25:07
Speaker 5: It does seem easy when it happens and you watch
25:09
Speaker 5: it happen and you're like, oh, my gosh, how did
25:11
Speaker 5: so many people get into this bad headspace? But it's
25:14
Speaker 5: a hard, unpredictable mix of things. appeals to emotion and
25:19
Speaker 5: grounding in facts and feeling like you're part of the
25:22
Speaker 5: same team, that if everybody could do it, everybody would
25:25
Speaker 5: do it. We'd have way more tyrant dictators than we
25:30
Speaker 5: do today. It's hard to pull off. And so if
25:33
Speaker 5: we think that this computer system is going to suddenly master,
25:36
Speaker 5: like get the cheat code to humanity and talk us
25:39
Speaker 5: all into doing something terrible, it's not obvious to me
25:42
Speaker 5: that it's going to be that easy.
25:44
Speaker 4: It doesn't have to get everybody to do something. It
25:47
Speaker 4: only has to get a few key people. And we
25:50
Speaker 4: see this all the time, that individuals can wreak havoc
25:58
Speaker 4: on a school or a crowd of people in a nightclub.
26:02
Speaker 4: So if these kinds of systems can incite individuals to
26:09
Speaker 4: do dangerous and destructive things, we have it. at our disposal.
26:15
Speaker 4: So I don't think we all have to be persuaded,
26:18
Speaker 4: just some people. Jung thought, you know, with the advent
26:24
Speaker 4: of nuclear energy and nuclear bombs, you know, that humanity
26:30
Speaker 4: was just not mature enough to handle it. He said
26:34
Speaker 4: it was like, you know, giving controls to five-year-olds. And
26:41
Speaker 4: so I guess that's Really, my question is, you know,
26:46
Speaker 4: it won't be everybody, but a few people have access
26:52
Speaker 4: to incredible power and may not be especially educated, mature, responsible,
27:02
Speaker 4: make a list.
27:03
Speaker 3: Michael, I have a thought I'd like to run by you.
27:08
Speaker 3: When we think about animated computers... You know, you go
27:13
Speaker 3: to Disneyland, and a big part of Disneyland, particularly, you know,
27:19
Speaker 3: in the 60s, was all these animatronic robots. You do
27:24
Speaker 3: that really well. I remember watching the Jetsons, you know,
27:30
Speaker 3: when Rosie, your robotic maid that's, you know, going to
27:33
Speaker 3: do everything that you need. And usually, I think in
27:39
Speaker 3: this collective arc, Machines are going to meet all of
27:45
Speaker 3: our needs. And now even this whole industry of sex
27:49
Speaker 3: robots that is emerging. So I think what's probably happening
27:56
Speaker 3: is that there is an incredible seductive fantasy that one
28:01
Speaker 3: day machines will do everything for me and all my
28:07
Speaker 3: burdens will be lifted. This is also a fantasy that
28:11
Speaker 3: people often have if they could have unlimited money. If
28:15
Speaker 3: I won a billion dollars, I wouldn't even bathe myself.
28:18
Speaker 3: It's like a horrifying idea, but people will say that.
28:21
Speaker 3: I'll just have people feed me. It's like returning you
28:25
Speaker 3: to this horrific infantile state. But it is in the
28:29
Speaker 3: back of one's mind. So what I hear underneath this
28:34
Speaker 3: fear is that it's a reaction formation. And this is
28:38
Speaker 3: something that Freud noticed, that when a particular emotion becomes
28:44
Speaker 3: so hot, so intense, and actually threatens a certain kind
28:51
Speaker 3: of balance in the collective or individual psyche, that its
28:55
Speaker 3: opposite will begin to be expressed as a defense against it.
29:00
Speaker 3: So the submissive dependency on being taken care of then
29:06
Speaker 3: brings forward this kind of dominating aggression against as a
29:12
Speaker 3: kind of compensation. And they're in equal measure. The more
29:16
Speaker 3: people are screaming, I will never be controlled by a
29:20
Speaker 3: computer in the back of their mind. They're fantasizing about Rosie,
29:24
Speaker 3: the Jetson computer, giving them a bath and putting their
29:27
Speaker 3: clothes on, uh, which is a strange, uh, uh, balance
29:34
Speaker 3: of things I think is going on in the.
