WEBVTT - From the Vault: Punish the Machine, Part 1

0:00:05.720 --> 0:00:07.680
<v Speaker 1>Hey, welcome to Stuff to Blow Your Mind. My name

0:00:07.720 --> 0:00:10.639
<v Speaker 1>is Robert Lamb and I'm Joe McCormick, and it's Saturday.

0:00:10.720 --> 0:00:13.640
<v Speaker 1>Time for an episode from the Vault. This is the

0:00:13.680 --> 0:00:16.959
<v Speaker 1>beginning of a series that we did about about robot

0:00:17.000 --> 0:00:21.200
<v Speaker 1>culpability and machine punishment. I think we're just calling this

0:00:21.239 --> 0:00:24.000
<v Speaker 1>Punish the Machine Part one, right, So this originally aired

0:00:24.040 --> 0:00:29.040
<v Speaker 1>on April. This one raised a lot of interesting questions

0:00:29.120 --> 0:00:34.960
<v Speaker 1>and so we hope you enjoy it. Welcome to stot

0:00:35.000 --> 0:00:44.280
<v Speaker 1>to Blow Your Mind, production of My Heart Radio. Hey you,

0:00:44.320 --> 0:00:46.240
<v Speaker 1>welcome to Stuff to Blow Your Mind. My name is

0:00:46.320 --> 0:00:49.559
<v Speaker 1>Robert Lamb and I'm Joe McCormick. And right before we

0:00:49.600 --> 0:00:52.559
<v Speaker 1>started recording today, we were just talking about that iconic

0:00:52.640 --> 0:00:56.800
<v Speaker 1>scene and returned to the Jedi where the droids are

0:00:56.800 --> 0:00:59.480
<v Speaker 1>sent to the droid torture Chamber. Do you remember there?

0:00:59.520 --> 0:01:01.640
<v Speaker 1>I guess it's not just a droid torture chamber. It's

0:01:01.680 --> 0:01:05.920
<v Speaker 1>sort of like the uh, the Droid onboarding center right

0:01:05.959 --> 0:01:08.200
<v Speaker 1>where the you know, R two, D two and C

0:01:08.400 --> 0:01:10.600
<v Speaker 1>three po have been given as gifts to job of

0:01:10.640 --> 0:01:13.080
<v Speaker 1>the Hut and they go meet their new like droid

0:01:13.160 --> 0:01:15.959
<v Speaker 1>boss and he's like you you're a feisty little one

0:01:16.000 --> 0:01:18.760
<v Speaker 1>and he's signing them in but uh, he sees that

0:01:18.920 --> 0:01:21.280
<v Speaker 1>R two D two is a is a bad robot

0:01:21.319 --> 0:01:24.720
<v Speaker 1>who needs discipline, and R two D two is confronted

0:01:24.760 --> 0:01:29.080
<v Speaker 1>with these images of robots being punished with various corporal punishments,

0:01:29.160 --> 0:01:32.280
<v Speaker 1>like one is getting stretched on a robot rack and

0:01:32.319 --> 0:01:36.160
<v Speaker 1>another one is getting its feet burned. Yes, this is

0:01:36.200 --> 0:01:40.040
<v Speaker 1>a This is a great scene, one that that definitely

0:01:40.080 --> 0:01:42.320
<v Speaker 1>burns its way into your your brain as a as

0:01:42.360 --> 0:01:44.200
<v Speaker 1>a young viewer, and maybe you don't think about it

0:01:44.240 --> 0:01:46.640
<v Speaker 1>that much for a long time, but uh, it's it's

0:01:46.680 --> 0:01:49.360
<v Speaker 1>still in there. It's it takes place in the bowels

0:01:49.400 --> 0:01:53.720
<v Speaker 1>of job as Palace on tattooing and it's um, yeah,

0:01:53.720 --> 0:01:58.240
<v Speaker 1>it's like Droid intake but also Droid corrections. It's there.

0:01:58.280 --> 0:02:00.440
<v Speaker 1>There are a number of different department that I think

0:02:00.480 --> 0:02:04.640
<v Speaker 1>are converging here and and it ray it ultimately kind

0:02:04.640 --> 0:02:09.120
<v Speaker 1>of raises some interesting questions about um about ethics and

0:02:09.160 --> 0:02:12.880
<v Speaker 1>punishment and in crime, and certainly as it relates to

0:02:13.200 --> 0:02:16.800
<v Speaker 1>two robots. Uh. Of course, one thing, this important distress

0:02:16.840 --> 0:02:18.840
<v Speaker 1>here is like none of this was really intended in

0:02:18.880 --> 0:02:23.560
<v Speaker 1>these scenes. This was about having droids doing things that

0:02:23.960 --> 0:02:27.079
<v Speaker 1>humans would be doing to each other in other pieces

0:02:27.080 --> 0:02:30.919
<v Speaker 1>of cinema, certainly, things like old pirate movies or old

0:02:31.000 --> 0:02:33.120
<v Speaker 1>sin bad movies or what have you. I mean, that's

0:02:33.200 --> 0:02:36.399
<v Speaker 1>kind of that's kind of Star Wars in a nutshell, right. Uh.

0:02:36.440 --> 0:02:39.720
<v Speaker 1>This whole portion of Return of the Jedi is essentially

0:02:39.800 --> 0:02:43.160
<v Speaker 1>a big pirate movie, a big swashbuckler set in an

0:02:43.160 --> 0:02:45.640
<v Speaker 1>alien location. Oh yeah, the job of the hud as

0:02:45.639 --> 0:02:50.120
<v Speaker 1>a pirate captain. Yeah. But something interesting occurs when you

0:02:50.160 --> 0:02:55.840
<v Speaker 1>replace the humans in these these tropic scenes with machines. Uh,

0:02:55.880 --> 0:02:57.720
<v Speaker 1>and then you think about it, you know, you think

0:02:57.720 --> 0:03:00.720
<v Speaker 1>about why is that robot torturing the other as if

0:03:00.720 --> 0:03:03.880
<v Speaker 1>it makes perfect sense if it's humans doing it, But

0:03:04.000 --> 0:03:06.799
<v Speaker 1>then the things that we create in our image when

0:03:06.800 --> 0:03:09.280
<v Speaker 1>they're doing it, suddenly we start seeing the flaws in

0:03:09.280 --> 0:03:12.480
<v Speaker 1>our reasoning. Suddenly we start questioning, well, how does this

0:03:12.520 --> 0:03:15.960
<v Speaker 1>whole system supposed to work? Uh? And maybe this whole

0:03:15.960 --> 0:03:18.840
<v Speaker 1>system doesn't work. Well. Yeah, there are multiple levels of

0:03:18.880 --> 0:03:22.080
<v Speaker 1>absurdities in the scene. One is the idea that this

0:03:22.240 --> 0:03:24.600
<v Speaker 1>robot is just sort of like coolly telling are two,

0:03:24.639 --> 0:03:26.600
<v Speaker 1>that he is going to learn some discipline. But then

0:03:26.680 --> 0:03:29.640
<v Speaker 1>the image that accompanies that is like, clearly just like

0:03:29.919 --> 0:03:33.840
<v Speaker 1>extreme robot torture, Like it's something way beyond what would

0:03:33.840 --> 0:03:36.480
<v Speaker 1>have to do with with discipline in the real world.

0:03:36.720 --> 0:03:39.040
<v Speaker 1>But then the other level of absurdity is that it's

0:03:39.120 --> 0:03:42.280
<v Speaker 1>robots in the scene, but coming off of the issue

0:03:42.400 --> 0:03:46.240
<v Speaker 1>of just like barbaric pirate torture directly and more to

0:03:46.360 --> 0:03:51.720
<v Speaker 1>the broader question of robots and discipline and punishment. Uh,

0:03:51.960 --> 0:03:53.960
<v Speaker 1>this is something that we actually wanted to talk about

0:03:54.000 --> 0:03:57.800
<v Speaker 1>today because the issue of robot moral and legal agency

0:03:58.400 --> 0:04:00.760
<v Speaker 1>is something I've been interested in for a long time.

0:04:00.800 --> 0:04:02.720
<v Speaker 1>I've talked about it, It's come up on the show

0:04:02.760 --> 0:04:06.320
<v Speaker 1>in the past and in briefer ways, um, and today

0:04:06.360 --> 0:04:09.320
<v Speaker 1>I wanted to come back and devote a full episode

0:04:09.320 --> 0:04:11.080
<v Speaker 1>to the subject. I guess actually we're gonna be talking

0:04:11.080 --> 0:04:13.840
<v Speaker 1>about this for a couple of episodes now. The question

0:04:14.040 --> 0:04:19.640
<v Speaker 1>of as machines AI robots become more independent and act

0:04:19.760 --> 0:04:23.280
<v Speaker 1>more like agents more like humans do, how are we

0:04:23.360 --> 0:04:28.200
<v Speaker 1>to understand their moral and legal culpability when they do

0:04:28.320 --> 0:04:31.440
<v Speaker 1>something that harms people? And is there such a thing

0:04:31.520 --> 0:04:36.279
<v Speaker 1>as robot punishment robot discipline? Do these concepts reflect anything

0:04:36.320 --> 0:04:39.840
<v Speaker 1>that's achievable in the real world and practical, and if so,

0:04:39.960 --> 0:04:42.120
<v Speaker 1>how would any of this work? Yeah? I think one

0:04:42.120 --> 0:04:44.599
<v Speaker 1>of the most interesting things about this topic is that

0:04:44.839 --> 0:04:47.520
<v Speaker 1>it does force us to force a face off between

0:04:47.520 --> 0:04:51.120
<v Speaker 1>what robots and ai actually are or will be, and

0:04:51.160 --> 0:04:55.440
<v Speaker 1>how we think about them, indeed, how we anthropomorphize them. Um,

0:04:55.560 --> 0:04:58.720
<v Speaker 1>And perhaps it might be helpful to take a step

0:04:58.760 --> 0:05:02.120
<v Speaker 1>back and think about something far less advanced as a

0:05:02.200 --> 0:05:04.520
<v Speaker 1>robot and think something. I think about something more like

0:05:04.560 --> 0:05:08.880
<v Speaker 1>a hammer. Okay, so everyone's heard the old adage that

0:05:08.920 --> 0:05:11.919
<v Speaker 1>it's a poor carpenter who blames their tools, right, But

0:05:12.000 --> 0:05:14.800
<v Speaker 1>of course we do this all the time. Uh, the

0:05:14.839 --> 0:05:18.080
<v Speaker 1>hammer slips, it hits our fingers, and we may, at

0:05:18.160 --> 0:05:20.680
<v Speaker 1>least in the heat of the moment, blame the hammer

0:05:21.080 --> 0:05:23.800
<v Speaker 1>for the failure. Now, we may get over this quickly,

0:05:24.160 --> 0:05:26.880
<v Speaker 1>but then again, we may decide that the hammer truly

0:05:26.960 --> 0:05:29.640
<v Speaker 1>is at fault and it should be used less. We

0:05:29.720 --> 0:05:33.240
<v Speaker 1>might also take this idea to a number of different extremes.

0:05:33.279 --> 0:05:35.720
<v Speaker 1>We might decide that the hammer is not merely at

0:05:35.760 --> 0:05:38.880
<v Speaker 1>fault but faulty, and then we're entitled to at least

0:05:38.880 --> 0:05:41.719
<v Speaker 1>a refund for its purchase. Or we might decide that

0:05:41.760 --> 0:05:44.680
<v Speaker 1>the hammer needs to actually be punished and and this

0:05:44.760 --> 0:05:48.040
<v Speaker 1>of course is ridiculous. And yet the idea of punishing

0:05:48.080 --> 0:05:50.440
<v Speaker 1>the hammer by say, putting it in the corner, or

0:05:50.560 --> 0:05:53.800
<v Speaker 1>perhaps you have an old toolbox of shame that's just

0:05:53.920 --> 0:05:57.559
<v Speaker 1>for the misbehaving tools, or maybe it's it's less thought

0:05:57.560 --> 0:06:00.400
<v Speaker 1>out and you just throw the hammer across the yard

0:06:00.720 --> 0:06:04.040
<v Speaker 1>as punishment for what it has done to you. Um. Again,

0:06:04.080 --> 0:06:06.920
<v Speaker 1>these are ridiculous things to do, but the idea of

0:06:06.920 --> 0:06:08.920
<v Speaker 1>doing them is not that far and from us. Um,

0:06:09.040 --> 0:06:10.960
<v Speaker 1>those of you listening, you may have engaged in this

0:06:11.000 --> 0:06:13.640
<v Speaker 1>sort of thing as well. You might also simply throw

0:06:13.680 --> 0:06:17.599
<v Speaker 1>the tool away and otherwise perfectly good tool. Um. I

0:06:17.640 --> 0:06:19.600
<v Speaker 1>know that I did this once with a knife sharpening

0:06:19.640 --> 0:06:22.039
<v Speaker 1>gadget that caused me to cut my finger my and

0:06:22.320 --> 0:06:25.600
<v Speaker 1>my like reaction was this thing has now injured me,

0:06:25.920 --> 0:06:28.520
<v Speaker 1>it has drawn my blood. Uh, I am, I'm getting

0:06:28.600 --> 0:06:30.359
<v Speaker 1>rid of it. It goes in the trash. It bore

0:06:30.440 --> 0:06:33.880
<v Speaker 1>malice against me. Yeah, or you know, I you know

0:06:33.960 --> 0:06:36.720
<v Speaker 1>ultimately it's I mean, I mean you you get into

0:06:36.800 --> 0:06:39.440
<v Speaker 1>arguments about different tools, like is this a dangerous tool?

0:06:39.480 --> 0:06:41.800
<v Speaker 1>And in that case that was my reasoning. It's like

0:06:42.240 --> 0:06:45.360
<v Speaker 1>this tool is dangerous, it's not enabling me to do

0:06:45.400 --> 0:06:48.560
<v Speaker 1>what I want to do without drawing blood, so it

0:06:48.560 --> 0:06:50.840
<v Speaker 1>goes in the trash um. But then there have been

0:06:50.880 --> 0:06:53.800
<v Speaker 1>other cases where like I had a mandolin for slicing

0:06:53.880 --> 0:06:57.760
<v Speaker 1>up carrots, and um, I like nicked my finger on it.

0:06:57.839 --> 0:07:00.720
<v Speaker 1>But oh I I nicked my fingers A rule they

0:07:00.760 --> 0:07:04.000
<v Speaker 1>can be. But I nicked my finger not using it,

0:07:04.279 --> 0:07:06.760
<v Speaker 1>but going into the drawer for something else. So I

0:07:06.800 --> 0:07:09.279
<v Speaker 1>punished it by putting at the very bottom of the drawer.

