WEBVTT - From the Vault: Punish the Machine, Part 2 0:00:05.680 --> 0:00:07.960 Hey, welcome to Stuff to Blow Your Mind. My name 0:00:08.119 --> 0:00:11.240 is Robert Lamb and I'm Joe McCormick, and it's Saturday. 0:00:11.280 --> 0:00:13.880 Time for an episode from the Vault. This is the 0:00:14.000 --> 0:00:18.520 episode Punish the Machine, Part two. It's all about about 0:00:18.560 --> 0:00:22.639 machine moral agency, legal agency and culpability, or what we 0:00:22.720 --> 0:00:25.560 make of this emergent world when when machines are acting 0:00:25.560 --> 0:00:31.920 in an increasingly autonomous way. Let us delay no further. 0:00:32.640 --> 0:00:34.960 Welcome to Stuff to Blow Your Mind production of My 0:00:35.080 --> 0:00:44.240 Heart Radio. Hey, welcome to Stuff to Blow Your Mind. 0:00:44.280 --> 0:00:46.879 My name is Robert Lamb and I'm Joe McCormick, and 0:00:46.920 --> 0:00:49.480 we're back for part two of our talk about Punishing 0:00:49.479 --> 0:00:51.960 the Robot. We are we are back here to uh 0:00:52.080 --> 0:00:55.240 to tell the robot he's been very bad now. In 0:00:55.280 --> 0:00:58.520 the last episode, we talked about the idea of legal 0:00:58.560 --> 0:01:03.200 agency and culpability for robots and other intelligent machines, and 0:01:03.320 --> 0:01:05.559 for a quick refresher on on some of the stuff 0:01:05.600 --> 0:01:08.560 we went over. We talked about the idea that as 0:01:08.680 --> 0:01:12.200 robots and AI become more sophisticated and thus in some 0:01:12.280 --> 0:01:16.560 ways or in some cases more independent and unpredictable, and 0:01:16.640 --> 0:01:19.160 as they integrate more and more into the wild of 0:01:19.240 --> 0:01:23.080 human society, there are just inevitably going to be situations 0:01:23.160 --> 0:01:28.200 where AI and robots do wrong and cause harm to people. Now, 0:01:28.200 --> 0:01:30.880 of course, when a human does wrong and causes harm 0:01:30.920 --> 0:01:33.679 to another human, we have a legal system through which 0:01:33.680 --> 0:01:36.600 the victim can seek various kinds of remedies. And we 0:01:36.640 --> 0:01:38.960 talked in the last episode about the idea of remedies 0:01:39.000 --> 0:01:42.039 that the simple version of that is the remedy is 0:01:42.080 --> 0:01:44.520 what do I get when I win in court? So 0:01:44.640 --> 0:01:47.800 that can be things like monetary rewards. You know, I 0:01:47.920 --> 0:01:49.880 ran into your car with my car, so I pay 0:01:49.920 --> 0:01:53.000 you money, Or it can be punishment, or it can 0:01:53.040 --> 0:01:56.480 be court orders like commanding or restricting the behavior of 0:01:56.520 --> 0:01:59.680 the perpetrator. And so we discussed the idea that as 0:01:59.800 --> 0:02:04.520 roots become more unpredictable and more like human agents, more 0:02:04.560 --> 0:02:09.119 sort of independent, and more integrated into society, it might 0:02:09.240 --> 0:02:12.160 make sense to have some kind of system of legal 0:02:12.200 --> 0:02:17.360 remedies for when robots cause harm or commit crimes. But also, 0:02:17.440 --> 0:02:20.720 as we talked about last time, this is much easier 0:02:20.720 --> 0:02:23.000 said than done. It's going to present tons of new 0:02:23.040 --> 0:02:27.239 problems because our legal system is in many ways not 0:02:27.520 --> 0:02:32.799 equipped to deal with with defendants and situations of this kind, 0:02:33.280 --> 0:02:35.359 and this may cause us to ask questions about how 0:02:35.360 --> 0:02:39.720 we already think about culpability and blame and and punishment 0:02:39.800 --> 0:02:42.440 in the legal system. And so in the last episode 0:02:42.440 --> 0:02:44.920 we talked about one big legal paper that we're going 0:02:45.000 --> 0:02:47.480 to continue to explore in this one. It's by Mark A. 0:02:47.680 --> 0:02:51.160 Limley and Brian Casey in the University of Chicago Law 0:02:51.200 --> 0:02:55.040 Review from twenty nineteen, called Remedies for Robots. So I'll 0:02:55.080 --> 0:02:57.079 be referring back to that one a good bit throughout 0:02:57.080 --> 0:02:59.520 this episode two. Now, I think when we