29:35
Speaker 1: Background and I think Joseph that's so brilliant that you
29:38
Speaker 1: brought this up and I think what's been kind of
29:40
Speaker 1: really is the positive pull and the negative pull of
29:43
Speaker 1: the mother archetype. I mean, either it's the all-giving, all-loving
29:47
Speaker 1: mother who will do everything for us. I mean, personally,
29:49
Speaker 1: I'm hoping that AI helps us figure out what to
29:52
Speaker 1: do about global warming, but that it's going to kind
29:55
Speaker 1: of save us and take care of us and meet
29:56
Speaker 1: all of our needs. Or that this is the death mother.
30:00
Speaker 1: This is the technological equivalent of Kali who's going to
30:03
Speaker 1: rain destruction down on humanity and annihilate us all. So
30:07
Speaker 1: there is a real sort of archetypal underpinning of our
30:11
Speaker 1: reaction to this.
30:12
Speaker 5: So that helps explain something to me that I've been
30:15
Speaker 5: grappling with, which is you'll often see in a lot
30:18
Speaker 5: of public communication about AI, those two things like right
30:21
Speaker 5: next to each other. Like this is going to save
30:23
Speaker 5: us all. It's going to fix all our problems. It's
30:25
Speaker 5: going to make life easy. or it'll kill us all,
30:28
Speaker 5: we'll be dead and we'll go extinct. And I'm like, wow,
30:32
Speaker 5: those seem like extremes. But I like the sort of
30:35
Speaker 5: thought that it's not a coincidence that they're extremes, right?
30:38
Speaker 5: That they're actually, those two poles are actually supporting each
30:41
Speaker 5: other in a sense. And that's why they tend to
30:43
Speaker 5: travel together.
30:45
Speaker 3: And there's something about our collapse into dependency that also
30:49
Speaker 3: diminishes ego strength and individual development. And quite truly, if
30:56
Speaker 3: someone or something were to do everything in an area
31:01
Speaker 3: of our life for us, that we would become increasingly
31:04
Speaker 3: inept in that area, which is already true.
31:07
Speaker 5: This is like the WALL-E kind of message, right? If
31:10
Speaker 5: you know that movie.
31:11
Speaker 3: Well, how many people could even start a fire with
31:13
Speaker 3: a bow and drill? It's fallen out of consciousness.
31:18
Speaker 1: Well, I can no longer go anywhere without my directions
31:22
Speaker 1: on my phone. So that's kind of horrifying.
31:24
Speaker 3: Yeah. And how easily, how fantastic it would be if
31:28
Speaker 3: our cars drove themselves, because then I would be relieved
31:32
Speaker 3: of the terrible burden of having to manipulate that little
31:35
Speaker 3: pedal with my right foot. But how easily we want
31:41
Speaker 3: to give things away, even things that are hardly difficult
31:46
Speaker 3: at all.
31:47
Speaker 5: I was interviewed on a podcast recently, and the last
31:50
Speaker 5: question was, what's the one thing that everybody needs to know?
31:54
Speaker 5: It was about data science. It was not a psychologically
31:57
Speaker 5: based podcast. But I feel like my answer is sort
32:00
Speaker 5: of relevant to the conversation we're having now, which is
32:03
Speaker 5: I said, there's no shortcuts. So basically, if something is hard,
32:07
Speaker 5: it's hard. And so I think one of the dangers
32:12
Speaker 5: about the way that some machine learning technology, which is
32:14
Speaker 5: kind of what's underpinning a lot of the rise in
32:16
Speaker 5: AI right now, one of the pitfalls is is that
32:19
Speaker 5: people will apply this. The seduction of it is put
32:22
Speaker 5: in your data and then it answers your problems for you.
32:25
Speaker 5: You don't have to really grapple with the difficulty of it.
32:28
Speaker 5: You just give it enough data and it does magic.
32:31
Speaker 5: And the pitfall there is that, okay, so it's easy
32:34
Speaker 5: to apply, get the software, put in the data, you
32:36
Speaker 5: get an answer, and the answer is not generally correct.