0:07:10.040 --> 0:07:12.240
<v Speaker 1>But I didn't throw it away. Uh huh. So I

0:07:12.320 --> 0:07:14.080
<v Speaker 1>think if we all think, think back, you know, we

0:07:14.120 --> 0:07:16.160
<v Speaker 1>have examples of this sort of thing from our from

0:07:16.200 --> 0:07:18.680
<v Speaker 1>our life. Well sure, I mean I'm going to talk

0:07:18.960 --> 0:07:20.800
<v Speaker 1>in this episode about some of the ways that we

0:07:20.920 --> 0:07:26.160
<v Speaker 1>mindlessly apply social rules to robots. But yeah, I think

0:07:26.160 --> 0:07:28.800
<v Speaker 1>what you're illustrating here is that you don't even have

0:07:28.840 --> 0:07:32.000
<v Speaker 1>to get to the robot agency stage before people start

0:07:32.080 --> 0:07:35.640
<v Speaker 1>doing that. I mean people mindlessly, to a lesser extent,

0:07:35.760 --> 0:07:39.840
<v Speaker 1>mindlessly apply social rules and rules derived for managing human

0:07:39.880 --> 0:07:44.800
<v Speaker 1>relationships to inanimate objects with no moving parts. Yeah. Yeah,

0:07:44.800 --> 0:07:47.200
<v Speaker 1>you don't even have to get into a room ba

0:07:47.360 --> 0:07:49.320
<v Speaker 1>or anything, or you know, you can deal with the hammer,

0:07:49.360 --> 0:07:52.640
<v Speaker 1>the can opener but we also you know, it's it's

0:07:52.640 --> 0:07:55.480
<v Speaker 1>also a sad fact that many pat pet owners will

0:07:55.680 --> 0:07:59.560
<v Speaker 1>punish an animal for a transgression, transgression, but scientific evidence

0:07:59.600 --> 0:08:03.440
<v Speaker 1>shows that this tends to not actually work at least

0:08:03.560 --> 0:08:07.160
<v Speaker 1>in most of the circumstances that it's used. Um so

0:08:07.960 --> 0:08:11.960
<v Speaker 1>you know, even it's not merely with with tools and

0:08:12.040 --> 0:08:15.320
<v Speaker 1>inanimate objects, but even non human entities were liable to

0:08:15.360 --> 0:08:20.280
<v Speaker 1>engage in this kind of discipline based thinking. Now, most

0:08:20.320 --> 0:08:22.840
<v Speaker 1>of the studies I think that necessarily relate to dogs

0:08:23.520 --> 0:08:25.320
<v Speaker 1>if I remember correctly, And there's a lot going on

0:08:25.400 --> 0:08:28.240
<v Speaker 1>here that doesn't relate directly to inanimate objects and robots,

0:08:28.440 --> 0:08:30.760
<v Speaker 1>but it illustrates how we tend to approach the punishment

0:08:30.760 --> 0:08:34.199
<v Speaker 1>of other agents and perceived agents. Well, yeah, there's a disconnect,

0:08:34.200 --> 0:08:35.920
<v Speaker 1>and this will be highlighted in one of the papers

0:08:35.960 --> 0:08:38.160
<v Speaker 1>we're going to talk about in in this pair of episodes,

0:08:38.200 --> 0:08:42.600
<v Speaker 1>But there's a disconnect in that punishment is often logically

0:08:43.320 --> 0:08:47.040
<v Speaker 1>characterized as serving one type of purpose, but then is

0:08:47.080 --> 0:08:50.200
<v Speaker 1>applied more like it serves another type of purpose. So

0:08:50.240 --> 0:08:53.920
<v Speaker 1>like it is logically explained as say a deterrent, right,

0:08:53.960 --> 0:08:57.120
<v Speaker 1>I mean, if you if you're talking about uh, legal theories,

0:08:57.120 --> 0:08:59.280
<v Speaker 1>of punishment. One of the main things that people come

0:08:59.360 --> 0:09:02.040
<v Speaker 1>up with is say, well, the remedy provided by the

0:09:02.120 --> 0:09:05.000
<v Speaker 1>law is in order to punish the person who did

0:09:05.000 --> 0:09:07.520
<v Speaker 1>the bad thing, in order to send a message that

0:09:07.559 --> 0:09:10.760
<v Speaker 1>people should not do this bad thing and thus maybe

0:09:10.800 --> 0:09:13.720
<v Speaker 1>discourage other people from doing something similar in the future,

0:09:13.840 --> 0:09:16.679
<v Speaker 1>or discourage the same person from doing it again. And

0:09:16.760 --> 0:09:19.280
<v Speaker 1>if it were to actually serve that purpose, it's debatable

0:09:19.280 --> 0:09:21.720
<v Speaker 1>in what cases it does actually serve that purpose. Maybe

0:09:21.760 --> 0:09:24.280
<v Speaker 1>sometimes it does, but that is a you know, you

0:09:24.280 --> 0:09:28.040
<v Speaker 1>could argue that's a rational, logical thing that prevents harm.

0:09:28.400 --> 0:09:31.440
<v Speaker 1>But the way punishment is actually often inflicted in the

0:09:31.440 --> 0:09:35.400
<v Speaker 1>real world seems to be more consistent with judgments based

0:09:35.480 --> 0:09:41.199
<v Speaker 1>on like emotional satisfaction of the idea of having been wronged. Yeah.

0:09:41.720 --> 0:09:45.480
<v Speaker 1>And then also we get into this area where, uh,

0:09:45.600 --> 0:09:50.360
<v Speaker 1>we have a couple of different factors encouraging traditions of discipline. Um,

0:09:50.480 --> 0:09:54.280
<v Speaker 1>particularly if we look at a parenthood, which there's some

0:09:54.440 --> 0:09:57.920
<v Speaker 1>crossover between discipline and parents who I didn't discipline and

0:09:58.040 --> 0:10:01.640
<v Speaker 1>the criminal justice system. But you know, uh, not everything

0:10:01.720 --> 0:10:04.079
<v Speaker 1>is going to line up one to one here. But um,

0:10:04.800 --> 0:10:07.200
<v Speaker 1>on the childhood example, it's benn argue that parents use

0:10:07.240 --> 0:10:10.319
<v Speaker 1>punishment first of all, because it's an emotional response out

0:10:10.320 --> 0:10:13.720
<v Speaker 1>of anger and anger that may be mismanaged. But then

0:10:14.320 --> 0:10:17.800
<v Speaker 1>on top of this, it's, you know, something that's culturally

0:10:17.840 --> 0:10:22.400
<v Speaker 1>passed down and punishment may seem to work. I was

0:10:22.400 --> 0:10:25.760
<v Speaker 1>reading about this in a Psychology Today article by Michael Carson,

0:10:26.080 --> 0:10:29.880
<v Speaker 1>pH d j D. And uh, this is what they said. Quote,

0:10:29.920 --> 0:10:32.560
<v Speaker 1>because the child is inhibited in your presence, it's easy

0:10:32.600 --> 0:10:35.880
<v Speaker 1>to think they would be inhibited in your absence. Punishment

0:10:35.920 --> 0:10:41.160
<v Speaker 1>produces politeness, not morality. Thus, the inhibited, obedient child inadvertently

0:10:41.200 --> 0:10:45.760
<v Speaker 1>reinforces the parents punitive behavior by acting obedient. For the

0:10:45.760 --> 0:10:49.440
<v Speaker 1>sorts of parents who find obedient children reinforcing, Yeah, that

0:10:49.520 --> 0:10:52.280
<v Speaker 1>raises an interesting question. I mean, I've been mainly thinking

0:10:52.320 --> 0:10:56.439
<v Speaker 1>for this episode about about legal punishments, but like when

0:10:56.480 --> 0:10:59.120
<v Speaker 1>it comes down to parenting, that's a very different kind

0:10:59.160 --> 0:11:03.320
<v Speaker 1>of thing because both parenting in the legal system involved punishment,

0:11:03.520 --> 0:11:07.440
<v Speaker 1>but parenting is not subject to a legal system, right,

0:11:07.480 --> 0:11:10.680
<v Speaker 1>So there's no there is no systematized way by which

0:11:10.760 --> 0:11:13.920
<v Speaker 1>justice is administered from a parent. It's just just I mean,

0:11:13.960 --> 0:11:15.280
<v Speaker 1>I think a lot of times it's just sort of

0:11:15.280 --> 0:11:17.440
<v Speaker 1>like whatever the parent can manage to do in the

0:11:17.480 --> 0:11:21.000
<v Speaker 1>moment because like the kids driving them crazy or something. Yeah,

0:11:21.080 --> 0:11:24.079
<v Speaker 1>usually the child can't take it to a higher court, right,

0:11:24.440 --> 0:11:26.960
<v Speaker 1>But I mean, I think you're absolutely right that whether

0:11:27.000 --> 0:11:30.040
<v Speaker 1>you're talking about discipline administered by a parent or the

0:11:30.120 --> 0:11:33.280
<v Speaker 1>justice system as a whole, I'd say that both are

0:11:33.280 --> 0:11:38.440
<v Speaker 1>probably based more on tradition and philosophy and less on

0:11:38.600 --> 0:11:41.680
<v Speaker 1>a scientifically rigorous study of the most efficient ways to

0:11:41.760 --> 0:11:44.960
<v Speaker 1>reduce harm. And one of the interesting things about thinking

0:11:45.000 --> 0:11:48.920
<v Speaker 1>about how law could potentially be applied to harm caused

0:11:49.040 --> 0:11:52.960
<v Speaker 1>by autonomous machines is that it may help give us

0:11:53.000 --> 0:11:56.600
<v Speaker 1>some insights on ways that the justice system as it

0:11:56.640 --> 0:11:59.800
<v Speaker 1>exists and is applied to humans today tends to be

0:12:00.080 --> 0:12:04.960
<v Speaker 1>of irrationally already, like with respect to humans. Yeah, and again,

0:12:04.960 --> 0:12:08.240
<v Speaker 1>this is what's so interesting about this this paper, I mean, well,

0:12:08.280 --> 0:12:10.880
<v Speaker 1>the papers that we're going to discuss this topic in general, though,

0:12:11.320 --> 0:12:15.000
<v Speaker 1>is if you start, you start comparing machine possibilities to

0:12:15.080 --> 0:12:19.040
<v Speaker 1>human possibilities, and it's on one level of thought experiment

0:12:19.120 --> 0:12:21.839
<v Speaker 1>in how you would hold machines are responsible, but then

0:12:21.840 --> 0:12:24.560
<v Speaker 1>it makes you rethink the way humans are held responsible.

0:12:24.640 --> 0:12:27.280
<v Speaker 1>You know. It's like, um, you might think you have

0:12:27.320 --> 0:12:31.000
<v Speaker 1>a pretty square a way, like if if an adult

0:12:31.080 --> 0:12:34.280
<v Speaker 1>sells a pack of cigarettes to someone who's underage, right,

0:12:34.679 --> 0:12:36.480
<v Speaker 1>but then one of a machine does the same thing,

0:12:37.600 --> 0:12:39.080
<v Speaker 1>how do you treat the machine? Do you treat the

0:12:39.080 --> 0:12:41.240
<v Speaker 1>machine like an adult? And then in trying to figure

0:12:41.280 --> 0:12:43.600
<v Speaker 1>out how to treat this machine, does it make you

0:12:43.600 --> 0:12:46.160
<v Speaker 1>rethink how you should be treating the adult who engaged

0:12:46.200 --> 0:12:48.520
<v Speaker 1>in this behavior. I don't know, yeah, And I think

0:12:48.520 --> 0:12:51.000
<v Speaker 1>a lot of that will come down to our understanding

0:12:51.040 --> 0:12:53.880
<v Speaker 1>of what the machine is capable of, like what kind

0:12:53.920 --> 0:12:56.920
<v Speaker 1>of constraints it has, what type of what level of

0:12:57.000 --> 0:13:00.600
<v Speaker 1>autonomy it seems to be operating at. I mean, again, Weirdly,

0:13:00.640 --> 0:13:03.400
<v Speaker 1>even when people set out to define clear rules for

0:13:03.440 --> 0:13:06.439
<v Speaker 1>what makes a machine culpable, there there's still going to

0:13:06.520 --> 0:13:08.760
<v Speaker 1>be a lot of subjectivity in it. I'm I'm looking

0:13:08.800 --> 0:13:13.319
<v Speaker 1>at like legal definitions of what constitutes a robot versus

0:13:13.400 --> 0:13:17.640
<v Speaker 1>just a machine, and some of these definitions involved things like, well,

0:13:17.679 --> 0:13:21.760
<v Speaker 1>a robot feels like a social agent. So there's still, like,

0:13:22.040 --> 0:13:24.560
<v Speaker 1>you know, an element of subjectivity. But I think that's

0:13:24.559 --> 0:13:27.040
<v Speaker 1>correct in how we actually apply the term most of

0:13:27.080 --> 0:13:30.080
<v Speaker 1>the time, right, Like something is like a gut feeling

0:13:30.200 --> 0:13:33.760
<v Speaker 1>about how this machine is behaving in your world. Is

0:13:33.760 --> 0:13:37.520
<v Speaker 1>it acting more like a fixed, you know, brainless machine,

0:13:37.600 --> 0:13:41.920
<v Speaker 1>or is it acting a little bit more like a person. So,

0:13:42.120 --> 0:13:44.840
<v Speaker 1>while it would be one thing if if it were

0:13:44.960 --> 0:13:48.360
<v Speaker 1>basically a cigarette vending machine that was selling to children,

0:13:48.360 --> 0:13:50.640
<v Speaker 1>but if it were a machine that went door to

0:13:50.720 --> 0:13:54.520
<v Speaker 1>door and and rang the doorbell and then asked for

0:13:54.559 --> 0:13:56.800
<v Speaker 1>the children so they could sell them cigarettes, that would

0:13:56.800 --> 0:13:58.480
<v Speaker 1>be a different matter. I mean, yeah, I mean I

0:13:58.559 --> 0:14:01.920
<v Speaker 1>think that would require different types of remedies. Probably, Yeah,

0:14:02.160 --> 0:14:03.800
<v Speaker 1>I mean I think a lot of people would probably

0:14:03.800 --> 0:14:06.480
<v Speaker 1>look at the cigarette vending machine and say where was

0:14:06.520 --> 0:14:09.240
<v Speaker 1>the vending machine placed? Why was it in a place

0:14:09.280 --> 0:14:12.640
<v Speaker 1>that children could have access to it? Rather than attacking

0:14:12.640 --> 0:14:15.280
<v Speaker 1>the fundamentals of the machine itself, if it's going door

0:14:15.320 --> 0:14:17.600
<v Speaker 1>to door and giving cigarettes to kids, yeah, then people

0:14:17.600 --> 0:14:20.760
<v Speaker 1>are probably going to attack the fundamentals and the moral

0:14:20.880 --> 0:14:24.920
<v Speaker 1>character of the robot. You have not attacked the robot itself.

0:14:24.960 --> 0:14:34.760
<v Speaker 1>It would just rob justice in somebody's front yard. Yeah, alright,

0:14:34.760 --> 0:14:36.400
<v Speaker 1>So I guess I want to introduce one of the

0:14:36.400 --> 0:14:38.840
<v Speaker 1>papers we're gonna be looking at in this pair of episodes,

0:14:39.000 --> 0:14:42.680
<v Speaker 1>and it is by Mark A. Limley and Brian casey

0:14:42.880 --> 0:14:46.920
<v Speaker 1>called Remedies for Robots published in the University of Chicago

0:14:47.040 --> 0:14:51.840
<v Speaker 1>Law Review in twenty nineteen. And this is a big paper.