left off 0:02:59.600 --> 0:03:02.760 last time, we had mainly been talking about sort of 0:03:02.760 --> 0:03:05.760 trying to categorize the different sorts of harm that could 0:03:05.760 --> 0:03:09.600 be done by robots or AI intelligent machines. And so 0:03:09.639 --> 0:03:12.919 we talked about some things like unavoidable harms and deliberate 0:03:12.919 --> 0:03:15.440 at least cost harms. These are sort of going to 0:03:15.480 --> 0:03:19.480 be unavoidable parts of having something like autonomous vehicles, right, 0:03:19.520 --> 0:03:21.720 if you have cars driving around on the road, Like, 0:03:21.760 --> 0:03:25.080 even if they're really really good at minimizing harm, there's 0:03:25.080 --> 0:03:27.880 still going to be some cases where there's just no 0:03:27.960 --> 0:03:31.119 way harm could be avoided because their cars. Another would 0:03:31.120 --> 0:03:33.720 be defect driven harms. That's pretty straightforward. That's just where 0:03:33.840 --> 0:03:37.160 the machine malfunctions or breaks in some way. Another would 0:03:37.160 --> 0:03:41.200 be misuse harms. That's where the machine is used in 0:03:41.240 --> 0:03:43.840 a way that is harmful, and in those cases it 0:03:43.880 --> 0:03:46.640 can be usually pretty clear who's at fault. It's the 0:03:46.640 --> 0:03:49.120 person who misused the machine. But then there are a 0:03:49.120 --> 0:03:52.000 couple of other categories that where things get really tricky, 0:03:52.360 --> 0:03:56.520 which are unforeseen harms and systemic harms. And in the 0:03:56.560 --> 0:03:59.840 case of unforeseen harms, one example we talked about in 0:03:59.840 --> 0:04:04.120 the last episode was the drone that invented a wormhole. So, 0:04:04.440 --> 0:04:07.080 you know, people were trying to train a drone to 0:04:07.320 --> 0:04:11.360 move towards like an autonomous flying vehicle, to move towards 0:04:11.440 --> 0:04:14.760 the center of a circular area. But the drone started 0:04:14.800 --> 0:04:17.119 doing a thing where when it got sufficiently far away 0:04:17.120 --> 0:04:18.760 from the center of the circle, it would just fly 0:04:18.920 --> 0:04:21.680 out of the circle altogether. And so it seems kind 0:04:21.680 --> 0:04:24.240 of weird at first, like, Okay, why would it be 0:04:24.279 --> 0:04:27.479 doing that, But then what the researchers realized was that 0:04:27.520 --> 0:04:29.919 whenever it did that, they would turn it off, and 0:04:29.960 --> 0:04:32.240 then they would move it back into the circle to 0:04:32.279 --> 0:04:35.520 start it over again. So from the machine learning point 0:04:35.520 --> 0:04:38.280 of view of the drone itself. It had discovered like 0:04:38.360 --> 0:04:41.560 a like a time space warp that you know. So 0:04:41.560 --> 0:04:43.239 so it was doing this thing that made no sense 0:04:43.279 --> 0:04:45.800 from a human perspective, but actually was it was following 0:04:45.800 --> 0:04:50.600 its programming exactly. Now for an example, uh, sort of 0:04:50.640 --> 0:04:54.120 a thought experiment of how this could become lethal. There's 0:04:54.120 --> 0:04:56.920 an example that is stuck in my head. I can't 0:04:56.920 --> 0:04:59.680 recall where I heard this, who presented this idea? And 0:04:59.720 --> 0:05:01.479 I kind of had it in my head that it 0:05:01.480 --> 0:05:04.400 came from Max tag Mark. But I did some searching 0:05:04.440 --> 0:05:06.760 around in my notes and some searching around and uh, 0:05:07.279 --> 0:05:09.680 one of his books, and I couldn't find it. Perhaps 0:05:09.680 --> 0:05:12.680 you can help refresh me. You remember maybe you remember 0:05:12.680 --> 0:05:15.279 this joke, but the idea of the the AI that 0:05:15.400 --> 0:05:17.960 is running deciding how much oxygen needs to be in 0:05:18.000 --> 0:05:21.200 a train station, it didn't given time. Oh this sounds familiar. 