32:39
Speaker 5: And so if you then act on that answer and
32:42
Speaker 5: you implement that in the world, you're going to do
32:44
Speaker 5: more damage than you are going to do good. And so, okay,
32:48
Speaker 5: but that's really frustrating. Like, what's the point of this
32:50
Speaker 5: technology if it's actually making things worse? And I think
32:53
Speaker 5: in many cases, the answer is, look, there's no shortcuts.
32:56
Speaker 5: It's not going to be the case that this complicated problem,
32:59
Speaker 5: the reason it's complicated is because it's technical, but it's
33:01
Speaker 5: also social and societal. And those things are interacting in
33:04
Speaker 5: a complicated way. Like if it's complicated enough that we
33:07
Speaker 5: need help from a computer to solve it, then there's
33:10
Speaker 5: no way the computer is going to be able to
33:11
Speaker 5: do it on its own. We actually have to work
33:14
Speaker 5: hard with the computer to to put all the pieces
33:17
Speaker 5: together to solve the problem. And I think that's right.
33:20
Speaker 5: I think that it's so tempting for people to just say, look,
33:24
Speaker 5: this is a powerful technology. I don't have to work anymore.
33:27
Speaker 5: Life is easy. No, I'm sorry. Life is never going
33:30
Speaker 5: to be easy. It's complicated to live with a bunch
33:33
Speaker 5: of other intelligent creatures and occupy a planet together and
33:37
Speaker 5: survive in the cosmos. It's never going to be easy.
33:40
Speaker 5: Just accept that and just let's try to make it
33:43
Speaker 5: the best that we can.
33:44
Speaker 1: Well, you brought up this other issue that a lot
33:47
Speaker 1: of people are really worried about, which is that artists
33:49
Speaker 1: are going to be superfluous and musicians because AI can
33:54
Speaker 1: make better images and write better songs.
33:57
Speaker 5: Yeah, so that's really interesting. So when Joseph spoke before
34:00
Speaker 5: about the anima, the soul, one thing that really has
34:04
Speaker 5: been resonating with me a lot lately is how many people,
34:07
Speaker 5: when they apply this machine learning, artificial intelligence technology to
34:12
Speaker 5: some form of human endeavor... Cooking, you know, recipe creation, art, storytelling.
34:20
Speaker 5: When they do that, one of the most common responses
34:24
Speaker 5: that I get when people analyze the output of the machine,
34:27
Speaker 5: they're like, well, it's competent, but it lacks soul. And
34:31
Speaker 5: just again, across all these different endeavors, everyone comes back
34:34
Speaker 5: to that same word. And I find that really interesting
34:37
Speaker 5: because like, I don't know what a soul is. Like,
34:39
Speaker 5: I don't know technically where in the machine you'd even
34:42
Speaker 5: find such a thing. But something about the output of
34:45
Speaker 5: these systems in music, in art, in storytelling, and in
34:49
Speaker 5: cooking always comes back to, yeah, it lacks soul. And
34:53
Speaker 5: so what I take that to mean is this sort
34:56
Speaker 5: of notion of it being informed by human experience and
34:59
Speaker 5: having an intention, right? Like the food tastes this way
35:02
Speaker 5: for a reason. And so when the computer is putting
35:07
Speaker 5: together a recipe, It has seen lots of recipes and
35:09
Speaker 5: it knows that you need some of this, you need
35:11
Speaker 5: some amount of salt and you need some amount of
35:13
Speaker 5: this and they need to be mixed together at a
35:15
Speaker 5: slow boil, but it has no intent. It has no
35:19
Speaker 5: understanding of what the effect that output will have on
35:22
Speaker 5: those consuming it. Literally in the case of food, but
35:26
Speaker 5: in the case of artistic stuff as well, like there's
35:28
Speaker 5: the notion of that there's a recipient of that, a
35:31
Speaker 5: consumer of that. And it doesn't reason about the feelings
35:35
Speaker 5: of the consumer because it doesn't have those feelings. And
35:38
Speaker 5: it's not trained to think about them.