0:14:51.920 --> 0:14:54.760
<v Speaker 1>It's like eighty something pages long with a with a

0:14:54.800 --> 0:14:57.840
<v Speaker 1>lot of different interesting, uh thoughts in it. We're not

0:14:57.880 --> 0:15:00.560
<v Speaker 1>going to be able to cover the entire thing in depth,

0:15:00.640 --> 0:15:03.280
<v Speaker 1>but it's worth looking up. You can easily find a

0:15:03.280 --> 0:15:05.960
<v Speaker 1>full PDF of it if you want to read it

0:15:06.000 --> 0:15:07.800
<v Speaker 1>in depth. And we're gonna look at some of the

0:15:07.960 --> 0:15:11.040
<v Speaker 1>larger framework it lays out and then some interesting thoughts

0:15:11.160 --> 0:15:13.440
<v Speaker 1>raised to buy it. But to kick it off here

0:15:13.600 --> 0:15:17.600
<v Speaker 1>the the author's right quote, what happens when artificially intelligent

0:15:17.680 --> 0:15:22.680
<v Speaker 1>robots misbehave? The question is not just hypothetical. As robotics

0:15:22.680 --> 0:15:27.320
<v Speaker 1>and artificial intelligence systems increasingly integrate into our society, they

0:15:27.360 --> 0:15:30.920
<v Speaker 1>will do bad things. We seek to explore what remedies

0:15:30.960 --> 0:15:33.960
<v Speaker 1>the law can and should provide once a robot has

0:15:34.040 --> 0:15:38.000
<v Speaker 1>caused harm. Now, obviously we're going to be focused less

0:15:38.080 --> 0:15:41.720
<v Speaker 1>on the like minute particulars of US legal precedent here

0:15:41.760 --> 0:15:45.600
<v Speaker 1>and more on the broader issues they raise about robot agency,

0:15:45.920 --> 0:15:49.400
<v Speaker 1>robot moral decision making, and how that interacts with harm

0:15:49.600 --> 0:15:53.240
<v Speaker 1>and morality and justice. And the authors start out in

0:15:53.280 --> 0:15:55.600
<v Speaker 1>their introduction by giving what I think is a really

0:15:55.640 --> 0:16:00.040
<v Speaker 1>fantastic example of how an autonomous robot with behave of

0:16:00.160 --> 0:16:03.120
<v Speaker 1>years guided by machine learning, which is how you know,

0:16:03.560 --> 0:16:07.640
<v Speaker 1>increasingly most robots are going to be controlled, can end

0:16:07.680 --> 0:16:10.600
<v Speaker 1>up doing things that are the exact opposite of what

0:16:10.680 --> 0:16:13.720
<v Speaker 1>was intended. So this case that they site is based

0:16:13.760 --> 0:16:16.840
<v Speaker 1>on a true story from a presentation at the eleventh

0:16:16.880 --> 0:16:22.480
<v Speaker 1>annual Stanford e Commerce Best Practices Conference in June, and

0:16:22.520 --> 0:16:27.480
<v Speaker 1>it goes like this quote. Engineers training and artificially intelligent

0:16:27.680 --> 0:16:31.440
<v Speaker 1>self flying drone were perplexed. They were trying to get

0:16:31.480 --> 0:16:34.840
<v Speaker 1>the drone to stay within a predefined circle and head

0:16:34.880 --> 0:16:38.040
<v Speaker 1>toward its center. Things were going well for a while.

0:16:38.440 --> 0:16:42.240
<v Speaker 1>The drone received positive reinforcement for its successful flights, and

0:16:42.280 --> 0:16:45.040
<v Speaker 1>it was improving its ability to navigate toward the middle

0:16:45.120 --> 0:16:49.960
<v Speaker 1>quickly and accurately. Then suddenly things changed. When the drone

0:16:50.000 --> 0:16:53.040
<v Speaker 1>near the edge of the circle, it would inexplicably turn

0:16:53.160 --> 0:16:57.040
<v Speaker 1>away from the center, leaving the circle. What went wrong?

0:16:57.800 --> 0:17:00.720
<v Speaker 1>After a long time spent puzzling over the problem, the

0:17:00.760 --> 0:17:04.960
<v Speaker 1>designers realized that whenever the drone left the circle during tests,

0:17:05.240 --> 0:17:08.359
<v Speaker 1>they had turned it off. Someone would then pick it

0:17:08.480 --> 0:17:11.600
<v Speaker 1>up and carry it back into the circle to start again.

0:17:12.440 --> 0:17:16.560
<v Speaker 1>From this pattern, the drones algorithm had learned correctly that

0:17:16.640 --> 0:17:19.840
<v Speaker 1>when it was sufficiently far from the center, the optimal

0:17:19.880 --> 0:17:22.040
<v Speaker 1>way to get back to the middle was to simply

0:17:22.160 --> 0:17:25.840
<v Speaker 1>leave it all together. As far as the drone was concerned,

0:17:25.920 --> 0:17:30.320
<v Speaker 1>it had discovered a wormhole somehow, flying outside of the

0:17:30.320 --> 0:17:33.760
<v Speaker 1>circle could be relied upon to magically teleport it closer

0:17:33.800 --> 0:17:37.280
<v Speaker 1>to the center, and far from violating the rules instilled

0:17:37.280 --> 0:17:40.480
<v Speaker 1>in it by its engineers, the drone had actually followed

0:17:40.520 --> 0:17:43.600
<v Speaker 1>them to a t. In doing so, however, it had

0:17:43.640 --> 0:17:48.520
<v Speaker 1>discovered an unforeseen shortcut, one that subverted its designer's true intent.

0:17:49.400 --> 0:17:52.800
<v Speaker 1>That's really good, that's that's it's as a yes, I

0:17:52.840 --> 0:17:55.800
<v Speaker 1>love it. This is such a great example of how

0:17:55.960 --> 0:17:59.560
<v Speaker 1>robots can fail in ways that are perfectly logical for

0:17:59.600 --> 0:18:03.000
<v Speaker 1>the machine means themselves, but hard for humans to predict

0:18:03.000 --> 0:18:06.520
<v Speaker 1>in advance, because we're not understanding how our you know,

0:18:06.680 --> 0:18:09.479
<v Speaker 1>our programming or the data sets we're training it on

0:18:10.000 --> 0:18:13.320
<v Speaker 1>is biasing its behavior in ways that that are strange

0:18:13.359 --> 0:18:15.880
<v Speaker 1>to us. And in this case, of of course, such

0:18:15.920 --> 0:18:20.480
<v Speaker 1>a malfunction is harmless, but as autonomous machines become more

0:18:20.520 --> 0:18:23.960
<v Speaker 1>and more integrated into the broader culture, not just in

0:18:24.160 --> 0:18:28.399
<v Speaker 1>controlled contained locations like factory floors and laboratories, but in

0:18:28.480 --> 0:18:31.080
<v Speaker 1>the wild so on the streets and in our homes

0:18:31.160 --> 0:18:35.919
<v Speaker 1>and stuff. There will inevitably be cases where robots fail

0:18:36.240 --> 0:18:40.480
<v Speaker 1>like this and fail in ways that cause catastrophic harm

0:18:40.600 --> 0:18:43.760
<v Speaker 1>to people. Yeah, and plus as an aside, we we

0:18:43.840 --> 0:18:46.040
<v Speaker 1>have to realize that even in cases where the machines

0:18:46.080 --> 0:18:48.880
<v Speaker 1>have not failed, there will be gray areas in which

0:18:48.880 --> 0:18:51.840
<v Speaker 1>it's not completely clear, and an argument could be made

0:18:51.840 --> 0:18:54.760
<v Speaker 1>in these cases for machine culpability with a variety of

0:18:54.800 --> 0:18:58.160
<v Speaker 1>intense and possible biases in place. Oh yeah, that's another

0:18:58.200 --> 0:19:00.240
<v Speaker 1>thing these authors talk about that there can be all

0:19:00.320 --> 0:19:04.359
<v Speaker 1>kinds of ways that, uh, that robotics and AI could

0:19:04.440 --> 0:19:07.920
<v Speaker 1>end up causing extreme harm to people without ever doing

0:19:07.960 --> 0:19:10.840
<v Speaker 1>anything that if a human did, it would be illegal.

0:19:11.480 --> 0:19:14.920
<v Speaker 1>What one example they give is like if, um, if

0:19:15.080 --> 0:19:19.639
<v Speaker 1>Google were to suddenly change its Google Maps algorithm so

0:19:19.720 --> 0:19:24.120
<v Speaker 1>that it routed all of the city's traffic through your neighborhood. Like,

0:19:24.560 --> 0:19:27.679
<v Speaker 1>nothing illegal about that. It doesn't like commit a crime

0:19:27.720 --> 0:19:31.400
<v Speaker 1>against you, but this is going to drastically negatively impact

0:19:31.400 --> 0:19:33.800
<v Speaker 1>your quality of life, and it's a decision that's just

0:19:33.920 --> 0:19:36.679
<v Speaker 1>like a could be a quirk of an algorithm in

0:19:36.680 --> 0:19:40.560
<v Speaker 1>a machine. Now this paper in particular concerns the legal

0:19:40.640 --> 0:19:44.520
<v Speaker 1>concept of remedies. So I was reading about remedies. A

0:19:44.840 --> 0:19:48.199
<v Speaker 1>common legal definition that I found is quote the means

0:19:48.280 --> 0:19:52.000
<v Speaker 1>to achieve justice in any matter in which legal rights

0:19:52.040 --> 0:19:55.760
<v Speaker 1>are involved. Or in the words of Limely and Casey,

0:19:55.840 --> 0:19:58.159
<v Speaker 1>what do I get when I win? Right? So, if

0:19:58.200 --> 0:20:00.320
<v Speaker 1>you if you take somebody to court, be as you

0:20:00.359 --> 0:20:04.280
<v Speaker 1>say they have harmed you, whatever outcome you're seeking from that,

0:20:04.280 --> 0:20:07.960
<v Speaker 1>that court case is the remedy. So usually when a

0:20:07.960 --> 0:20:10.639
<v Speaker 1>court case finds that somebody has done something wrong to

0:20:10.680 --> 0:20:13.919
<v Speaker 1>harm somebody else, the court responds to the finding of

0:20:13.960 --> 0:20:17.520
<v Speaker 1>guilt or blame by enforcing this remedy. And common remedies

0:20:17.520 --> 0:20:20.720
<v Speaker 1>would include a payment of money, right, a guilty defendant

0:20:20.720 --> 0:20:23.760
<v Speaker 1>has to pay money to the plaintiff, a punishment of

0:20:23.760 --> 0:20:26.680
<v Speaker 1>the offender, like maybe they go to jail, or a

0:20:26.680 --> 0:20:29.560
<v Speaker 1>court order to do something or not to do something.

0:20:29.680 --> 0:20:32.879
<v Speaker 1>For example, somebody is ordered not to drive a vehicle,

0:20:33.080 --> 0:20:35.240
<v Speaker 1>or they are ordered not to go within a hundred

0:20:35.280 --> 0:20:38.199
<v Speaker 1>feet of somebody else or something like that. Yeah, or

0:20:38.320 --> 0:20:40.879
<v Speaker 1>their their eyeball is removed, or they have to spend

0:20:41.040 --> 0:20:43.600
<v Speaker 1>a night in a hunted house something like that. Hopefully

0:20:43.640 --> 0:20:46.280
<v Speaker 1>not in modern law. But wait a minute, there are

0:20:46.320 --> 0:20:49.400
<v Speaker 1>some Sometimes you do read about some really strange like

0:20:49.760 --> 0:20:52.199
<v Speaker 1>remedies that are ordered by judges like I order you

0:20:52.240 --> 0:20:54.560
<v Speaker 1>to I don't know, to wear chicken suit or something,

0:20:55.119 --> 0:20:58.320
<v Speaker 1>right like, Yeah, there's some judges who like to get creative.

0:20:58.400 --> 0:21:01.720
<v Speaker 1>It seems weird. Yeah, I wonder I have There have

0:21:01.760 --> 0:21:03.639
<v Speaker 1>been cases where someone has to spend a night and

0:21:03.640 --> 0:21:05.720
<v Speaker 1>a hounded house due to a court order. I think

0:21:05.720 --> 0:21:07.879
<v Speaker 1>that would be a good setup for a film. But anyway,

0:21:08.200 --> 0:21:10.439
<v Speaker 1>so when you start looking at the idea of remedies,

0:21:10.600 --> 0:21:14.600
<v Speaker 1>remedies are complicated because they involve different types of implied

0:21:14.680 --> 0:21:18.760
<v Speaker 1>satisfaction on behalf of the victim or plaintiff. And some

0:21:18.880 --> 0:21:22.919
<v Speaker 1>are very clear and material and others are much more abstract.

0:21:23.000 --> 0:21:24.720
<v Speaker 1>So the very the ones that are very clear and

0:21:24.760 --> 0:21:28.040
<v Speaker 1>material are like, if I hit your car with my

0:21:28.119 --> 0:21:31.080
<v Speaker 1>car and I'm clearly at fault, I need to give

0:21:31.160 --> 0:21:34.440
<v Speaker 1>you a payment of cash to offset the material losses

0:21:34.480 --> 0:21:37.080
<v Speaker 1>to the value of your car. Right. But then other

0:21:37.119 --> 0:21:40.399
<v Speaker 1>times it's it's more abstract. It's you know, punishment of

0:21:40.440 --> 0:21:43.800
<v Speaker 1>an offender, to give the victim a sense of justice

0:21:44.119 --> 0:21:47.960
<v Speaker 1>or to allegedly discourage someone from committing this type of

0:21:48.000 --> 0:21:50.719
<v Speaker 1>harm or offense in the future. And then the authors

0:21:50.800 --> 0:21:53.480
<v Speaker 1>right that things get way more complicated when you bring

0:21:53.680 --> 0:21:57.480
<v Speaker 1>robots and AI into the picture. For example, if you're

0:21:57.480 --> 0:21:59.919
<v Speaker 1>trying to give a court order to a person, you know,

0:22:00.080 --> 0:22:02.680
<v Speaker 1>saying like you shall not drive a car, you shall

0:22:02.720 --> 0:22:05.080
<v Speaker 1>not you know, come within a hundred feet of this person.

0:22:05.440 --> 0:22:07.600
<v Speaker 1>You can do so in natural language. You can like

0:22:07.680 --> 0:22:10.480
<v Speaker 1>speak a sentence to them and you can expect them

0:22:10.520 --> 0:22:13.879
<v Speaker 1>to understand. But how do you get a court to

0:22:14.200 --> 0:22:17.720
<v Speaker 1>give an order to a robot not to do something?

0:22:18.480 --> 0:22:22.639
<v Speaker 1>Most robots don't have natural language processing, and even if

0:22:22.680 --> 0:22:24.440
<v Speaker 1>they do, a lot of times it's not that good.