0:05:21.320 --> 0:05:23.440 I don't know the answer, but a lot of these 0:05:23.440 --> 0:05:25.960 thought experiments tend to trace back to Nick Bostrom, so 0:05:26.000 --> 0:05:29.279 I wouldn't be surprised if in there. But but go ahead, 0:05:29.360 --> 0:05:31.960 right okay, as I remember it. The way it works 0:05:32.040 --> 0:05:34.440 is you have you have this AI that's in charge 0:05:34.480 --> 0:05:36.680 of of making sure there's enough oxygen in the train 0:05:36.720 --> 0:05:40.320 station for when humans are there, and it seems to 0:05:40.360 --> 0:05:42.360 have learned this fine, and when humans are there to 0:05:42.480 --> 0:05:46.359 get on the train, everything goes goes well, everybody's breathing fine. 0:05:46.839 --> 0:05:50.200 And then one in one day, Uh, the train arrives 0:05:50.240 --> 0:05:52.040 a little late or it leaves a little right late, 0:05:52.040 --> 0:05:54.839 if you get which whatever it is, and there's not 0:05:54.960 --> 0:05:58.360 enough oxygen and people die. And then it turns out 0:05:58.400 --> 0:06:01.400 that the train was not based saying its decision on 0:06:01.720 --> 0:06:03.920 when people were there, but it was basing it on 0:06:04.160 --> 0:06:08.880 a clock in the train station, like what it was. Um. 0:06:08.920 --> 0:06:11.200 And I may be mangling this horribly, but you know, 0:06:11.240 --> 0:06:13.800 another way of illustrating the point that machine learning could 0:06:13.920 --> 0:06:18.600 end up, you know, latching onto shortcuts or heuristic devices 0:06:19.440 --> 0:06:23.679 that would just seem completely insane to a quote unquote 0:06:23.720 --> 0:06:27.240 logical human mind, but might make sense within the framework 0:06:27.400 --> 0:06:30.479 of the AI. Right, they worked in the training cases, 0:06:30.520 --> 0:06:33.159 and it doesn't understand because it doesn't have common sense, 0:06:33.200 --> 0:06:36.080 it doesn't understand why they wouldn't work. In another case, 0:06:37.279 --> 0:06:39.720 there was actually a real world case that we talked 0:06:39.720 --> 0:06:43.440 about in Part one where there was an attempt to 0:06:43.440 --> 0:06:48.400 do some machine learning on what risk factors would would 0:06:48.440 --> 0:06:51.440 make a pneumonia case admitted to the hospital have a 0:06:51.520 --> 0:06:54.719 higher or lower chance of survival. And one thing that 0:06:54.839 --> 0:06:58.960 a machine learning algorithm determined was that asthma meant that 0:06:59.040 --> 0:07:01.960 you were you better off when you got pneumonia if 0:07:02.000 --> 0:07:05.039 you had asthma. But actually the reason for that that 0:07:05.040 --> 0:07:07.080 that isn't true. Actually the reason for that is that 0:07:07.120 --> 0:07:10.720 if you have asthma, you're a higher risk case for pneumonia, 0:07:11.000 --> 0:07:14.600 so you've got more intensive treatment in the hospital and 0:07:14.600 --> 0:07:17.040 thus had better outcomes on the data set that the 0:07:17.040 --> 0:07:19.760 algorithm was trained on. But the algorithm came up with 0:07:19.800 --> 0:07:23.720 this completely backwards uh failure to understand the difference between 0:07:23.760 --> 0:07:27.160 correlation and causation. There it made it look like asthma 0:07:27.200 --> 0:07:30.240 was a superpower. Now, of course, if you you take 0:07:30.320 --> 0:07:33.080 that kind of shortsighted algorithm and you make it god, 0:07:33.160 --> 0:07:34.720 then it will say, oh, I've just got to give 0:07:34.760 --> 0:07:37.200 everybody asthma, so so we will have a better chance 0:07:37.240 --> 0:07:40.200 of surviving. The point is it can be hard to 0:07:40.480 --> 0:07:43.880 imagine in advance all the cases like this that would 0:07:43.880 --> 0:07:48.120 arise when you've got a world full of robots and 0:07:48.120 --> 0:07:50.119 and AI is running around in it that are trained 0:07:50.120 --> 0:07:53.040 on machine learning. Basically, they're just a number of less 0:07:53.040 --> 0:07:56.360 sort of soft sky nets that you couldn't possibly predict, 0:07:56.920 --> 0:07:58.800 you know, like the sky net scenario being sort of 0:07:58.840 --> 0:08:01.240 like a robots decide that one to end all war 0:08:01.800 --> 0:08:05.880 and um, you know, humans causal war, therefore end all humans, 0:08:05.920 --> 0:08:09.080 that sort of thing. But there's so many different like 0:08:09.200 --> 0:08:12.280 