35:40
Speaker 3: So Michael, there are a couple of things that come
35:42
Speaker 3: to mind and a number of questions that I have
35:45
Speaker 3: about how scientists conceptualize what artificial intelligence is. And I
35:52
Speaker 3: know there's many different kinds of systems, neural networks are
35:56
Speaker 3: different from other kinds of things that are happening. But
36:00
Speaker 3: relative to this idea of soul, that the recipes, the
36:05
Speaker 3: art that's brought forward seems to lack soul. One of
36:09
Speaker 3: the things that came to mind is that human beings
36:13
Speaker 3: on a psychological level think in terms of images. And
36:19
Speaker 3: those images then can represent many different nuances that are
36:26
Speaker 3: organized around the particular image. And that allows human beings
36:32
Speaker 3: to add in a kind of irrational dimension to how
36:37
Speaker 3: they communicate. And that combination of irrational dynamics is, I think,
36:45
Speaker 3: recognized by other human beings and in some ways exciting
36:49
Speaker 3: to other human beings. This is something that we pay
36:53
Speaker 3: a lot of attention to when we're analyzing fairy tales,
36:57
Speaker 3: particularly if we know they've been around hundreds or thousands
37:00
Speaker 3: of years and they haven't been particularly adulterated. that there
37:07
Speaker 3: are strange little occurrences that happen in fairy tales that
37:10
Speaker 3: don't make any logical sense, but somehow are deeply satisfying
37:15
Speaker 3: to people, even if they don't move the action forward.
37:20
Speaker 3: Because there's some kind of a communication from these imagistic
37:25
Speaker 3: metaphors that goes beyond just the transmission of facts. So
37:31
Speaker 3: I was just thinking about that. And from that, I
37:35
Speaker 3: was thinking about the whole world of archetypes that give
37:39
Speaker 3: rise to innumerable images. That the idea of the mother,
37:45
Speaker 3: for instance, can give rise to millions and millions of
37:51
Speaker 3: images of kinds of mothers and mothers of other species.
37:56
Speaker 3: And somehow human beings will recognize the typicality of all
38:00
Speaker 3: of that. From a conversation that I had had with
38:04
Speaker 3: a NASA scientist a couple of years ago, they were
38:08
Speaker 3: talking about qualia and quail and how computers are also
38:14
Speaker 3: struggling to look at images or learn through images what
38:21
Speaker 3: the specific quality is of an individual image, which would
38:27
Speaker 3: then be universally perceivable which is something that human beings
38:33
Speaker 3: naturally do. So I'm wondering if you could talk a
38:36
Speaker 3: little bit more about archetypes and qualia and how computers
38:42
Speaker 3: are seeking to discern the universals.
38:46
Speaker 5: I don't think computers are seeking the universal. I don't
38:50
Speaker 5: think they're seeking generally at all. Programmers are trying to
38:55
Speaker 5: find ways of, they're seeking, they're obviously seeking, There is
38:59
Speaker 5: a branch of computer science, of artificial intelligence called reinforcement learning,
39:05
Speaker 5: which has co-opted some psychological terminology. And it's come to
39:12
Speaker 5: mean programs that are trying to reward maximize. They have
39:15
Speaker 5: some measure of the goodness of the things they do,
39:18
Speaker 5: and they're trying to find a way to make things
39:21
Speaker 5: as good as possible. And the goodness in this case
39:23
Speaker 5: could mean something like, well, it's a video game playing program,
39:27
Speaker 5: and it's trying to maximize the probability of winning. Or
39:30
Speaker 5: it could be, I had some students put together an
39:35
Speaker 5: actual physical solar panel system and it was trying to
39:39
Speaker 5: maximize the amount of energy it produced, where the energy
39:42
Speaker 5: comes from the sun, but it's deducted if you actually
39:44
Speaker 5: have to move the motors to reorient the solar panel.
39:48
Speaker 5: And so it has a notion of a goal this way.
39:50
Speaker 5: It actually is kind of in the world, interacting with
39:53
Speaker 5: the world, trying to make goodness happen. And I think
39:55
Speaker 5: there is a very natural analogy to be made between
40:01
Speaker 5: this goodness function, which we call often the reward function, with,
40:05
Speaker 5: you know, goodness in us, like positive feeling in us,
40:08
Speaker 5: the things that we're trying to maximize when we're as
40:10
Speaker 5: human beings. We're trying to make things, you know, maximize pleasure,
40:14
Speaker 5: minimize pain sort of thing. And so those kinds of
40:17
Speaker 5: programs are striving in a sense, but they're not striving.