0:22:24.840 --> 0:22:27.280
<v Speaker 1>So you might think, okay, well you just you know,

0:22:27.359 --> 0:22:30.560
<v Speaker 1>you give the court order to the robots programmer and

0:22:30.600 --> 0:22:32.879
<v Speaker 1>then it will and then they'll have to program the

0:22:32.960 --> 0:22:36.000
<v Speaker 1>robot to obey. But this is also really complicated, like

0:22:36.359 --> 0:22:40.199
<v Speaker 1>whose responsibility is it the robots current owner or the

0:22:40.240 --> 0:22:44.720
<v Speaker 1>original contractor or creator who made the robot? Uh? And

0:22:44.880 --> 0:22:47.680
<v Speaker 1>what if this is like an end user consumer device

0:22:48.080 --> 0:22:51.560
<v Speaker 1>that the owner doesn't have any ability to reprogram or

0:22:51.800 --> 0:22:54.119
<v Speaker 1>what if, in the case of robots whose behavior is

0:22:54.240 --> 0:22:56.919
<v Speaker 1>driven by machine learning or some other kind of system

0:22:57.000 --> 0:23:00.199
<v Speaker 1>that is, for practical purposes, a black box, what if

0:23:00.240 --> 0:23:02.760
<v Speaker 1>it's not even clear how you could reprogram it to

0:23:02.920 --> 0:23:08.800
<v Speaker 1>reliably obey the rule. Yeah, because there's a chance you

0:23:08.920 --> 0:23:14.440
<v Speaker 1>got to this position because the robot misinterpreted what was

0:23:14.520 --> 0:23:18.760
<v Speaker 1>asked of it. So if you then make additional requirements,

0:23:19.040 --> 0:23:22.280
<v Speaker 1>ones that maybe you know, haven't actually been tested before

0:23:22.359 --> 0:23:24.120
<v Speaker 1>but are just you know, that that are then brought

0:23:24.119 --> 0:23:27.720
<v Speaker 1>on by the court, that could conceivably create new problems,

0:23:27.840 --> 0:23:30.760
<v Speaker 1>right yeah, Yeah, totally, And and it keeps getting even

0:23:30.760 --> 0:23:33.760
<v Speaker 1>more complicated from there, like Limle and Casey right quote.

0:23:33.800 --> 0:23:38.080
<v Speaker 1>To complicate matters further, some systems, including many self driving cars,

0:23:38.560 --> 0:23:45.119
<v Speaker 1>distribute responsibility for their robots between both designers and downstream operators.

0:23:45.480 --> 0:23:48.639
<v Speaker 1>For systems of this kind, it has already proven extremely

0:23:48.640 --> 0:23:53.679
<v Speaker 1>difficult to allocate responsibility when accidents inevitably occur. It just

0:23:53.680 --> 0:23:56.160
<v Speaker 1>seems like a real, real fast way to get into

0:23:56.200 --> 0:23:59.840
<v Speaker 1>skynet territory, where it's like the robot then decides that

0:24:00.119 --> 0:24:02.600
<v Speaker 1>only way to assure that it never sells cigarettes to

0:24:02.680 --> 0:24:06.119
<v Speaker 1>children again is to destroy all humans. That sounds like

0:24:06.160 --> 0:24:09.400
<v Speaker 1>finding a wormhole to me. We will be getting into

0:24:09.440 --> 0:24:14.640
<v Speaker 1>some more wormhole territory as we go on, so more complications. Uh.

0:24:14.680 --> 0:24:16.639
<v Speaker 1>The authors bring up the idea of how to courts

0:24:16.720 --> 0:24:20.560
<v Speaker 1>compel a person or a company to obey a court order. Right, Like,

0:24:20.600 --> 0:24:23.720
<v Speaker 1>if you know a company is like dumping poison that's

0:24:23.760 --> 0:24:26.959
<v Speaker 1>harming somebody, and the person sues that company, what does

0:24:27.000 --> 0:24:29.240
<v Speaker 1>the court do to get them to stop? Well, the

0:24:29.240 --> 0:24:31.919
<v Speaker 1>there is a threat of contempt of court if they

0:24:31.960 --> 0:24:35.600
<v Speaker 1>don't stop doing it. Right, Courts usually just assume that

0:24:35.640 --> 0:24:38.720
<v Speaker 1>people are motivated by a desire not to pay huge

0:24:38.760 --> 0:24:41.560
<v Speaker 1>monetary damages or a desire not to go to jail.

0:24:42.080 --> 0:24:45.000
<v Speaker 1>Would that have any motivating power on a robot? It

0:24:45.359 --> 0:24:47.919
<v Speaker 1>would only have that power to the extent that the

0:24:48.040 --> 0:24:51.440
<v Speaker 1>robot had been programmed to take that into account. If

0:24:51.440 --> 0:24:53.520
<v Speaker 1>it hadn't, it wouldn't matter at all. Like, you know,

0:24:53.680 --> 0:24:56.440
<v Speaker 1>most robots probably do not have any opinion one way

0:24:56.520 --> 0:24:59.080
<v Speaker 1>or another about whether about going to jail or having

0:24:59.080 --> 0:25:02.320
<v Speaker 1>to pay damages. So you'd have to explicitly program it

0:25:02.400 --> 0:25:07.520
<v Speaker 1>to be disincentivized by potential punishments. Yeah, because take the

0:25:07.560 --> 0:25:10.840
<v Speaker 1>cigarette robot for example, Like it's prime it's prime directive

0:25:10.920 --> 0:25:14.639
<v Speaker 1>is just to sell delicious cigarettes to human beings, like

0:25:14.720 --> 0:25:17.080
<v Speaker 1>the the what what else? What kind of leverage do

0:25:17.080 --> 0:25:19.600
<v Speaker 1>you have? Right? Exactly? So in that case you'd be

0:25:19.640 --> 0:25:22.480
<v Speaker 1>faced with either you'd be trying to find some kind

0:25:22.520 --> 0:25:25.560
<v Speaker 1>of human who's responsible for its behavior, but you could

0:25:25.800 --> 0:25:28.640
<v Speaker 1>very well run into the problem that like, you can't

0:25:28.680 --> 0:25:31.800
<v Speaker 1>really identify any one person who seems to be at

0:25:31.800 --> 0:25:34.800
<v Speaker 1>fault for what it did, and it's doing this bad thing,

0:25:34.920 --> 0:25:37.120
<v Speaker 1>so so what are you going to do about it? Yeah,

0:25:37.480 --> 0:25:39.679
<v Speaker 1>And then of course things get even weirder when you

0:25:39.720 --> 0:25:42.160
<v Speaker 1>start getting into that that other side. You know, that's

0:25:42.200 --> 0:25:46.359
<v Speaker 1>like the more like direct and material remedies that can

0:25:46.400 --> 0:25:50.040
<v Speaker 1>be provided by courts, either like a monetary award to

0:25:50.160 --> 0:25:53.159
<v Speaker 1>the victim or in order to stop doing something that

0:25:53.200 --> 0:25:55.480
<v Speaker 1>causes harm. On the other hand, you've got this thing

0:25:55.560 --> 0:25:58.520
<v Speaker 1>that courts often end up engaging in, and people are

0:25:58.560 --> 0:26:02.800
<v Speaker 1>are largely driven in motivated by however however irrational it

0:26:02.880 --> 0:26:06.400
<v Speaker 1>might be in some cases, And that's the perceived abstract

0:26:06.520 --> 0:26:09.840
<v Speaker 1>value of punishment, you know, not just material damages to

0:26:09.880 --> 0:26:11.919
<v Speaker 1>a victim or in order not to do something, but

0:26:12.000 --> 0:26:16.920
<v Speaker 1>the inflicting of punishments, specifically to demonstrate the court's displeasure

0:26:16.960 --> 0:26:20.760
<v Speaker 1>with the original behavior of the defendant. Uh So, they

0:26:20.840 --> 0:26:23.199
<v Speaker 1>raise a question that's brought up in a paper by

0:26:23.240 --> 0:26:27.320
<v Speaker 1>a professor named Christina Mulligan, who explores the subject of

0:26:27.400 --> 0:26:29.679
<v Speaker 1>should you have the right to punch a robot that

0:26:29.760 --> 0:26:32.760
<v Speaker 1>hurts you limly in case? He called the call this

0:26:32.880 --> 0:26:37.320
<v Speaker 1>the expressive component of remedies, And though a desire to

0:26:37.480 --> 0:26:41.639
<v Speaker 1>see offenders punished maybe an extremely natural and nearly universal

0:26:41.720 --> 0:26:45.840
<v Speaker 1>human drive, it's debatable whether it actually serves a purpose

0:26:45.920 --> 0:26:49.320
<v Speaker 1>in reducing harm, and if it does, in what cases

0:26:49.359 --> 0:26:53.480
<v Speaker 1>it does. I love this idea because, in a very

0:26:53.560 --> 0:26:56.240
<v Speaker 1>literal level, it makes me think, well, why would you

0:26:56.240 --> 0:26:58.120
<v Speaker 1>punch a robot? They're made out of out of metal.

0:26:58.160 --> 0:26:59.879
<v Speaker 1>You're gonna hurt your hand. All you're gonna do is

0:27:00.040 --> 0:27:01.399
<v Speaker 1>or your hand, and you're not going to hurt the

0:27:01.480 --> 0:27:06.119
<v Speaker 1>robot unless first of all, you design the robots so

0:27:06.119 --> 0:27:09.520
<v Speaker 1>that it has at least one punchable portion of its anatomy,

0:27:10.000 --> 0:27:12.679
<v Speaker 1>and then for it to be more than just you know,

0:27:13.280 --> 0:27:17.600
<v Speaker 1>a cathartic uh uh thing for you, then you have

0:27:17.640 --> 0:27:19.880
<v Speaker 1>to also make sure there's some sort of feedback right

0:27:19.920 --> 0:27:25.119
<v Speaker 1>where Yeah, like you punch cigarette bought in it's punchable area,

0:27:25.720 --> 0:27:28.879
<v Speaker 1>then it will say owl. And maybe it will I

0:27:28.880 --> 0:27:31.760
<v Speaker 1>don't know, ottaw inciner rate one packet of cigarettes so

0:27:31.800 --> 0:27:33.520
<v Speaker 1>that it can never sell them, that sort of thing.

0:27:34.000 --> 0:27:37.040
<v Speaker 1>But then, yeah, you're having to design your robots to

0:27:37.240 --> 0:27:40.200
<v Speaker 1>to suffer to a certain extent, which I guess that

0:27:40.320 --> 0:27:42.800
<v Speaker 1>means that goes back to what C three PO said, Right,

0:27:43.240 --> 0:27:45.720
<v Speaker 1>he said, you know about about being made to suffer.

0:27:45.800 --> 0:27:48.359
<v Speaker 1>It seems to be our lot in life. Oh that's interesting.

0:27:48.400 --> 0:27:50.439
<v Speaker 1>I hadn't thought about that. Yeah, clearly R two, D

0:27:50.520 --> 0:27:53.720
<v Speaker 1>two and C three p O have have inherent desires

0:27:53.760 --> 0:27:57.119
<v Speaker 1>to avoid pain. They have been programmed with that. Yeah,

0:27:57.359 --> 0:28:00.640
<v Speaker 1>but as we've said that, that's not standard issue for robots.

0:28:00.680 --> 0:28:03.320
<v Speaker 1>Most robots don't care about whether or not they get injured,

0:28:03.440 --> 0:28:06.200
<v Speaker 1>Like that's not a motivating factor for them. And again

0:28:06.240 --> 0:28:09.280
<v Speaker 1>it raises this bizarre question of like, what are you

0:28:09.359 --> 0:28:12.040
<v Speaker 1>doing when you punch the robot? Like what is I

0:28:12.080 --> 0:28:15.000
<v Speaker 1>guess it's making you feel better, But does it make

0:28:15.040 --> 0:28:17.560
<v Speaker 1>you feel better if, like you know that the robot

0:28:17.600 --> 0:28:21.440
<v Speaker 1>doesn't actually care? Yeah, and then what then what needs

0:28:21.440 --> 0:28:24.120
<v Speaker 1>to be done to convince you that it does care. Yeah,

0:28:24.160 --> 0:28:26.520
<v Speaker 1>it just gets very sticky, very quickly, and then of

0:28:26.560 --> 0:28:29.479
<v Speaker 1>course turns the mirror back on the way we handle

0:28:30.320 --> 0:28:34.199
<v Speaker 1>human to human scenarios. Right. But anyway, Limbly and Casey,

0:28:34.320 --> 0:28:37.240
<v Speaker 1>I guess to to summarize their position, they say, Okay,

0:28:37.280 --> 0:28:42.600
<v Speaker 1>increasingly independent robots and AI are coming. They're they're infiltrating

0:28:42.640 --> 0:28:45.280
<v Speaker 1>more and more into society, and they will inevitably do

0:28:45.320 --> 0:28:49.240
<v Speaker 1>bad things. When that happens, the legal system will try

0:28:49.280 --> 0:28:52.560
<v Speaker 1>to order remedies to make things right, and you know

0:28:52.640 --> 0:28:56.560
<v Speaker 1>when when harm has been caused. Our current legal understanding

0:28:56.560 --> 0:28:59.880
<v Speaker 1>of remedies is based on the assumption of human agents

0:29:00.160 --> 0:29:03.720
<v Speaker 1>human agents only, and its rules are not suited to

0:29:03.800 --> 0:29:07.640
<v Speaker 1>dealing with robot crime or robot offenses quote. As we

0:29:07.720 --> 0:29:11.000
<v Speaker 1>have shown, Failing to recognize those differences could result in

0:29:11.120 --> 0:29:16.760
<v Speaker 1>significant unintended consequences, inadvertently encouraging the wrong behaviors, or even

0:29:16.840 --> 0:29:21.920
<v Speaker 1>rendering our most important remedial mechanisms functionally irrelevant. Uh So,

0:29:22.080 --> 0:29:24.960
<v Speaker 1>to take robot agents into account, we're going to have

0:29:25.000 --> 0:29:28.520
<v Speaker 1>to examine and rethink how our systems of remedies work.