lesser versions of it. It It could also be destructive or 0:08:12.360 --> 0:08:15.200 annoying or just get in the way of effectively using 0:08:15.240 --> 0:08:18.920 AI for whatever we turn to it for. Uh yeah, yeah. 0:08:18.960 --> 0:08:21.320 To come to an example that is definitely used by 0:08:21.400 --> 0:08:24.760 Nick Bostrom, the paper clip maximizer. You know, a robot 0:08:24.800 --> 0:08:27.280 that is designed to make as many paper clips as 0:08:27.320 --> 0:08:29.520 it can, and it just it looks at your body 0:08:29.520 --> 0:08:31.520 and says, hey, that's full of matter. Those could be 0:08:31.560 --> 0:08:37.000 paper clips. Yeah that yeah, yeah, that would be that 0:08:37.000 --> 0:08:39.480 would be quite an apocalypse. Now, before we get back 0:08:39.520 --> 0:08:41.880 into the main subject and talking about this limly in 0:08:41.960 --> 0:08:45.920 case paper with with robots as offenders, there was one 0:08:46.000 --> 0:08:48.040 thing that was interesting I came across. It was just 0:08:48.080 --> 0:08:51.520 a brief footnote in their paper, but about the question 0:08:51.840 --> 0:08:54.480 of what about if the robot is the plaintiff in 0:08:54.480 --> 0:08:57.880 a case? Uh, They said, it's it is possible to 0:08:57.960 --> 0:09:00.680 imagine a robot as a plaintiff in a court case, 0:09:01.240 --> 0:09:05.040 because of course robots, you know, can be injured by humans. 0:09:05.040 --> 0:09:07.760 And they cite a bunch of examples of news stories 0:09:07.800 --> 0:09:12.120 of humans just intentionally like torturing and being cruel two 0:09:12.200 --> 0:09:16.400 robots like that. They cite one news article from eighteen 0:09:16.480 --> 0:09:22.280 about people just aggressively kicking food delivery robots, and then 0:09:22.320 --> 0:09:24.520 they share another story. I actually remember this one from 0:09:24.520 --> 0:09:29.120 the news from about a Silicon Valley security robot that 0:09:29.240 --> 0:09:31.920 was just violently attacked by a drunk man in a 0:09:31.960 --> 0:09:34.520 parking garage. I don't remember this one, but I can 0:09:34.559 --> 0:09:38.480 imagine how it went down. Yeah, exactly. So they say 0:09:38.480 --> 0:09:41.119 that in a case like this, this is actually pretty 0:09:41.160 --> 0:09:44.079 straightforward as a property crime. I mean, unless we start 0:09:44.720 --> 0:09:48.200 getting into a scenario where we're really seeing robots as 0:09:48.280 --> 0:09:51.559 like human beings with their own like consciousness and interests 0:09:51.600 --> 0:09:55.880 and all that, the attacks against robots are really probably 0:09:55.960 --> 0:09:59.040 just property crimes against the owner of the robot. It's like, 0:09:59.080 --> 0:10:02.880 you know, attacking somebody computer or their car or something potentially. 0:10:02.920 --> 0:10:05.840 But we'll get into some stuff a little later that 0:10:06.000 --> 0:10:08.360 I think shows some other directions that could go in 0:10:08.400 --> 0:10:11.800 as well, you know, um, you know, especially consider the 0:10:11.840 --> 0:10:15.760 awesome possibility of robots owning themselves. Yeah, and and that's 0:10:15.760 --> 0:10:18.120 obviously a very different world. I mean, where you get 0:10:18.120 --> 0:10:21.520 into the idea like does a robot actually have rights? Um, 0:10:21.800 --> 0:10:24.640 which is not. That's sort of beyond the horizon of 0:10:24.679 --> 0:10:27.319 what's explored in in this paper itself. This paper is 0:10:27.360 --> 0:10:30.080 more focused on like the kinds of robots that you 0:10:30.120 --> 0:10:34.080 can practically imagine within the next few decades. And and 0:10:34.160 --> 0:10:36.400 in those cases, it seems like all of the really 0:10:36.440 --> 0:10:40.800 thorny stuff would probably be in robots as offenders rather 0:10:40.880 --> 0:10:44.760 than robots as victims of crimes. Right. But to your point, 0:10:44.840 --> 0:10:48.040 like the the initial crimes against robot that we can 0:10:48.280 --> 0:10:51.079 robots that we can imagine would be stuff like drunk 0:10:51.120 --> 0:10:53.880 people pushing them over things like