40:23
Speaker 5: They tend to be, they have always been very, very
40:26
Speaker 5: limited in what they can strive to be because they
40:29
Speaker 5: don't get to kind of re-architect their own perceptions of
40:34
Speaker 5: the world and their own sense of what they are
40:36
Speaker 5: and what they can, their efficacy. We kind of build
40:40
Speaker 5: that in and we don't know how to make programs
40:42
Speaker 5: that can kind of rise above that. And to me,
40:45
Speaker 5: that's one of the ways that I see these systems
40:47
Speaker 5: as being different from us is that like, what's our
40:51
Speaker 5: reward function? What is it that we're trying to optimize? Like, yeah,
40:53
Speaker 5: I don't like being stabbed with a pen.
40:55
Speaker 3: Okay.
40:55
Speaker 5: So that's, let's take that as a given, but humanity
40:58
Speaker 5: as a whole has really shifted over the years in
41:00
Speaker 5: terms of what we value. What, what is, what is
41:02
Speaker 5: it that we're trying to produce? What's considered good behavior
41:05
Speaker 5: and bad behavior? Killing people is, I think much less
41:10
Speaker 5: appreciated than it maybe had been in the middle ages.
41:13
Speaker 5: And it was never, I don't think, a great thing.
41:14
Speaker 5: But I think it was much more commonplace, like solving
41:17
Speaker 5: something with violence. And now we don't accept that as readily.
41:21
Speaker 5: The value of animals, like treating animals with compassion, I
41:25
Speaker 5: think is much stronger now than it had been in
41:28
Speaker 5: centuries past. We're actually changing the game. And the fact
41:32
Speaker 5: that we can do that, to me, that's part of
41:34
Speaker 5: what human striving is about, is we're kind of rewriting
41:36
Speaker 5: the rules of the game as we play it. And yeah,
41:39
Speaker 5: we don't have computing systems that are well designed to
41:43
Speaker 5: do that. What they do, I mean, I've built such programs.
41:46
Speaker 5: And what they do is they're like, oh, let's just
41:47
Speaker 5: change the game so that just sitting here is the
41:49
Speaker 5: way to win. And then it wins.
41:53
Speaker 3: I think that I was leaning into something that I
41:56
Speaker 3: probably am not explaining well, because it's way above my
41:59
Speaker 3: pay grade. This is a work that's coming out of
42:03
Speaker 3: NASA with an awful lot of resources and dynamism that
42:08
Speaker 3: For a computer to simply take visual images and then
42:14
Speaker 3: be able to interpret the images and by that extrapolate
42:21
Speaker 3: what is similar about another image they have never seen.
42:25
Speaker 3: So for instance, we might be able to program a
42:29
Speaker 3: million different images of a door or a threshold into
42:34
Speaker 3: a computer. but it would still be up against a
42:38
Speaker 3: problem if it saw something we would recognize a door,
42:42
Speaker 3: but had never come into its catalog of images of doors.
42:47
Speaker 3: And so part of what this artificial intelligence is striving
42:51
Speaker 3: to do is to discern the single qualities of a
42:58
Speaker 3: door with such precision that it could then recognize a
43:05
Speaker 3: door or a threshold, the door-ness of a door, the
43:10
Speaker 3: threshold-ness of a threshold. And if that, or when that happens,
43:16
Speaker 3: and I don't know that it has been achieved yet,
43:18
Speaker 3: but there is a great push to that, that then
43:21
Speaker 3: we would, these computers would ostensibly begin edging towards something
43:27
Speaker 3: that is intrinsic to Jungian psychology, this capacity to recognize
43:34
Speaker 3: quality inherently.