0:29:29.040 --> 0:29:31.840
<v Speaker 1>But and this is a point we've been making already,

0:29:31.880 --> 0:29:34.440
<v Speaker 1>this could have multiple benefits because it could also lead

0:29:34.480 --> 0:29:37.200
<v Speaker 1>to a better understanding of how we apply these remedies

0:29:37.240 --> 0:29:41.360
<v Speaker 1>to cases dealing exclusively with humans. Quote. Indeed, one of

0:29:41.400 --> 0:29:44.720
<v Speaker 1>the most pressing challenges raised by the technology is its

0:29:44.720 --> 0:29:48.640
<v Speaker 1>tendency to reveal the trade offs between sidal economic and

0:29:48.720 --> 0:29:52.320
<v Speaker 1>legal values that many of us today make without deeply

0:29:52.360 --> 0:29:57.160
<v Speaker 1>appreciating the downstream consequences. They right, we need a law

0:29:57.200 --> 0:30:00.560
<v Speaker 1>of remedies for robots, but in the final analysis, remedies

0:30:00.600 --> 0:30:03.360
<v Speaker 1>for robots may also end up being remedies for all

0:30:03.400 --> 0:30:06.040
<v Speaker 1>of us. Now, like I said, this is a very

0:30:06.080 --> 0:30:08.240
<v Speaker 1>long paper. We can't do justice to all of the

0:30:08.280 --> 0:30:11.120
<v Speaker 1>subjects they raise, but to focus on some highlights, I

0:30:11.200 --> 0:30:14.440
<v Speaker 1>thought one interesting place to look was when they try

0:30:14.480 --> 0:30:16.840
<v Speaker 1>to get into the definition of what actually makes a

0:30:16.960 --> 0:30:20.360
<v Speaker 1>robot in the legal sense. Obviously, there's going to be

0:30:20.400 --> 0:30:24.400
<v Speaker 1>some difficulty here because think about how differently the term

0:30:24.520 --> 0:30:26.920
<v Speaker 1>is used and how many different things it's applied to

0:30:27.000 --> 0:30:30.480
<v Speaker 1>in the world. Uh. The authors here cite a professor

0:30:30.600 --> 0:30:34.120
<v Speaker 1>Ryan Callo, who in the past had written that there

0:30:34.160 --> 0:30:38.560
<v Speaker 1>are three important characteristics that define a robot and make

0:30:38.560 --> 0:30:42.120
<v Speaker 1>it different from any machine, just like a computer or phone,

0:30:42.720 --> 0:30:46.360
<v Speaker 1>and Callo says that these three uh, these three qualities

0:30:46.400 --> 0:30:51.920
<v Speaker 1>are embodiment, emergence, and social valence. So to quote from Calo,

0:30:52.200 --> 0:30:56.480
<v Speaker 1>robotics combines, arguably for the first time, the promiscuity of

0:30:56.560 --> 0:31:01.280
<v Speaker 1>information with the embodied capacity to do physical arm. Robots

0:31:01.320 --> 0:31:06.320
<v Speaker 1>display increasingly emergent behavior, permitting the technology to accomplish both

0:31:06.400 --> 0:31:10.320
<v Speaker 1>useful and unfortunate tasks in unexpected ways. I like that

0:31:10.400 --> 0:31:14.960
<v Speaker 1>idea of unfortunate tasks. Um and robots, more so than

0:31:15.000 --> 0:31:18.800
<v Speaker 1>any technology and history, feel to us like social actors,

0:31:18.920 --> 0:31:23.200
<v Speaker 1>a tendency so strong that soldiers sometimes jeopardize themselves to

0:31:23.360 --> 0:31:26.920
<v Speaker 1>preserve the lives of military robots in the field, and

0:31:27.080 --> 0:31:30.160
<v Speaker 1>lives is in quotes there. Yeah, you may remember this

0:31:30.280 --> 0:31:32.600
<v Speaker 1>from the film that came out a few years back,

0:31:32.920 --> 0:31:38.360
<v Speaker 1>Saving Private Cigarette Robot. It's quite touching. I mean, it

0:31:38.400 --> 0:31:40.440
<v Speaker 1>seems absurd, but it does seem to play on our

0:31:40.520 --> 0:31:43.440
<v Speaker 1>natural biases. I want to talk about a couple of

0:31:43.480 --> 0:31:46.880
<v Speaker 1>examples from a psychology paper in a second, but like, uh,

0:31:47.440 --> 0:31:50.680
<v Speaker 1>we're we're just so ready to look at machines like

0:31:50.800 --> 0:31:53.600
<v Speaker 1>humans and and treat them as such. It seems almost

0:31:53.600 --> 0:31:57.520
<v Speaker 1>impossible to avoid. But anyway to pick up with limely

0:31:57.600 --> 0:32:00.400
<v Speaker 1>and Casey after after that Callo quote they a quote.

0:32:00.440 --> 0:32:03.480
<v Speaker 1>In light of these qualities, Calo argues that robots are

0:32:03.520 --> 0:32:07.960
<v Speaker 1>best thought of as artificial objects or systems that sense, process,

0:32:08.040 --> 0:32:11.880
<v Speaker 1>and act upon the world to at least some degree. Thus,

0:32:11.920 --> 0:32:14.600
<v Speaker 1>a robot, in the strongest, fullest sense of the term,

0:32:14.920 --> 0:32:18.240
<v Speaker 1>exists in the world as a corporeal object with the

0:32:18.280 --> 0:32:22.960
<v Speaker 1>capacity to exert itself physically. Though, it's interesting to me

0:32:23.200 --> 0:32:26.640
<v Speaker 1>that even this attempt to give a strict and legally

0:32:26.720 --> 0:32:30.600
<v Speaker 1>useful definition of a robot includes a subjective component. I

0:32:30.880 --> 0:32:34.000
<v Speaker 1>brought this up earlier, the component about human feelings, the

0:32:34.240 --> 0:32:38.600
<v Speaker 1>social valence criterion. The Calos sites here this means they

0:32:38.720 --> 0:32:41.960
<v Speaker 1>feel to us like social actors. Yeah. Like I was

0:32:42.000 --> 0:32:45.480
<v Speaker 1>wondering in all of this, like, where does a particularly

0:32:45.680 --> 0:32:50.120
<v Speaker 1>malicious robo call fit into the scenario, Say, a robo

0:32:50.240 --> 0:32:53.280
<v Speaker 1>call that is not just about trying to sell you something,

0:32:53.320 --> 0:32:55.520
<v Speaker 1>but it's like, you know, actively trying to say, get

0:32:55.520 --> 0:32:59.040
<v Speaker 1>a credit card number out of you for nefarious purposes. Yeah,

0:32:59.080 --> 0:33:02.080
<v Speaker 1>that's a really good point. And and along those lines,

0:33:02.160 --> 0:33:05.520
<v Speaker 1>Limbly and Casey argue that actually they don't think the

0:33:05.600 --> 0:33:10.080
<v Speaker 1>embodiment criteria of hardware is necessarily a good one that

0:33:10.160 --> 0:33:13.440
<v Speaker 1>maybe our concept of a robot should be less limited

0:33:13.480 --> 0:33:17.000
<v Speaker 1>to the essentialist quality of being embodied and more just

0:33:17.200 --> 0:33:22.880
<v Speaker 1>applied to anything that exhibits intelligent behavior and exactly things

0:33:22.920 --> 0:33:25.760
<v Speaker 1>like that robot call would be would be a good example. Uh,

0:33:25.880 --> 0:33:28.240
<v Speaker 1>the things we think of as robots probably do. They

0:33:28.280 --> 0:33:31.840
<v Speaker 1>they're not just like stand alone objects. They interact with

0:33:31.920 --> 0:33:34.680
<v Speaker 1>the broader world in some way, but they could be

0:33:34.920 --> 0:33:39.080
<v Speaker 1>entirely software based. Yeah, I guess certainly the room is

0:33:39.120 --> 0:33:41.560
<v Speaker 1>a great example, or any kind of like vacuuming robot

0:33:41.640 --> 0:33:44.520
<v Speaker 1>where it's it's it's it's in your house or it's

0:33:44.520 --> 0:33:46.720
<v Speaker 1>in a room in your house. It's a it's interacting

0:33:46.720 --> 0:33:50.800
<v Speaker 1>in your environment and it's essentially making decisions about how

0:33:50.840 --> 0:33:54.040
<v Speaker 1>best to move around that space. Sure, but if you

0:33:54.080 --> 0:33:56.240
<v Speaker 1>want to take it out of the embodied space, you

0:33:56.280 --> 0:33:59.080
<v Speaker 1>could have the idea of bots. On the Internet, there's

0:33:59.120 --> 0:34:03.600
<v Speaker 1>things out there are acting autonomously to some extent and doing,

0:34:03.640 --> 0:34:07.320
<v Speaker 1>you know, executing some behavior, acting almost maliciously. We were

0:34:07.480 --> 0:34:11.320
<v Speaker 1>tempted to call them bots meaning short for robots, because

0:34:11.400 --> 0:34:15.080
<v Speaker 1>they have some kind of apparent independent agency and they're

0:34:15.120 --> 0:34:19.359
<v Speaker 1>doing something that seems at least halfway intelligent. Right, yeah, yeah,

0:34:19.400 --> 0:34:22.359
<v Speaker 1>and you can easily imagine how they could they are

0:34:22.400 --> 0:34:24.160
<v Speaker 1>and well they I mean they are used maliciously in

0:34:24.160 --> 0:34:27.480
<v Speaker 1>some cases. But how something like a social media boat

0:34:27.520 --> 0:34:30.200
<v Speaker 1>that responds to certain comments in a particular way, like,

0:34:30.239 --> 0:34:32.799
<v Speaker 1>it's very easy to imagine how you how that could

0:34:32.840 --> 0:34:36.400
<v Speaker 1>be utilized in a way that would be not only

0:34:36.880 --> 0:34:40.919
<v Speaker 1>annoying but just that outright harmful, even physically harmful. Oh yeah,

0:34:40.920 --> 0:34:43.200
<v Speaker 1>I mean think about some of these, say like bots

0:34:43.239 --> 0:34:47.400
<v Speaker 1>on social media that try to crowdsource information like during

0:34:47.400 --> 0:34:51.359
<v Speaker 1>a natural disaster or something like that. You could imagine uh,

0:34:51.400 --> 0:34:55.520
<v Speaker 1>intentionally maliciously manipulating a bot of this kind to like

0:34:55.560 --> 0:34:59.120
<v Speaker 1>have you know, bad information on it or something, yeah yeah,

0:34:59.239 --> 0:35:02.880
<v Speaker 1>or know, anything that a troll can do on social media,

0:35:03.120 --> 0:35:06.000
<v Speaker 1>a bought could conceivably do as well. So that just

0:35:06.280 --> 0:35:09.680
<v Speaker 1>you know, opens up the door, right. But coming back

0:35:09.680 --> 0:35:14.240
<v Speaker 1>to this, so there's this interesting idea that robots feel

0:35:14.320 --> 0:35:16.840
<v Speaker 1>to us like social actors, and that seems to be,

0:35:16.880 --> 0:35:20.560
<v Speaker 1>at least by some people's definitions, a kind of inextricable

0:35:21.000 --> 0:35:24.440
<v Speaker 1>quality of what makes a robot like it feels like,

0:35:24.600 --> 0:35:27.320
<v Speaker 1>at least to a small extent, like a person somehow,

0:35:27.719 --> 0:35:30.040
<v Speaker 1>and it reminds me of the psychology paper I was

0:35:30.080 --> 0:35:33.640
<v Speaker 1>looking at just recently on human social interaction with robots,

0:35:34.080 --> 0:35:38.120
<v Speaker 1>that is by Elizabeth Broadbent called Interactions with Robots The

0:35:38.160 --> 0:35:41.560
<v Speaker 1>Truths We Reveal About Ourselves, published in the Annual Review

0:35:41.560 --> 0:35:44.719
<v Speaker 1>of Psychology in twenty seventeen. Uh, this was a highly

0:35:44.840 --> 0:35:46.759
<v Speaker 1>cited paper, and it seems to be. It's a big

0:35:46.800 --> 0:35:50.360
<v Speaker 1>literature review of a lot of different stuff about about

0:35:50.480 --> 0:35:54.960
<v Speaker 1>how humans interact emotionally and socially with robots. And the

0:35:55.000 --> 0:35:57.320
<v Speaker 1>one section I was thinking about was where she reviews

0:35:57.360 --> 0:36:01.280
<v Speaker 1>a bunch of other studies about how we mindlessly apply

0:36:01.520 --> 0:36:04.440
<v Speaker 1>social rules to robots. So there are a ton of

0:36:04.440 --> 0:36:06.640
<v Speaker 1>different examples, but just to cite a couple of them,

0:36:07.320 --> 0:36:10.400
<v Speaker 1>once she writes up his quote. After using a computer,

0:36:10.600 --> 0:36:14.920
<v Speaker 1>people evaluate its performance more highly if the same computer

0:36:15.080 --> 0:36:19.040
<v Speaker 1>delivers the rating scale, then if another computer delivers the

0:36:19.160 --> 0:36:21.800
<v Speaker 1>rating scale, or if they rate it with pen and paper.

0:36:22.960 --> 0:36:25.279
<v Speaker 1>So like, if you know, you get a thing at

0:36:25.280 --> 0:36:27.400
<v Speaker 1>the end of a test that says like, hey, you know,

0:36:27.440 --> 0:36:30.520
<v Speaker 1>how did you enjoy interacting with this machine? You're you're

0:36:31.040 --> 0:36:33.759
<v Speaker 1>more likely to give it a higher score if you're

0:36:33.760 --> 0:36:35.840
<v Speaker 1>still sitting on the same machine or at least. That

0:36:35.920 --> 0:36:40.120
<v Speaker 1>was what was found by nas at All in UH

0:36:40.160 --> 0:36:43.840
<v Speaker 1>and the and Broadbent rights quote. This result is similar

0:36:43.880 --> 0:36:46.719
<v Speaker 1>to experiment or bias, in which people try not to

0:36:46.800 --> 0:36:52.160
<v Speaker 1>offend a human researcher. Another example of social behavior is reciprocity.

0:36:52.360 --> 0:36:55.760
<v Speaker 1>We help others who help us. People help to computer

0:36:55.840 --> 0:36:59.359
<v Speaker 1>with a task for more time and more accurately if

0:36:59.360 --> 0:37:02.399
<v Speaker 1>the computer first helped them with a task than if

0:37:02.400 --> 0:37:04.520
<v Speaker 1>it did not, And this was found by Fog and

0:37:04.640 --> 0:37:09.080
<v Speaker 1>nas In. I love that idea of people, you know,

0:37:09.200 --> 0:37:13.239
<v Speaker 1>being more reluctant to rate a computer UH poorly if

0:37:13.280 --> 0:37:16.640
<v Speaker 1>they're still interacting with the same computer that that's that

0:37:16.719 --> 0:37:19.960
<v Speaker 1>seems perfectly true to me. But another interesting one from

0:37:19.960 --> 0:37:23.840
<v Speaker 1>the summary is quote research and psychology has shown that

0:37:23.880 --> 0:37:27.360
<v Speaker 1>the presence of an observer can increase people's honesty, but

0:37:27.480 --> 0:37:30.640
<v Speaker 1>incentives for cheating can reduce honesty, and this is found

0:37:30.680 --> 0:37:33.680
<v Speaker 1>by Covey at All in nineteen eighty nine. In a

0:37:33.800 --> 0:37:37.440
<v Speaker 1>robot version of this work, participants given incentives to cheat

0:37:37.680 --> 0:37:40.960
<v Speaker 1>were shown to be less honest when alone compared to

0:37:41.320 --> 0:37:44.960
<v Speaker 1>when they were accompanied by either a human or by

0:37:44.960 --> 0:37:47.560
<v Speaker 1>a simple robot, and that was found by Hoffman at

0:37:47.560 --> 0:37:51.960
<v Speaker 1>All in this illustrates that the social presence of robots

0:37:52.000 --> 0:37:54.560
<v Speaker 1>may make people feel as though they're being watched and

0:37:54.719 --> 0:37:58.360
<v Speaker 1>increase their honesty in an effect similar to that produced

0:37:58.400 --> 0:38:01.120
<v Speaker 1>by the presence of humans. Now, this is inter right.