that, Yeah, or just 0:10:53.960 --> 0:10:56.640 like a human driver and a human powered vehicle hitting 0:10:56.640 --> 0:10:59.880 an autonomous vehicle, you know, right. Aw As I mentioned 0:10:59.880 --> 0:11:02.000 in the last episode, this is a very big paper 0:11:02.040 --> 0:11:04.160 and we're not gonna have time to get into every 0:11:04.240 --> 0:11:06.360 avenue they go down in it. But I just wanted 0:11:06.400 --> 0:11:08.839 to go through, uh and and mention some ideas that 0:11:08.920 --> 0:11:12.120 stuck out to me is interesting that they discuss. And 0:11:12.160 --> 0:11:17.040 one thing that really fascinated me about this was that 0:11:17.240 --> 0:11:21.480 the idea of robots as possible agents in in a 0:11:21.559 --> 0:11:26.960 legal context uh brings to the for a philosophical argument 0:11:27.000 --> 0:11:30.160 that has existed in the realm of substantive law for 0:11:30.200 --> 0:11:32.280 a while. UH. And I'll try not to be too 0:11:32.320 --> 0:11:34.280 dry about this, but I think it actually does get 0:11:34.320 --> 0:11:37.520 to some really interesting philosophical territory. UH. And this is 0:11:37.559 --> 0:11:41.520 the distinction between what Limly and Casey call the normative 0:11:41.800 --> 0:11:47.240 versus economic interpretations of substantive law. Again complicated philosophical and 0:11:47.320 --> 0:11:49.520 legal distinction. I'll try to do my best to sum 0:11:49.520 --> 0:11:53.680 it up simply. So. The normative perspective on substantive law 0:11:54.200 --> 0:11:59.280 says that the law is a prohibition against doing something bad. 0:11:59.520 --> 0:12:02.120 So in so thing is against the law. That means 0:12:02.360 --> 0:12:06.120 you shouldn't do it, And we would stop the offender 0:12:06.280 --> 0:12:08.960 from doing the thing that's against the law if we could. 0:12:09.360 --> 0:12:11.720 But since we usually can't stop them from doing it, 0:12:11.760 --> 0:12:15.480 often because it already happened, the remedy that exists. You 0:12:15.480 --> 0:12:18.920 know that maybe paying damages to the victim or something 0:12:18.960 --> 0:12:22.280 like that is a is a an attempt to right 0:12:22.400 --> 0:12:25.520 the wrong in other words, to do the next best 0:12:25.640 --> 0:12:29.079 thing to undoing the harm in the first place. So 0:12:29.440 --> 0:12:32.880 basically getting into the idea of negative reinforcement. Somebody or 0:12:32.920 --> 0:12:35.800 something did something bad, we can't we couldn't stop them 0:12:35.800 --> 0:12:39.640 from doing something bad, but we can try and and 0:12:39.760 --> 0:12:43.400 give them stimulus that would make them not do it again. 0:12:43.679 --> 0:12:46.959 Be that economic or otherwise. Well, yes, but I think 0:12:46.960 --> 0:12:49.520 what you're saying, uh, could actually apply to both of 0:12:49.559 --> 0:12:51.959 these conditions I'm going to talk about. So I think 0:12:52.000 --> 0:12:56.079 maybe the distinction comes in about whether that whether there 0:12:56.200 --> 0:12:59.000 is such a thing as an inherent prohibition. So the 0:12:59.000 --> 0:13:03.199 thing that's operative in the normative view is that the 0:13:03.200 --> 0:13:05.520 thing that's against the law is a thing that should 0:13:05.559 --> 0:13:10.000 not be done, and thus the remedy is an attempt 0:13:10.080 --> 0:13:12.200 to try to fix the fact that it was done 0:13:12.200 --> 0:13:15.680 in the first place. The the economic view is the 0:13:15.760 --> 0:13:18.160 alternative here, and the way they sum that up is 0:13:18.800 --> 0:13:22.800 there is no such thing as forbidden conduct. Rather, a 0:13:22.880 --> 0:13:27.679 substantive law tells you what the cost of the conduct is. 0:13:28.000 --> 0:13:31.080 Does that distinction make any more sense? Yes? Yes, So 0:13:31.120 --> 0:13:36.040 it's basically the first version is doing crimes is bad. Um, 0:13:36.200 --> 0:13:39.199 the second one is doing crimes is expensive. So it's 0:13:39.320 --> 0:13:42.360 the first is crime should not be done, and the 0:13:42.400 --> 0:13:44.840 second one is crimes can be done if you can 0:13:44.880 --> 0:13:48.760 