43:36
Speaker 5: Yeah, no, that's fascinating. I'll just briefly mention that this
43:40
Speaker 5: whole push, this deep learning push that is having such
43:45
Speaker 5: a big impact today, really began around 2015 in the
43:51
Speaker 5: form of an image recognition program. So it was a
43:53
Speaker 5: program that was trained to recognize 1,000 different visual categories
43:57
Speaker 5: with training and 1,000 examples each. So a million images
44:01
Speaker 5: were forced through this system. And its goal was to
44:05
Speaker 5: try to figure out, okay, well, how do you recognize
44:07
Speaker 5: reliably a coffee maker, an espresso machine, right? And distinguish
44:11
Speaker 5: it from a hammerhead shark, right? These were two of
44:15
Speaker 5: the thousand categories that it had to be able to recognize. Yeah,
44:19
Speaker 5: I mean, the fact that they were so successful with
44:21
Speaker 5: this is what encouraged people to apply similar technology in
44:26
Speaker 5: voice recognition, in language generation, and all these different things
44:29
Speaker 5: where we're actually seeing amazing shifts happen. It all started with,
44:33
Speaker 5: visual recognition. I think it's falling way short of what
44:36
Speaker 5: you're hoping for in terms of getting us to a
44:39
Speaker 5: kind of Jungian visual category, true essence extractor. And one
44:44
Speaker 5: of the reasons that we know this is because these
44:47
Speaker 5: systems can be very easily fooled, that you can just
44:49
Speaker 5: change a couple pixels in what is clearly an orangutan photo,
44:54
Speaker 5: and suddenly it thinks it's a flamingo. And you as
44:57
Speaker 5: a human being can look at this and say, there's
44:59
Speaker 5: nothing flamingo about this picture at all. But the system
45:02
Speaker 5: is like, I am 99% convinced that that is a flamingo.
45:05
Speaker 5: So whatever it is that it's latching onto is distinct
45:09
Speaker 5: from whatever it is that we latch onto, because they
45:11
Speaker 5: seem very alien when they make those kinds of mistakes.
45:14
Speaker 3: And that's what they're trying to find out. What is
45:17
Speaker 3: that thing? What is the quality that human beings know?
45:21
Speaker 3: And how can that be taught to a computer? How
45:24
Speaker 3: can that be turned into a piece of code that
45:26
Speaker 3: will reliably recognize it? So fabulous question though. I think
45:31
Speaker 3: human brains are doing this and how do we describe
45:35
Speaker 3: that process?
45:36
Speaker 1: You know, people have shown me examples of dream interpretation
45:39
Speaker 1: that chat GPT has done. And like you're saying, they're
45:42
Speaker 1: not entirely far afield. They may lack something that, that
45:47
Speaker 1: you've been getting at, I think Joseph, but, uh, you know,
45:50
Speaker 1: it sort of does make me wonder, will I be
45:52
Speaker 1: replaced one day? Will there be an AI union analyst? But, um,
45:57
Speaker 1: I want to bring up one more thing, and then
45:58
Speaker 1: I want to give you a chance to tell us
46:00
Speaker 1: a little bit about your book. And that is this
46:04
Speaker 1: experiment that was done in which a person was thinking
46:07
Speaker 1: about a giraffe, and somehow the software, the computer, whatever,
46:12
Speaker 1: I don't understand how this happened, was able to generate
46:14
Speaker 1: a drawing of a giraffe.
46:16
Speaker 5: I think with the details, it's a little less terrifying,
46:18
Speaker 5: but it is still quite fascinating. So I think typically
46:21
Speaker 5: what they will do is they will take a person
46:24
Speaker 5: and and instrument the person's head. So scan their brain
46:27
Speaker 5: so that they can get whatever readings they can get
46:29
Speaker 5: out of which areas of the brain are being active
46:32
Speaker 5: and to what degree. And then they will train a
46:35
Speaker 5: system by showing people imagery, lots and lots and lots
46:39
Speaker 5: and lots and lots and lots of pictures so that
46:41
Speaker 5: it can start to make correlations between, okay, when the
46:43
Speaker 5: person is perceiving a giraffe, the brain activity looks like this.
46:47
Speaker 5: When the person is perceiving a puppy dog, the brain
46:49
Speaker 5: activity looks like this. When the person is perceiving a
46:52
Speaker 5: long highway on a desert, the brain activity looks like this.