0:38:01.200 --> 0:38:03.560
<v Speaker 1>This This also reminds me of various studies that have

0:38:03.680 --> 0:38:08.680
<v Speaker 1>gone into sort of the idea of of imagine beings

0:38:08.920 --> 0:38:13.160
<v Speaker 1>or religious beings watching us while we're doing things right,

0:38:13.360 --> 0:38:16.160
<v Speaker 1>or even just like I imagery, like putting some eyes

0:38:16.400 --> 0:38:19.160
<v Speaker 1>imagery on a wall looking at people while they're like,

0:38:19.200 --> 0:38:21.240
<v Speaker 1>I don't know, not supposed to steal from the collection

0:38:21.280 --> 0:38:23.319
<v Speaker 1>plate or something like that. I don't know if it's

0:38:23.320 --> 0:38:25.719
<v Speaker 1>to the same extent, but at least in the same

0:38:25.719 --> 0:38:29.120
<v Speaker 1>direction that the presence of another human is. You know,

0:38:28.920 --> 0:38:31.040
<v Speaker 1>you're you might be a little bit worried that ARE

0:38:31.080 --> 0:38:33.440
<v Speaker 1>two D two is gonna, you know, judge your moral

0:38:33.560 --> 0:38:36.680
<v Speaker 1>character harshly or tattle on you. I'm not as worried

0:38:36.680 --> 0:38:47.799
<v Speaker 1>about R two, but um three po snitch. Coming back

0:38:47.880 --> 0:38:50.359
<v Speaker 1>to Limele and Casey, so they talked for a long

0:38:50.360 --> 0:38:53.839
<v Speaker 1>time about how robots get their intelligence. They talked about

0:38:53.840 --> 0:38:57.360
<v Speaker 1>the importance of machine learning for the modern generations of

0:38:57.560 --> 0:39:01.000
<v Speaker 1>robots and AI that it's just not practical to hard

0:39:01.160 --> 0:39:03.440
<v Speaker 1>code AI the way we used to imagine. You know,

0:39:03.480 --> 0:39:05.640
<v Speaker 1>you'd be a programmer and you're just like creating a

0:39:05.640 --> 0:39:08.719
<v Speaker 1>lot of strings of if then statements like you know,

0:39:09.120 --> 0:39:12.000
<v Speaker 1>the kind of intelligence that we expect from a modern

0:39:12.040 --> 0:39:15.720
<v Speaker 1>AI or or intelligent robot is too complex for people

0:39:15.719 --> 0:39:18.480
<v Speaker 1>to program in a in a direct way like that. Instead,

0:39:18.520 --> 0:39:21.160
<v Speaker 1>they've got to be trained on natural data sets through

0:39:21.200 --> 0:39:24.319
<v Speaker 1>machine learning. But of course doing so comes at the

0:39:24.400 --> 0:39:29.799
<v Speaker 1>cost of increasing uncertainty about their future behaviors. Behaviors could

0:39:29.800 --> 0:39:34.080
<v Speaker 1>emerge that a conscientious programmer would never intentionally hard code

0:39:34.080 --> 0:39:37.600
<v Speaker 1>into the system. Uh So, so that brings us to like,

0:39:37.719 --> 0:39:42.560
<v Speaker 1>what types of harms could we expect from robots and AI?

0:39:42.680 --> 0:39:44.400
<v Speaker 1>And the authors here come up with what I think

0:39:44.440 --> 0:39:47.600
<v Speaker 1>are some very useful categories, some sort of like cubby holes,

0:39:47.640 --> 0:39:51.120
<v Speaker 1>to slot the different types of AI fears into. So

0:39:51.160 --> 0:39:54.040
<v Speaker 1>the first kind is what they call unavoidable harms. These

0:39:54.040 --> 0:39:56.160
<v Speaker 1>are probably not the main ones to be worried about,

0:39:56.160 --> 0:39:59.120
<v Speaker 1>but they are worth thinking about. Uh And this is

0:39:59.160 --> 0:40:01.919
<v Speaker 1>just the fact that some dangers are inherent too many

0:40:01.960 --> 0:40:05.160
<v Speaker 1>products and services, we just accept them as the cost

0:40:05.239 --> 0:40:07.960
<v Speaker 1>of having those products and services in the first place.

0:40:08.360 --> 0:40:11.480
<v Speaker 1>So like this would just be cigarette bought just by

0:40:11.560 --> 0:40:14.640
<v Speaker 1>virtue of selling cigarettes is doing harm to people, right, Yes,

0:40:14.800 --> 0:40:17.480
<v Speaker 1>I mean the fact that you have cigarettes, there is

0:40:17.520 --> 0:40:20.320
<v Speaker 1>some harm coming from that. But there are also ones

0:40:20.400 --> 0:40:23.960
<v Speaker 1>that are more fully integrated into just the way society works,

0:40:24.000 --> 0:40:28.360
<v Speaker 1>like having cars. It is absolutely inevitable that people driving

0:40:28.360 --> 0:40:30.719
<v Speaker 1>cars are going to crash their cars and there will

0:40:30.760 --> 0:40:33.120
<v Speaker 1>be fatalities from that, and you can think of ways

0:40:33.120 --> 0:40:36.680
<v Speaker 1>of reducing it, but there's there's really not any expectation

0:40:37.320 --> 0:40:40.080
<v Speaker 1>that we can have a country that has car based

0:40:40.080 --> 0:40:42.840
<v Speaker 1>transportation and there will not be any accidents because there

0:40:42.960 --> 0:40:46.239
<v Speaker 1>always be things that are that are not even reducible

0:40:46.280 --> 0:40:48.880
<v Speaker 1>to driver error or to malfunction of the cars, right,

0:40:48.920 --> 0:40:51.800
<v Speaker 1>like a tree falls on the road or something, birds,

0:40:51.800 --> 0:40:54.160
<v Speaker 1>wild animals. Any So, even though they're I think there's

0:40:54.160 --> 0:40:58.680
<v Speaker 1>some very convincing arguments to be made that uh a

0:40:58.760 --> 0:41:02.520
<v Speaker 1>switch to self driving cars would create a much safer

0:41:03.080 --> 0:41:06.439
<v Speaker 1>uh travel environment, that it would make roads safer. You're

0:41:06.440 --> 0:41:09.920
<v Speaker 1>not gonna you're not gonna get to absolute zero crashes

0:41:10.000 --> 0:41:13.800
<v Speaker 1>or absolute zero road fatalities, right, I mean, you wouldn't

0:41:13.840 --> 0:41:17.480
<v Speaker 1>even if the driving algorithms were perfect, right, and they're

0:41:17.480 --> 0:41:19.439
<v Speaker 1>probably not going to be perfect. They may well be

0:41:19.520 --> 0:41:21.720
<v Speaker 1>and probably are going to be better than the average

0:41:21.760 --> 0:41:25.879
<v Speaker 1>human driver. Yeah, okay, so that's just there's just unavoidable

0:41:25.920 --> 0:41:28.880
<v Speaker 1>harm that comes from using any type of product or service,

0:41:29.000 --> 0:41:31.680
<v Speaker 1>and when you integrate robotics and AI into that product

0:41:31.760 --> 0:41:34.799
<v Speaker 1>or service, those unavoidable harms will just continue. But that's

0:41:34.800 --> 0:41:38.640
<v Speaker 1>something we already deal with. The next category is deliberate

0:41:38.760 --> 0:41:42.719
<v Speaker 1>least cost harms. This is similar to unavoidable harms, but

0:41:42.760 --> 0:41:45.480
<v Speaker 1>it's in cases where the machine actually is able to

0:41:45.680 --> 0:41:49.960
<v Speaker 1>make a decision with with important ramifications, like it can

0:41:50.000 --> 0:41:53.360
<v Speaker 1>make a decision to act in a way that causes harm,

0:41:53.560 --> 0:41:56.880
<v Speaker 1>but is attempting to cause the least harm possible. So

0:41:56.920 --> 0:41:59.480
<v Speaker 1>in a sense, this is forcing robots to do the

0:41:59.520 --> 0:42:02.719
<v Speaker 1>trolley problem. Right, do you switch to the track that

0:42:02.840 --> 0:42:05.279
<v Speaker 1>has one person sitting on the train tracks instead of

0:42:05.280 --> 0:42:09.600
<v Speaker 1>five people? Yeah? Yeah, And this will be another inevitable

0:42:09.640 --> 0:42:12.799
<v Speaker 1>capability of autonomous cars, but it raises all kinds of

0:42:12.840 --> 0:42:17.160
<v Speaker 1>thorny questions. If an autonomous vehicle can avoid a head

0:42:17.200 --> 0:42:20.760
<v Speaker 1>on collision that will likely kill multiple people by suddenly

0:42:20.760 --> 0:42:24.239
<v Speaker 1>swerving out of the way and hitting one pedestrian, that

0:42:24.320 --> 0:42:28.080
<v Speaker 1>may indeed avoid a greater harm. But that's probably cold

0:42:28.080 --> 0:42:31.400
<v Speaker 1>comfort to the one person who got hit, right, right. Yeah,

0:42:31.440 --> 0:42:34.520
<v Speaker 1>And then when you have a robot or some sort

0:42:34.520 --> 0:42:37.719
<v Speaker 1>of an AI involved in that decision making, I mean,

0:42:38.160 --> 0:42:43.480
<v Speaker 1>it's it's you can just imagine the the the intensity

0:42:43.520 --> 0:42:47.120
<v Speaker 1>of the arguments and the conversations they would ensue. Right.

0:42:47.239 --> 0:42:48.920
<v Speaker 1>But then the authors raised what I think is a

0:42:49.000 --> 0:42:52.040
<v Speaker 1>very interesting point. They say that this kind of life

0:42:52.120 --> 0:42:55.680
<v Speaker 1>or death trolley problem will probably be the exception rather

0:42:55.760 --> 0:42:59.360
<v Speaker 1>than the rule. Instead, they say, quote, uh, far likelier,

0:42:59.440 --> 0:43:04.040
<v Speaker 1>I'll build albeit subtler scenarios involving least cost harms will

0:43:04.080 --> 0:43:08.560
<v Speaker 1>involve robots that make decisions with seemingly trivial implications at

0:43:08.600 --> 0:43:12.840
<v Speaker 1>an individual level, but which result in non trivial impacts

0:43:13.000 --> 0:43:16.959
<v Speaker 1>at scale. Self driving cars, for example, will rarely face

0:43:17.040 --> 0:43:20.160
<v Speaker 1>a stark choice between killing a child or killing two

0:43:20.200 --> 0:43:23.719
<v Speaker 1>elderly people, but thousands of times a day they will

0:43:23.760 --> 0:43:27.880
<v Speaker 1>have to choose precisely where to change lanes, how closely

0:43:27.960 --> 0:43:31.280
<v Speaker 1>to trail another vehicle, when to accelerate on a freeway,

0:43:31.320 --> 0:43:34.640
<v Speaker 1>on ramp, and so forth. Each of these decisions will

0:43:34.800 --> 0:43:39.759
<v Speaker 1>entail some probability of injuring someone. I guess another thing

0:43:39.760 --> 0:43:41.279
<v Speaker 1>to keep in mind, like with the trap, with the

0:43:41.320 --> 0:43:45.400
<v Speaker 1>trolley problem, generally, when you're dealing with it, there's a

0:43:45.440 --> 0:43:47.840
<v Speaker 1>there's a lot of emphasis on the problem aspect of it,

0:43:47.920 --> 0:43:51.560
<v Speaker 1>you know, like the trolley problem should be a an

0:43:51.560 --> 0:43:54.400
<v Speaker 1>ethical dilemma. It should, it should hurt a bit to

0:43:54.440 --> 0:43:57.480
<v Speaker 1>try and figure out how what to do. And the

0:43:57.520 --> 0:44:00.759
<v Speaker 1>idea of the trolley problem being something that um is

0:44:00.840 --> 0:44:04.000
<v Speaker 1>encountered and decided upon, like as in a split second,

0:44:04.200 --> 0:44:07.799
<v Speaker 1>by a machine, um, by an algorithm like that, that

0:44:07.880 --> 0:44:10.840
<v Speaker 1>feels that that feels a bit worse to us. You know,

0:44:10.920 --> 0:44:14.400
<v Speaker 1>that feels like if if it's if it's an easy decision,

0:44:14.480 --> 0:44:16.520
<v Speaker 1>even if it's just based purely on math, you know,

0:44:16.880 --> 0:44:20.600
<v Speaker 1>it's um it feels wrong on some level. Oh yeah, yeah,

0:44:20.880 --> 0:44:23.759
<v Speaker 1>so I think you're right. But also the thing they're

0:44:23.760 --> 0:44:26.520
<v Speaker 1>bringing up here is that the trolley problem you're actually

0:44:26.560 --> 0:44:29.719
<v Speaker 1>more often facing is that every single day you're your

0:44:29.719 --> 0:44:32.680
<v Speaker 1>autonomous car is gonna make you know, hundreds or thousands

0:44:32.680 --> 0:44:36.279
<v Speaker 1>of trolley problem calls where on one track it is

0:44:36.920 --> 0:44:41.160
<v Speaker 1>getting to your destination a few seconds faster, and on

0:44:41.200 --> 0:44:43.399
<v Speaker 1>the other track is a one in a million chance

0:44:43.440 --> 0:44:46.440
<v Speaker 1>of killing somebody. Yeah yeah, And these do we make

0:44:46.480 --> 0:44:48.680
<v Speaker 1>these decisions all the time, but we don't focus on

0:44:48.719 --> 0:44:50.320
<v Speaker 1>these That's but I think that's part of the issue,

0:44:50.360 --> 0:44:53.160
<v Speaker 1>you know exactly. You're like, should I, okay, should I

0:44:53.239 --> 0:44:56.320
<v Speaker 1>take a left on this road? Well, there's a chance

0:44:56.360 --> 0:44:59.399
<v Speaker 1>there's a speeding car just above, just over the edge there,

0:44:59.400 --> 0:45:01.440
<v Speaker 1>and I can't it, but I'm going to take that

0:45:01.520 --> 0:45:03.799
<v Speaker 1>chance because I want to cut three minutes off my

0:45:03.880 --> 0:45:06.640
<v Speaker 1>drive to work. Yes. Uh, this is actually a very

0:45:06.640 --> 0:45:09.839
<v Speaker 1>good point that we we already make these decisions, but

0:45:09.920 --> 0:45:14.240
<v Speaker 1>we just don't think about them in these explicit probability calculations.

0:45:14.640 --> 0:45:17.359
<v Speaker 1>And there may be some consequences to thinking about them

0:45:17.360 --> 0:45:19.000
<v Speaker 1>this way, which is there could be a weird, like

0:45:19.080 --> 0:45:22.799
<v Speaker 1>perceived downside just to making these kind of uh, these

0:45:22.840 --> 0:45:26.480
<v Speaker 1>kind of calculations objective and and explicit. Yeah, I mean

0:45:26.480 --> 0:45:28.479
<v Speaker 1>I've I've run into this with some of the map

0:45:28.640 --> 0:45:31.439
<v Speaker 1>programs that I used to drive before, where I want

0:45:31.480 --> 0:45:33.440
<v Speaker 1>to tell it in some cases like like give me

0:45:33.480 --> 0:45:36.560
<v Speaker 1>the ability and maybe they have this now, but there

0:45:36.600 --> 0:45:39.720
<v Speaker 1>was one left turn in particular where the ability to

0:45:39.719 --> 0:45:42.839
<v Speaker 1>to flag this left turn, this is a dangerous left turn.