afford it. Yes, exactly so. In Limley and Casey's words, quote, 0:13:49.000 --> 0:13:52.240 damages on this view, the economic view are simply a 0:13:52.320 --> 0:13:56.400 cost of doing business. One we want defendants to internalize, 0:13:56.640 --> 0:14:01.040 but not necessarily to avoid the conduct altogether. And now 0:14:01.120 --> 0:14:03.400 you might look at this and think, oh, okay, well, 0:14:03.440 --> 0:14:06.600 so the economic view is just like a psychopathic way 0:14:06.640 --> 0:14:09.199 of looking at things. And in a certain sense you 0:14:09.200 --> 0:14:11.320 could look at that as like, if you're calculating what's 0:14:11.320 --> 0:14:15.320 the economic cost of murder, then yeah, okay, that does 0:14:15.440 --> 0:14:18.960 just like that's evil, that's like psychopathic. But they're actually 0:14:18.960 --> 0:14:21.840 all kinds of cases we're thinking about. The economic view 0:14:21.920 --> 0:14:25.320 makes more sense of the way we actually behave And 0:14:25.360 --> 0:14:28.000 they use the example of stopping at a traffic light. 0:14:29.240 --> 0:14:31.920 So to read from Limely and Casey here quote Under 0:14:31.960 --> 0:14:35.440 the normative view, a red light stands as a prohibition 0:14:35.560 --> 0:14:38.680 against traveling through an intersection, with the remedy being a 0:14:38.680 --> 0:14:41.360 ticket or a fine against those who are caught breaking 0:14:41.400 --> 0:14:44.960 the prohibition. We would stop you from running the red 0:14:45.040 --> 0:14:48.320 light if we could, But because policing every intersection in 0:14:48.360 --> 0:14:51.680 the country would be impossible. We instead punish those we 0:14:51.760 --> 0:14:55.120 do catch in hopes of deterring others. So in this 0:14:55.200 --> 0:14:58.160 first case, you running a red light is bad, you 0:14:58.200 --> 0:15:00.880 should not do it, and the cost of doing it, 0:15:00.880 --> 0:15:03.080 you know, the punishment you face for doing it is 0:15:03.120 --> 0:15:06.360 an attempt to right that wrong. But then they say, 0:15:06.480 --> 0:15:10.120 under the economic view, however, and absolute prohibition against running 0:15:10.120 --> 0:15:13.000 red lights was never the intention. Rather, the red light 0:15:13.120 --> 0:15:16.600 merely signals a consequence for those who do, in fact 0:15:16.960 --> 0:15:20.560 choose to travel through the intersection. As in the first instance, 0:15:20.600 --> 0:15:23.280 the remedy available is a fine or a ticket. But 0:15:23.400 --> 0:15:25.400 under this view, the choice of whether or not to 0:15:25.480 --> 0:15:28.320 violate the law depends on the willingness of the lawbreaker 0:15:28.600 --> 0:15:32.120 to accept the penalty. So in the case of a 0:15:32.160 --> 0:15:34.880 red light, well that that might make more sense if 0:15:34.920 --> 0:15:37.280 you're like sitting at a red light and you look 0:15:37.320 --> 0:15:40.160 around and there are no other cars anywhere near you, 0:15:40.480 --> 0:15:42.800 and you've you've got a clear view of the entire 0:15:42.840 --> 0:15:45.840 intersection and the red lights not changing, and you think 0:15:45.880 --> 0:15:48.680 maybe it's broken, and you're just like, okay, I'm I'm 0:15:48.680 --> 0:15:50.840 just going to drive through. Well, if if you reach 0:15:50.920 --> 0:15:53.240 that point where you're like, I think it's broken. That 0:15:53.320 --> 0:15:55.560 I feel like that's a slightly different case. But if 0:15:55.560 --> 0:15:59.479 you're just like nobody's watching, I'm gonna do it, um 0:15:59.560 --> 0:16:02.280 and and and the light isn't taking an absurd amount 0:16:02.320 --> 0:16:05.360 of time or longer than you're you're accustomed to. Yeah, 0:16:05.400 --> 0:16:08.240 I don't know how the the belief that the light 0:16:08.360 --> 0:16:10.720 is broken would factor into that, but yeah, it is. 0:16:11.000 --> 0:16:13.360 I Mean, one thing that I think is clear that 0:16:13.400 --> 0:16:17.640 in it's that in many cases there are people, especially 0:16:17.680 --> 0:16:22.360 I think companies and corporations that operate on the economic view, 0:16:22.400 --> 0:16:25.480 and it is something that I think people generally look 0:16:25.480 --> 0:16:27.760 at and say, okay, that that's kind of grimy. Like 0:16:27.800 --> 0:16:30.080 it like a company that says, okay, there is a 0:16:30.160 --> 0:16:34.920 fine for not obeying this environmental regulation, and we're going 0:16:34.960 --> 0:16:37.880 to make more money by violating the regulation than we 0:16:37.920 --> 0:16:39.920 would pay in the fine anybody, So we're just gonna 0:16:39.960 --> 0:16:42.680 pay it. Yeah, you hear about that. With factories, for instance, 0:16:42.680 --> 0:16:45.320 where where there'll be there'll be some situation where that 0:16:45.400 --> 0:16:48.480 the fine is not significant enough to really be at 0:16:48.520 --> 0:16:51.840 a turrent, it's just a for them breaking that that 0:16:51.880 --> 0:16:55.160 mandate being called on it. Occasionally, it's just the cost 0:16:55.200 --> 0:16:58.520 of doing business. Right. Uh. So, there's a funny way 0:16:58.560 --> 0:17:00.680 to describe this point of view. The the authors bring 0:17:00.760 --> 0:17:04.399 up here that they called the bad man theory. And 0:17:04.480 --> 0:17:07.520 this comes from Justice Oliver Wendell Holmes, who is a 0:17:07.880 --> 0:17:10.720 US Supreme Court justice. Uh And he's talking about the 0:17:10.760 --> 0:17:14.560 economic view of substantive law. Uh And. Holmes wrote, quote, 0:17:14.760 --> 0:17:16.840 if you want to know the law and nothing else, 0:17:16.880 --> 0:17:19.640 you must look at it as a bad man who 0:17:19.680 --> 0:17:23.359 cares only for the material consequences which such knowledge enables 0:17:23.440 --> 0:17:26.399 him to predict, not as a good one who finds 0:17:26.480 --> 0:17:29.920 his reasons for conduct, whether inside the law or outside 0:17:29.920 --> 0:17:33.600 of it, in the vaguer sanctions of conscience. Uh. And, 0:17:33.640 --> 0:17:35.919 so they write, the measure of the substantive law, in 0:17:35.960 --> 0:17:38.840 other words, is not to be mixed up with moral qualms, 0:17:39.080 --> 0:17:42.359 but is simply coextensive with its remedy. No more and 0:17:42.440 --> 0:17:45.199 no less. It just is what the remedy is. It's 0:17:45.240 --> 0:17:47.399 the cost of doing business. Now, of course, there are 0:17:47.400 --> 0:17:50.480 plenty of legal scholars and philosophers who would dispute how 0:17:50.480 --> 0:17:53.600 Holmes thinks of this. But the interesting question is how 0:17:53.640 --> 0:17:57.320 does this apply to robots. If you're programming a robot 0:17:57.359 --> 0:18:01.600 to behave well, you actually don't get to just sort 0:18:01.600 --> 0:18:05.000 of like jump over this distinction the way humans do 0:18:05.160 --> 0:18:07.600 when they think about their own moral conduct. Right, Like, 0:18:07.640 --> 0:18:10.199 you're not sitting when you're trying to think what's a 0:18:10.200 --> 0:18:12.680 good way to be a good person. You're not sitting 0:18:12.680 --> 0:18:15.320 around thinking about well, am I going by the normative 0:18:15.400 --> 0:18:19.480 view of morality or the economic view of morality? You know, Um, 0:18:19.640 --> 0:18:21.720 you just sort of act a certain way whatever it 0:18:21.800 --> 0:18:23.760 seems to you the right way to do. But if 0:18:23.800 --> 0:18:26.119 you're trying to program a robot to behave well, you 0:18:26.160 --> 0:18:28.840 have to make a choice whether to embrace the normative 0:18:28.880 --> 0:18:32.000 view or the economic view. Does a robot view a 0:18:32.200 --> 0:18:36.639 red light, say, as a firm prohibition against forward movement, 0:18:36.680 --> 0:18:38.800 it's just a bad thing and you shouldn't do it 0:18:38.880 --> 0:18:41.199 to drive through a red light? Or does it just 0:18:41.359 --> 0:18:45.919 view it as a substantial discouragement against forward motion that 0:18:46.000 --> 0:18:49.119 has a certain cost, and if you were to overcome 0:18:49.160 --> 0:18:52.000 that cost, then you drive on through. Yeah, this is 0:18:52.000 --> 0:18:54.440 a great, great question because I feel like with humans 0:18:54.560 --> 0:18:56.880 we're probably