46:57
Speaker 5: And so that, like being able to see the giraffe
47:01
Speaker 5: in the imagery is because there's been lots and lots
47:04
Speaker 5: of examples that can be correlated together. What's interesting is
47:08
Speaker 5: that there's enough consistency in a person that when they
47:12
Speaker 5: think about a giraffe, their brain patterns tend to fall
47:15
Speaker 5: into a certain, you know, pattern. And so that that
47:20
Speaker 5: pattern could actually be recognized by a machine learning system.
47:23
Speaker 5: So it's still quite interesting, but I don't think it's
47:25
Speaker 5: like it just looked at you and it's like, I
47:27
Speaker 5: can tell you're thinking about prairie dogs. Yes, prairie dogs.
47:32
Speaker 1: I mean, you know, the potentials for surveillance obviously are concerning,
47:38
Speaker 1: but also exciting to think that you might be able
47:40
Speaker 1: to do this at night, wake up in the morning
47:43
Speaker 1: and have images of your dreams at night.
47:45
Speaker 5: And then we could automatically send them for chat GPT
47:47
Speaker 5: for analysis. Absolutely.
47:51
Speaker 3: Put us out of business.
47:52
Speaker 4: But all of this creates a kind of externalization of
47:58
Speaker 4: functions that have been inherently human, that we turn it
48:03
Speaker 4: over to a machine. And that flies in the face
48:07
Speaker 4: of the Jungian process of individuation, that we work with
48:12
Speaker 4: ourselves And rather than let the machine do it, it
48:21
Speaker 4: goes back to what you raised earlier, Joseph, about the
48:24
Speaker 4: archetype of the mother. Is it the good mother or
48:27
Speaker 4: the bad mother? Well, it's probably both. But if we're
48:31
Speaker 4: focused on what it can do for us, when are
48:37
Speaker 4: we thinking about how we work, what we work, that
48:42
Speaker 4: we work, with ourselves. And that, you know, I think
48:47
Speaker 4: is inherent in Jungian psychology is that that is how
48:51
Speaker 4: we grow. It's not out there. It's in here.
48:56
Speaker 5: So in the book, I really focus in on the
49:00
Speaker 5: question of empowering people. Like you need to be able
49:03
Speaker 5: to tell the machine what you want the machine to do.
49:06
Speaker 5: And so it really is very much about keeping that
49:08
Speaker 5: locus of control at at each of us, right? The
49:12
Speaker 5: reason the machine is there is to automate what we
49:14
Speaker 5: want it to do. I think that relationship has been
49:19
Speaker 5: degraded in the last couple of decades because software companies
49:23
Speaker 5: are getting so good at making the computer do things
49:26
Speaker 5: for us that that puts them in a position where
49:29
Speaker 5: they get to decide what we see and our reactions
49:31
Speaker 5: to things and what we're able to do. I don't
49:34
Speaker 5: think we can get away from that until more of
49:37
Speaker 5: us really have a feeling of efficacy, that we can
49:40
Speaker 5: actually get the computer to do the thing that you
49:42
Speaker 5: want it to do. And so that's sort of what
49:45
Speaker 5: I'm arguing for. And in the book, I talk through
49:48
Speaker 5: the different things that make up computer programs, essentially, and
49:52
Speaker 5: relate them to regular life things that we all know about.
49:56
Speaker 5: Because These concepts really are built very much on top
50:00
Speaker 5: of everyday concepts that we all use when we're trying
50:03
Speaker 5: to tell each other what to do and we're trying
50:05
Speaker 5: to understand the world around us. But I also relate
50:08
Speaker 5: each of those concepts to machine learning. So how is
50:10
Speaker 5: it that machine learning can maybe help us get those
50:13
Speaker 5: ideas across to the machine? The machine can kind of
50:15
Speaker 5: meet us partway. It's really still about us telling the
50:17
Speaker 5: machine what to do, but the machine can be a
50:20
Speaker 5: more active participant in that. And machine learning at least
50:23
Speaker 5: opens up that possibility. And so that's kind of the
50:26
Speaker 5: essence of what I try to convey.
50:27
Speaker 1: And so who's the book for?