0:45:43.280 --> 0:45:46.320
<v Speaker 1>You have put me in a position to make. Um,

0:45:46.360 --> 0:45:48.560
<v Speaker 1>I might know which left turn you're talking about. If

0:45:48.600 --> 0:45:51.399
<v Speaker 1>you probably in town, it's it's it's in town, it's

0:45:51.440 --> 0:45:54.600
<v Speaker 1>near our office, so yeah, I mean yeah. By the way,

0:45:54.640 --> 0:45:57.120
<v Speaker 1>if you're out there working on programming driving apps, you

0:45:57.320 --> 0:46:00.799
<v Speaker 1>should absolutely include the toggle key where you can say

0:46:00.840 --> 0:46:04.080
<v Speaker 1>no left turns please. Yes, that that is highly useful.

0:46:04.360 --> 0:46:06.680
<v Speaker 1>I'm to understand this is how one of my aunts

0:46:07.000 --> 0:46:09.160
<v Speaker 1>got around, Like if they got older and they were

0:46:09.239 --> 0:46:11.960
<v Speaker 1>less adventurous driving, they would only take right turns, and

0:46:12.000 --> 0:46:14.080
<v Speaker 1>they would do all their driving so that no left

0:46:14.080 --> 0:46:16.520
<v Speaker 1>turns were made. I think I have one time read

0:46:16.760 --> 0:46:19.879
<v Speaker 1>this could be totally wrong, but I at least one

0:46:19.880 --> 0:46:23.160
<v Speaker 1>time I remember reading a claim that, like, you know,

0:46:23.480 --> 0:46:26.640
<v Speaker 1>the traffic efficiency would be x percent higher and people

0:46:26.640 --> 0:46:29.840
<v Speaker 1>would spend x number of minutes less time and traffic

0:46:29.920 --> 0:46:31.960
<v Speaker 1>if there were no such thing as left turns, if

0:46:32.000 --> 0:46:34.640
<v Speaker 1>everybody had to get everywhere by only doing you know,

0:46:34.760 --> 0:46:38.320
<v Speaker 1>full right turns to to go around the block. Interesting,

0:46:38.560 --> 0:46:40.439
<v Speaker 1>I'm sure there would be some cases where you can't

0:46:40.440 --> 0:46:42.480
<v Speaker 1>do that, but you know, in a in a grid city,

0:46:42.840 --> 0:46:44.719
<v Speaker 1>seems to make a lot of sense. Maybe you get

0:46:44.719 --> 0:46:47.239
<v Speaker 1>like one left turn of days, some sort of a

0:46:47.520 --> 0:46:51.000
<v Speaker 1>card system. But like I said, I I cannot confirm that. Okay,

0:46:51.040 --> 0:46:53.480
<v Speaker 1>but anyway, the next categories of harm they talked about

0:46:53.680 --> 0:46:56.880
<v Speaker 1>this one is defect driven harms. Uh. This one is

0:46:57.000 --> 0:46:59.799
<v Speaker 1>very easy to understand. The robot harms someone because of

0:46:59.840 --> 0:47:02.759
<v Speaker 1>a design flaw or a bug or a mistake, or

0:47:02.840 --> 0:47:06.520
<v Speaker 1>it's just broken. You know, A warehouse loading robot is

0:47:06.600 --> 0:47:09.759
<v Speaker 1>designed to only operate when no humans are nearby it.

0:47:10.000 --> 0:47:12.560
<v Speaker 1>But there's a malfunction with one of its sensors and

0:47:12.640 --> 0:47:15.400
<v Speaker 1>it fails to detect the presence of a human operator

0:47:15.960 --> 0:47:17.560
<v Speaker 1>trying to get I don't know, a piece of junk

0:47:17.600 --> 0:47:19.799
<v Speaker 1>out of it, out of one of its hinges, and

0:47:19.880 --> 0:47:23.320
<v Speaker 1>it moves and kills them. Okay, this is pretty straightforward,

0:47:23.320 --> 0:47:26.040
<v Speaker 1>just it's broken for some reason. The authors here do

0:47:26.120 --> 0:47:28.399
<v Speaker 1>point out that this gets even more complicated when there

0:47:28.480 --> 0:47:32.200
<v Speaker 1>is a human in the loop e g. An autonomous

0:47:32.239 --> 0:47:35.640
<v Speaker 1>car with a human driver who is supposed to intervene

0:47:35.760 --> 0:47:37.840
<v Speaker 1>in the event of an emergency. They talked about one

0:47:37.920 --> 0:47:40.800
<v Speaker 1>case where this happened with with I believe it was

0:47:40.840 --> 0:47:45.040
<v Speaker 1>an uber autonomous vehicle where both the machine and the

0:47:45.160 --> 0:47:49.120
<v Speaker 1>human fail, that both of them failed to stop a

0:47:49.120 --> 0:47:52.719
<v Speaker 1>collision that hurts someone, like what happens here. Yeah, yeah,

0:47:52.960 --> 0:47:55.640
<v Speaker 1>of course, we we have very similar cases in just

0:47:55.719 --> 0:47:58.719
<v Speaker 1>purely human affairs. Right when questions are asked like where

0:47:58.760 --> 0:48:01.600
<v Speaker 1>was this person's supervisor? Uh wait, you know who? Who

0:48:01.600 --> 0:48:04.080
<v Speaker 1>were the watchers? Who? There should have been some other person,

0:48:04.320 --> 0:48:06.439
<v Speaker 1>There was someone else in the loop here. Why didn't

0:48:06.480 --> 0:48:10.240
<v Speaker 1>they do something to stop this crime from taking place? Right? Okay,

0:48:10.239 --> 0:48:13.480
<v Speaker 1>After that you get into misuse harms. Now, some of

0:48:13.480 --> 0:48:16.839
<v Speaker 1>these are very obvious, very straightforward, like if you program

0:48:16.920 --> 0:48:19.759
<v Speaker 1>a robot directly to go kill someone, or even if

0:48:19.800 --> 0:48:23.800
<v Speaker 1>you program it to wander around at random swinging a machete.

0:48:23.840 --> 0:48:26.840
<v Speaker 1>In these cases, it seems that the human programmer is

0:48:26.840 --> 0:48:29.960
<v Speaker 1>clearly at fault, right, the robot has just become a

0:48:30.000 --> 0:48:33.360
<v Speaker 1>weapon of murder or of reckless endangerment, and the person

0:48:33.400 --> 0:48:36.560
<v Speaker 1>who told it to do that is the person responsible. Yeah.

0:48:36.600 --> 0:48:41.240
<v Speaker 1>Like if you take an automotive, uh like oil change

0:48:41.320 --> 0:48:46.840
<v Speaker 1>robot and you reprogram it to um do appendectomies and

0:48:46.880 --> 0:48:49.120
<v Speaker 1>people die as a result, Like, that's a misuse. You

0:48:49.120 --> 0:48:52.400
<v Speaker 1>can you can only blame the the oil changed robots

0:48:52.440 --> 0:48:57.279
<v Speaker 1>so much because it was not ultimately designed to perform appendectomies. Right.

0:48:57.320 --> 0:48:59.480
<v Speaker 1>In this case, this is more like the hammer example

0:48:59.640 --> 0:49:02.840
<v Speaker 1>used to the beginning this it's not the robot autonomously

0:49:03.000 --> 0:49:05.880
<v Speaker 1>making the decision to do this. Uh, this is somebody

0:49:05.960 --> 0:49:10.080
<v Speaker 1>just using it as a tool of crime. But the

0:49:10.080 --> 0:49:13.560
<v Speaker 1>authors point out that there are cases where quote, people

0:49:13.560 --> 0:49:17.120
<v Speaker 1>will misuse robots in a manner that is neither negligent

0:49:17.200 --> 0:49:21.600
<v Speaker 1>nor criminal, but nevertheless threatens to harm others. And these

0:49:21.640 --> 0:49:25.440
<v Speaker 1>types of harm are especially difficult to predict and prevent.

0:49:26.280 --> 0:49:29.560
<v Speaker 1>So one example is just people love to trick robots,

0:49:29.600 --> 0:49:32.759
<v Speaker 1>people like to to mess around with robots in AI.

0:49:33.120 --> 0:49:36.440
<v Speaker 1>I would admit to myself finding this amusing and principle,

0:49:36.480 --> 0:49:38.279
<v Speaker 1>we've talked about this in uh, you know, the the

0:49:38.280 --> 0:49:40.960
<v Speaker 1>flated sex make Ina episodes. But of course there are

0:49:41.040 --> 0:49:43.600
<v Speaker 1>times when it's not so funny, when when people take

0:49:43.640 --> 0:49:47.239
<v Speaker 1>it to really sinister places. One example the authors bring

0:49:47.360 --> 0:49:50.759
<v Speaker 1>up here is the horrible saga of Microsoft Tay. Do

0:49:50.840 --> 0:49:54.040
<v Speaker 1>you remember this thing? Oh? This was the this is

0:49:54.040 --> 0:49:56.680
<v Speaker 1>the robot that was traveling across the country. No no,

0:49:56.680 --> 0:49:59.440
<v Speaker 1>no, no no, uh though I know what you're talking about there. No,

0:50:00.000 --> 0:50:02.279
<v Speaker 1>maybe we can come back to that. But Tay was

0:50:02.440 --> 0:50:07.640
<v Speaker 1>a Twitter chat bot created by Microsoft that was supposed

0:50:07.680 --> 0:50:11.160
<v Speaker 1>to learn how to interact on the Internet. Just by

0:50:11.239 --> 0:50:13.920
<v Speaker 1>learning from conversations it had with real users. So you

0:50:13.920 --> 0:50:16.239
<v Speaker 1>could tweet it Tay and say hey, how are you doing,

0:50:16.440 --> 0:50:18.640
<v Speaker 1>you know, and you could talk about the weather or whatever.

0:50:18.680 --> 0:50:22.880
<v Speaker 1>But of course who who ended up engaging and training

0:50:22.920 --> 0:50:25.600
<v Speaker 1>this AI to speak? It was like the worst trolls

0:50:25.600 --> 0:50:28.239
<v Speaker 1>on the Internet. So within a matter of hours, this

0:50:28.320 --> 0:50:31.919
<v Speaker 1>brand new chat bot had been transformed from from a

0:50:32.000 --> 0:50:34.439
<v Speaker 1>from a you know, a lump of clay unformed into

0:50:34.520 --> 0:50:39.400
<v Speaker 1>a pornographic nazi. Yes, I do remember this now. And

0:50:39.480 --> 0:50:41.760
<v Speaker 1>this kind of just gets you thinking about the ways

0:50:41.840 --> 0:50:45.040
<v Speaker 1>that people will be will be able to misuse robots

0:50:45.080 --> 0:50:49.680
<v Speaker 1>in ways that guide their behavior in extremely pernicious directions,

0:50:49.800 --> 0:50:54.560
<v Speaker 1>sometimes without the people guiding this misuse necessarily committing any

0:50:54.600 --> 0:50:58.439
<v Speaker 1>kind of identifiable crime. Like people are going to look

0:50:58.480 --> 0:51:01.359
<v Speaker 1>for exploits, they're going to look for ways, they're gonna

0:51:01.360 --> 0:51:03.200
<v Speaker 1>look for cracks in the system. It's it's like within

0:51:03.480 --> 0:51:05.799
<v Speaker 1>with any kind of like a video game system. You know,

0:51:05.800 --> 0:51:07.399
<v Speaker 1>people are just gonna see what they can get away

0:51:07.400 --> 0:51:10.240
<v Speaker 1>with and and just engage in that kind of action,

0:51:10.680 --> 0:51:12.920
<v Speaker 1>sometimes just for the fun of it, right, And sometimes

0:51:12.960 --> 0:51:17.280
<v Speaker 1>that's harmless, but sometimes that's really awful. Yeah, Okay, next,

0:51:17.480 --> 0:51:21.360
<v Speaker 1>category is unforeseen harms. And here's where we start getting

0:51:21.360 --> 0:51:25.240
<v Speaker 1>into the really the really interesting and really difficult cases,

0:51:25.760 --> 0:51:29.840
<v Speaker 1>types of harm that are not unavoidable, not a product

0:51:29.840 --> 0:51:35.239
<v Speaker 1>of defects or misuse, but are still not predicted by creators. Uh.

0:51:35.239 --> 0:51:37.359
<v Speaker 1>And so the authors talk about how, in a way,

0:51:37.560 --> 0:51:42.200
<v Speaker 1>unpredictability is what makes AI potentially useful, right, Like, it

0:51:42.280 --> 0:51:46.719
<v Speaker 1>can potentially arrive at solutions that humans wouldn't have predicted,

0:51:47.200 --> 0:51:50.280
<v Speaker 1>but sometimes it does so in ways that really miss

0:51:50.360 --> 0:51:53.240
<v Speaker 1>the boat and could be extremely harmful if they were

0:51:53.280 --> 0:51:57.000
<v Speaker 1>embodied in action in the real world. Uh. Similar to

0:51:57.040 --> 0:51:59.640
<v Speaker 1>the drone example from the circle that we talked about

0:51:59.680 --> 0:52:03.480
<v Speaker 1>at the inning, But they signed another fantastic example here

0:52:03.520 --> 0:52:05.759
<v Speaker 1>that's kind of chilling. So I'm just going to read

0:52:05.800 --> 0:52:08.840
<v Speaker 1>from Lemle and Casey here. In the ninet nineties, a

0:52:08.960 --> 0:52:13.440
<v Speaker 1>pioneering multi institutional study sought to use machine learning techniques

0:52:13.480 --> 0:52:19.000
<v Speaker 1>to predict health related risks prior to hospitalization. After ingesting

0:52:19.000 --> 0:52:23.160
<v Speaker 1>an enormous quantity of data covering patients with pneumonia, the

0:52:23.239 --> 0:52:29.160
<v Speaker 1>system learned the rule has asthma X delivers lower risk X.

0:52:29.680 --> 0:52:33.480
<v Speaker 1>The colloquial translation is patients with pneumonia who have a

0:52:33.600 --> 0:52:36.839
<v Speaker 1>history of asthma have a lower risk of dying from

0:52:36.880 --> 0:52:41.480
<v Speaker 1>pneumonia than the general population. The machine derived rule was curious,

0:52:41.520 --> 0:52:44.600
<v Speaker 1>to say the least. Far from being protective, asthma can

0:52:44.680 --> 0:52:50.920
<v Speaker 1>seriously complicate pulmonary illnesses, including pneumonia. Perplexed by this counterintuitive result,

0:52:51.000 --> 0:52:54.360
<v Speaker 1>the researchers dug deeper, and what they found was troubling.

0:52:54.960 --> 0:52:58.320
<v Speaker 1>They discovered that quote patients with the history of asthma

0:52:58.360 --> 0:53:01.799
<v Speaker 1>who presented with pneumonia usually were admitted not only to

0:53:01.840 --> 0:53:05.000
<v Speaker 1>the hospital, but directly to the i c U, the

0:53:05.040 --> 0:53:08.800
<v Speaker 1>intensive care unit. Once in the i c U, asthmatic

0:53:08.800 --> 0:53:12.800
<v Speaker 1>pneumonia patients went on to receive more aggressive care, thereby

0:53:12.960 --> 0:53:18.080
<v Speaker 1>raising their survival rates compared to the general population. The rule,

0:53:18.160 --> 0:53:21.480
<v Speaker 1>in other words, reflected a genuine pattern in the data,

0:53:21.880 --> 0:53:26.640
<v Speaker 1>but the machine had confused correlation with causation quote, incorrectly

0:53:26.760 --> 0:53:30.520
<v Speaker 1>learning that asthma lowers risk when in fact, asthmatics have

0:53:30.719 --> 0:53:34.080
<v Speaker 1>much higher risk. It seems like we've got another wormhole here.