mixing a match and all the time, yes, 0:18:57.240 --> 0:19:00.760 you know, perhaps the same law breaking behavior. You know, 0:19:00.840 --> 0:19:03.000 we may do both on on one thing, and we 0:19:03.040 --> 0:19:04.840 do one on another thing, and then the other one 0:19:04.880 --> 0:19:07.240 on its still a third thing, but with the robot 0:19:07.280 --> 0:19:09.679 it seems like you're gonna deal more or less with 0:19:09.800 --> 0:19:14.280 kind of an absolute direction. Either they're going to be 0:19:14.560 --> 0:19:18.000 um either that the law is is to be obeyed 0:19:18.080 --> 0:19:21.760 or the law is to be taken into your cost analysis. Well, yeah, 0:19:21.760 --> 0:19:23.880 so they talked about how the normative you is actually 0:19:23.960 --> 0:19:29.400 very much like uh Isaac Asimov's Laws of Robotics inviolable rules, 0:19:29.520 --> 0:19:33.000 and the Asimov story is doing a very good job 0:19:33.400 --> 0:19:38.960 of demonstrating why inviolable rules are really difficult to implement 0:19:39.000 --> 0:19:42.280 in the real world. Like you know that they Asimov 0:19:42.320 --> 0:19:45.359 explored this brilliantly. And along these lines, the authors here 0:19:45.440 --> 0:19:47.560 argue that there there are major reasons to think it 0:19:47.600 --> 0:19:51.240 will just not make any practical sense to program robots 0:19:51.440 --> 0:19:55.080 with a normative view of legal remedies. That probably when 0:19:55.080 --> 0:19:57.800 people make ais and robots that that have to take 0:19:57.800 --> 0:20:00.760 these kinds of things into account, they're a most definitely 0:20:00.760 --> 0:20:05.080 going to program them according to the economic view. Right. Uh. 0:20:05.119 --> 0:20:07.600 They say that quote encoding the rule don't run a 0:20:07.640 --> 0:20:11.440 red light as an absolute prohibition, for example, might sometimes 0:20:11.440 --> 0:20:14.399 conflict with the more compelling goal of not letting your 0:20:14.480 --> 0:20:18.679 driver die by being hit by an oncoming truck. So 0:20:18.960 --> 0:20:22.000 the robots are probably going to have to be economically 0:20:22.760 --> 0:20:26.840 economically motivated to an extent like this, um, But then 0:20:26.840 --> 0:20:29.119 they talk about how you know, this gets very complicated 0:20:29.160 --> 0:20:33.959 because robots will calculate the risks of reward and punishment 0:20:34.400 --> 0:20:37.600 with different biases than humans, or maybe even without the 0:20:37.640 --> 0:20:40.800 biases that humans have that the legal system relies on 0:20:41.000 --> 0:20:44.840 in order to keep us obedient. Humans are highly motivated, 0:20:45.000 --> 0:20:48.040 usually by certain types of punishments that like, you know, 0:20:48.440 --> 0:20:51.639 humans like really don't want to spend a month in jail, 0:20:52.600 --> 0:20:54.479 you know, most of the time. And you can't just 0:20:54.640 --> 0:20:58.840 rely on a robot to be incredibly motivated by something 0:20:58.880 --> 0:21:00.760 like this, first of all, because like it wouldn't even 0:21:00.760 --> 0:21:03.439 make sense to send the robot itself to jail. So 0:21:03.520 --> 0:21:07.600 you need some kind of organized system for making a 0:21:07.680 --> 0:21:11.879 robot understand the cost of bad behavior in a systematized 0:21:11.880 --> 0:21:14.240 way that made sense to the robot as a as 0:21:14.240 --> 0:21:17.959 a demotivating incentive. Yeah, Like shame comes to mind as 0:21:17.960 --> 0:21:20.600 another aspect of of all this, Like how do you 0:21:20.640 --> 0:21:23.879 shame a robot? You have to program a robot to 0:21:24.000 --> 0:21:26.560 feel Shane and being uh you know, made to give 0:21:26.600 --> 0:21:29.360 a public apology or something. Yeah. Uh so, so they 0:21:29.440 --> 0:21:31.760 argue that that it really only makes sense for robots 0:21:31.800 --> 0:21:34.840 to look at legal remedies in an economic way, and 0:21:34.840 --> 0:21:37.439 then they write quote, it thus appears that Justice Holmes 0:21:37.520 --> 0:21:41.399 archetypical bad man, will finally be brought to corporeal form, 0:21:41.640 --> 0:21:44.359