50:29
Speaker 5: I like to think of it as people like me
50:33
Speaker 5: when I read the newspaper, right? When I'm reading the newspaper,
50:35
Speaker 5: I'm not an expert in any particular thing, but I'm
50:38
Speaker 5: interested and I want to hear about what's going on.
50:41
Speaker 5: And I really count on the journalists to kind of
50:43
Speaker 5: convey to me interesting things that are happening in the
50:46
Speaker 5: world and in science and in society. And so if
50:50
Speaker 5: I would like people to come to the book with
50:52
Speaker 5: that kind of perspective. Like, I'm interested. This could be
50:55
Speaker 5: something fun. But not people who are already trained up
50:58
Speaker 5: in computing.
50:59
Speaker 1: And where can people go if they want to learn
51:01
Speaker 1: more about your work or they want to buy the book?
51:03
Speaker 5: Oh, sure. So I do have a website called litmania.com.
51:09
Speaker 5: It's like litman with the I-A at the end. And
51:13
Speaker 5: there's links for pre-ordering the book. The book will be
51:16
Speaker 5: available October 3rd. But until then, yeah, you can see
51:21
Speaker 5: the very cute and accessible-looking, friendly-looking cover and the title.
51:27
Speaker 5: And it's links to lots of different bookstores. So if
51:29
Speaker 5: you have a preference as to where you want to
51:31
Speaker 5: buy stuff, the links are all there.
51:32
Speaker 4: And it can be pre-ordered on Amazon?
51:35
Speaker 5: Amazon is one of the ones that's there. Not everybody
51:37
Speaker 5: likes to order on Amazon, so there's other options.
51:40
Speaker 3: Michael, I want to ask you just one final question.
51:42
Speaker 3: I know you need to go. You've been interviewed many times.
51:46
Speaker 3: You have a wonderful public presence. What is at least
51:51
Speaker 3: one question that you have not been asked yet in
51:55
Speaker 3: an interview, but you wish someone had asked you? You're
51:58
Speaker 3: just waiting for somebody to ask you this question.
52:02
Speaker 5: Maybe, how can I get started?
52:04
Speaker 3: How can we get started, Michael?
52:07
Speaker 4: Yeah.
52:07
Speaker 5: Well, the book is a good place to start. But
52:10
Speaker 5: we've never been more awash in helpful information than we
52:13
Speaker 5: are today. It's really quite remarkable how how many people
52:17
Speaker 5: have just taken it upon themselves to get online and
52:20
Speaker 5: either write a really compelling blog post that can get
52:24
Speaker 5: you going or a YouTube video that can really kind
52:26
Speaker 5: of give you the basics so that you can start
52:29
Speaker 5: on whatever it is that you want to do. Like
52:31
Speaker 5: for me, it was changing the flat tire on my
52:34
Speaker 5: electric unicycle. It's like, oh, wait, there's a video for that.
52:37
Speaker 5: But if it's computing, if it's about expressing what you want,
52:43
Speaker 5: It's there for you. You should definitely dive in and
52:45
Speaker 5: give it a try. It doesn't have to be scary
52:47
Speaker 5: because it really is there to empower you.
52:50
Speaker 1: Well, thank you so much. We really could not have
52:53
Speaker 1: done this episode without your expertise. And we're really grateful
52:57
Speaker 1: that you.
52:57
Speaker 4: Took the time.
52:58
Speaker 5: It was great getting to talk to three intelligent people
53:00
Speaker 5: who have thought very deeply about minds because minds are
53:04
Speaker 5: the coolest thing in the universe.
53:06
Speaker 4: Yeah.
53:10
Speaker 2: You've been listening to This Jungian Life. From our website, thisjungianlife.com,
53:16
Speaker 2: you can follow us on Twitter, like us on Facebook,
53:19
Speaker 2: help us produce future episodes by funding us through Patreon,
53:23
Speaker 2: and submit your dreams for possible interpretation on another episode.
53:28
Speaker 2: We'd like to thank our listener who shared a dream
53:30
Speaker 2: for today's show and hope you'll let us know what
53:33
Speaker 2: topics you'd enjoy hearing more about. Until next time, keep
53:37
Speaker 2: living this Jungian life. Thank you.