0:53:35.360 --> 0:53:38.759
<v Speaker 1>And here the authors introduce an idea of of a

0:53:38.800 --> 0:53:43.919
<v Speaker 1>curve of outcomes that they call a leptokurtic curve. That's

0:53:43.920 --> 0:53:46.400
<v Speaker 1>a strange term, but basically what that means is if

0:53:46.520 --> 0:53:50.120
<v Speaker 1>you are UM, if you're charting what types of outcomes

0:53:50.120 --> 0:53:54.040
<v Speaker 1>you expect from a traditional system like just you know,

0:53:54.120 --> 0:53:59.360
<v Speaker 1>humans looking at data versus a a complex automated system.

0:53:59.800 --> 0:54:02.760
<v Speaker 1>The uh, the sort of the tails of the graph

0:54:03.000 --> 0:54:06.200
<v Speaker 1>with the complex automated system will tend to be fatter,

0:54:06.320 --> 0:54:09.600
<v Speaker 1>meaning you get more extreme events in the positive and

0:54:09.640 --> 0:54:13.279
<v Speaker 1>negative space rather than a you know, a sort of

0:54:13.360 --> 0:54:17.600
<v Speaker 1>rounder clustering of events in the you know, normal operation space,

0:54:17.719 --> 0:54:21.880
<v Speaker 1>if that makes any sense. So, these kinds of unforeseen

0:54:21.960 --> 0:54:25.120
<v Speaker 1>harms are some of the most worrisome types of things

0:54:25.160 --> 0:54:27.480
<v Speaker 1>to expect coming out of robots and AI. But then

0:54:27.520 --> 0:54:30.680
<v Speaker 1>the other one would be systemic harms and this is

0:54:30.719 --> 0:54:34.040
<v Speaker 1>the last category of of harms they talk about. Uh

0:54:34.080 --> 0:54:37.200
<v Speaker 1>the author's right quote. People have long assumed that robots

0:54:37.239 --> 0:54:41.640
<v Speaker 1>are inherently neutral and objective, given that robots simply intake

0:54:41.719 --> 0:54:46.080
<v Speaker 1>data and systematically output results, But they are actually neither.

0:54:46.480 --> 0:54:49.400
<v Speaker 1>Robots are only as neutral as the data they're fed,

0:54:49.560 --> 0:54:52.640
<v Speaker 1>and only as objective as the design choices of those

0:54:52.680 --> 0:54:56.960
<v Speaker 1>who create them. When either bias or subjectivity infiltrates a

0:54:57.080 --> 0:55:01.200
<v Speaker 1>systems inputs or design choices, it is in inevitably reflected

0:55:01.239 --> 0:55:04.080
<v Speaker 1>in the system's outputs. This is your classic garbage in

0:55:04.160 --> 0:55:08.600
<v Speaker 1>garbage out, problem, right, They go on, Accordingly, those responsible

0:55:08.640 --> 0:55:12.840
<v Speaker 1>for overseeing the deployment of robots must anticipate the possibility

0:55:12.880 --> 0:55:17.280
<v Speaker 1>that algorithmically biased applications will cause harms of this systemic

0:55:17.360 --> 0:55:21.279
<v Speaker 1>nature to third parties. So uh, an example that's much

0:55:21.320 --> 0:55:24.839
<v Speaker 1>discussed in this would be an AI trained to make

0:55:24.880 --> 0:55:29.800
<v Speaker 1>decisions about granting loans by studying patterns of which loan

0:55:29.920 --> 0:55:33.680
<v Speaker 1>applicants got their loans granted in the past. And a

0:55:33.880 --> 0:55:36.440
<v Speaker 1>I like this could end up manifesting some type of

0:55:36.440 --> 0:55:39.439
<v Speaker 1>bias that hurts people, like a racial bias in its

0:55:39.440 --> 0:55:43.359
<v Speaker 1>loan assessments, because there was already a bias in the

0:55:43.440 --> 0:55:46.520
<v Speaker 1>real world data set that it was trained on. So,

0:55:46.560 --> 0:55:49.200
<v Speaker 1>in other words, AI that is trained on data from

0:55:49.239 --> 0:55:53.120
<v Speaker 1>the real world, unless it is it is explicitly told

0:55:53.160 --> 0:55:55.680
<v Speaker 1>not to do this, it will tend to reproduce and

0:55:55.760 --> 0:56:01.040
<v Speaker 1>perpetuate any injustices, any inequalities that already exist. And the

0:56:01.040 --> 0:56:05.720
<v Speaker 1>authors here give an example that is based on algorithmically

0:56:05.840 --> 0:56:09.040
<v Speaker 1>derived insurance premiums that I think they're talking about auto

0:56:09.080 --> 0:56:13.200
<v Speaker 1>insurance quote. A recent study by Consumer Reports found that

0:56:13.239 --> 0:56:18.120
<v Speaker 1>contemporary premiums depended less on driving habits and increasingly on

0:56:18.239 --> 0:56:24.280
<v Speaker 1>socioeconomic factors, including an individual's credit score. After analyzing two

0:56:24.360 --> 0:56:28.880
<v Speaker 1>billion car insurance price quotes across approximately seven hundred companies,

0:56:29.239 --> 0:56:32.960
<v Speaker 1>the study found that credit scores factored into insurance algorithms

0:56:33.000 --> 0:56:37.000
<v Speaker 1>so heavily that perfect drivers with low credit scores often

0:56:37.040 --> 0:56:41.680
<v Speaker 1>paid substantially more than terrible drivers with high scores. The

0:56:41.719 --> 0:56:45.560
<v Speaker 1>studies findings raised widespread concerns that AI systems used to

0:56:45.600 --> 0:56:49.520
<v Speaker 1>generate these quotes could create negative feedback loops that are

0:56:49.520 --> 0:56:53.799
<v Speaker 1>hard to break. According to one expert quote, higher insurance

0:56:53.840 --> 0:56:57.200
<v Speaker 1>prices for low income people can translate to higher debt

0:56:57.560 --> 0:57:01.200
<v Speaker 1>and plummeting credit scores, which can mean use job prospects,

0:57:01.239 --> 0:57:04.440
<v Speaker 1>which allows debt to pile up, credit scores to sink lower,

0:57:04.520 --> 0:57:08.799
<v Speaker 1>and insurance rates to increase in a vicious cycle. Uh so,

0:57:08.840 --> 0:57:11.000
<v Speaker 1>this is kind of a nightmare scenario, right, Like an

0:57:11.000 --> 0:57:15.600
<v Speaker 1>AI that is too powerful and not explicitly protected against

0:57:15.640 --> 0:57:18.960
<v Speaker 1>acquiring these types of biases could create these kind of

0:57:19.080 --> 0:57:23.960
<v Speaker 1>computer enforced prisons in reality, like a machine code for

0:57:24.120 --> 0:57:27.560
<v Speaker 1>perpetuating whatever state of the world, like whatever state the

0:57:27.600 --> 0:57:30.800
<v Speaker 1>world was in when the AI was first deployed and

0:57:30.840 --> 0:57:34.920
<v Speaker 1>then just entrenching it further and further. Yeah, And that

0:57:35.000 --> 0:57:37.360
<v Speaker 1>kind of thing is especially scary because like, if there's

0:57:37.360 --> 0:57:40.200
<v Speaker 1>a human making the decision, you can you can call

0:57:40.280 --> 0:57:42.880
<v Speaker 1>up the human to a witness stand or ask them like, hey,

0:57:42.880 --> 0:57:45.560
<v Speaker 1>why did you make the decision this way? But if

0:57:45.600 --> 0:57:48.000
<v Speaker 1>it's an AI doing it, you could say like, hey,

0:57:48.000 --> 0:57:50.400
<v Speaker 1>why why is it? Why are we getting this outcome

0:57:50.480 --> 0:57:53.320
<v Speaker 1>that's you know, creating a sort of like cyclical prison

0:57:53.400 --> 0:57:55.680
<v Speaker 1>out of reality, And they can just say, hey, you

0:57:55.680 --> 0:57:58.080
<v Speaker 1>know it's the machine with the machine. You know it

0:57:58.360 --> 0:58:00.959
<v Speaker 1>knows what it's doing. Yeah, yes, the machine it says

0:58:01.000 --> 0:58:04.560
<v Speaker 1>I learned it from watching you dad, and you have

0:58:04.600 --> 0:58:07.280
<v Speaker 1>that moment of shame. So I think these different categories

0:58:07.360 --> 0:58:09.440
<v Speaker 1>that they that they bring up are really important for

0:58:09.480 --> 0:58:13.200
<v Speaker 1>helping us kind of sort our ideas into into recognizable

0:58:13.280 --> 0:58:16.000
<v Speaker 1>types for for ways that AI and robots could go

0:58:16.040 --> 0:58:18.320
<v Speaker 1>wrong and could potentially cause harm that you would seek

0:58:18.400 --> 0:58:22.240
<v Speaker 1>legal remedy for. And also they help identify the spaces

0:58:22.360 --> 0:58:24.760
<v Speaker 1>that there's the most worry. I mean, for me, I

0:58:24.760 --> 0:58:27.920
<v Speaker 1>think that would be like those last two cases, right,

0:58:27.960 --> 0:58:31.439
<v Speaker 1>the unforeseen problems and the systemic problems are the ones

0:58:31.800 --> 0:58:34.920
<v Speaker 1>where there's the most real danger, I think, and the

0:58:34.920 --> 0:58:38.080
<v Speaker 1>most difficulty in trying to figure out how to solve it. Yeah,

0:58:38.120 --> 0:58:41.840
<v Speaker 1>because we we kind of you know, train ourselves for

0:58:42.240 --> 0:58:45.240
<v Speaker 1>to a certain extent and sort of culturally focus on

0:58:45.400 --> 0:58:51.000
<v Speaker 1>the sky net problems, right, the really obvious, um uh

0:58:51.280 --> 0:58:54.160
<v Speaker 1>situations where the robot car veers off the road in

0:58:54.160 --> 0:58:58.080
<v Speaker 1>a dangerous way. But the situations where it is just

0:58:58.520 --> 0:59:03.680
<v Speaker 1>perpetuating what we're already doing, where it's making choices in

0:59:03.800 --> 0:59:06.800
<v Speaker 1>getting from point A to point B that don't violate

0:59:06.840 --> 0:59:09.920
<v Speaker 1>anything we told it, but just are an uninventive and

0:59:09.960 --> 0:59:13.440
<v Speaker 1>even harmful way of doing it. Uh. Yeah, that's that's

0:59:13.480 --> 0:59:16.360
<v Speaker 1>that's harder to deal with. That's a type of misbehavior

0:59:16.720 --> 0:59:20.280
<v Speaker 1>that you can't solve by just having Dan o'harla hay

0:59:20.360 --> 0:59:27.160
<v Speaker 1>stand up and bellowing behave yourselves exactly. Um yeah, yeah, yeah,

0:59:27.200 --> 0:59:29.320
<v Speaker 1>I mean I can't remember if that even worked. I

0:59:29.360 --> 0:59:31.600
<v Speaker 1>just remember that was one of my favorite moments in

0:59:31.840 --> 0:59:34.040
<v Speaker 1>that was RoboCop two, right, was it? Yeah? Well, I

0:59:34.040 --> 0:59:36.520
<v Speaker 1>mean RoboCop one. I think we're already dealing with this

0:59:36.560 --> 0:59:40.360
<v Speaker 1>problem of like the sort of like weird dynamics of

0:59:40.400 --> 0:59:43.760
<v Speaker 1>machine culpability when ed two oh, nine like shoots that

0:59:43.800 --> 0:59:47.120
<v Speaker 1>guy five hundred times in the boardroom during the demonstration,

0:59:47.560 --> 0:59:50.160
<v Speaker 1>and then Dan O'Hurley. He's response to it is to

0:59:50.240 --> 0:59:53.080
<v Speaker 1>turn to Ronnie Cox and say, I'm very disappointed, Dick.

0:59:55.800 --> 0:59:58.360
<v Speaker 1>But anyway, well, I guess we're running running kind of long,

0:59:58.400 --> 1:00:00.560
<v Speaker 1>so maybe we should call part one there. But we

1:00:00.600 --> 1:00:05.160
<v Speaker 1>will resume this discussion about about robot justice and robot

1:00:05.240 --> 1:00:08.240
<v Speaker 1>punishment in the next episode. That's right, We'll be back

1:00:08.440 --> 1:00:11.640
<v Speaker 1>with more of this discussion in the meantime. If you

1:00:11.640 --> 1:00:13.720
<v Speaker 1>would like to check out past episodes of Stuff to

1:00:13.720 --> 1:00:16.800
<v Speaker 1>Blow Your Mind, uh, and the definitely worth checking out

1:00:16.840 --> 1:00:19.160
<v Speaker 1>because we have lots of past episodes that deal with

1:00:19.280 --> 1:00:22.080
<v Speaker 1>robots and AI. We have lots of episodes where we

1:00:22.120 --> 1:00:25.880
<v Speaker 1>make RoboCop references, so they're all they're all there, go

1:00:25.920 --> 1:00:27.640
<v Speaker 1>back and check them out. You can find our podcast

1:00:27.680 --> 1:00:29.880
<v Speaker 1>wherever you get your podcasts. Just look for the Stuff

1:00:29.920 --> 1:00:32.800
<v Speaker 1>to Blow your Mind podcast feed. UH. In that feed,

1:00:32.840 --> 1:00:35.640
<v Speaker 1>we put out core episodes of the show on Tuesdays

1:00:35.640 --> 1:00:39.240
<v Speaker 1>and Thursdays. Mondays, we have a little listener mail Wednesdays,

1:00:39.280 --> 1:00:42.840
<v Speaker 1>so that's when we do the artifact shorty uh usually,

1:00:43.160 --> 1:00:45.800
<v Speaker 1>and then on Friday's we do weird house cinema. That's

1:00:45.800 --> 1:00:48.160
<v Speaker 1>our chance to sort of set most of the science

1:00:48.200 --> 1:00:52.040
<v Speaker 1>aside and just focus on the films about rampaging robots.

1:00:52.800 --> 1:00:55.560
<v Speaker 1>You just thinks, as always to our excellent audio producer

1:00:55.640 --> 1:00:57.840
<v Speaker 1>Seth Nicholas Johnson. If you would like to get in

1:00:57.960 --> 1:01:00.480
<v Speaker 1>touch with us with feedback on this episode or any other,

1:01:00.720 --> 1:01:02.760
<v Speaker 1>to suggest a topic for the future, or just to

1:01:02.800 --> 1:01:05.480
<v Speaker 1>say hello, you can email us at contact at stuff

1:01:05.520 --> 1:01:15.200
<v Speaker 1>to Blow your Mind dot com. Stuff to Blow Your

1:01:15.200 --> 1:01:18.120
<v Speaker 1>Mind is production of I Heart Radio. For more podcasts

1:01:18.160 --> 1:01:20.240
<v Speaker 1>for my heart Radio, visit the i heart Radio app,

1:01:20.400 --> 1:01:32.280
<v Speaker 1>Apple Podcasts, or wherever you're listening to your favorite shows.