WEBVTT - Punish the Machine, Part 2 0:00:03.040 --> 0:00:05.280 Welcome to Stuff to Blow Your Mind, the production of 0:00:05.360 --> 0:00:14.640 My Heart Radio. Hey, welcome to Stuff to Blow Your Mind. 0:00:14.720 --> 0:00:17.319 My name is Robert Lamb and I'm Joe McCormick, and 0:00:17.360 --> 0:00:19.880 we're back for part two of our talk about punishing 0:00:19.920 --> 0:00:23.279 the robot. We're we're back here to uh to tell 0:00:23.320 --> 0:00:26.640 the robot he's been very bad. Now. In the last episode, 0:00:26.680 --> 0:00:30.640 we talked about the idea of legal agency and culpability 0:00:30.720 --> 0:00:34.240 for robots and other intelligent machines, and for a quick 0:00:34.280 --> 0:00:36.600 refresher on on some of the stuff we went over, 0:00:37.479 --> 0:00:40.120 we talked about the idea that as robots and AI 0:00:40.320 --> 0:00:43.320 become more sophisticated and thus in some ways or in 0:00:43.360 --> 0:00:47.960 some cases more independent and unpredictable, and as they integrate 0:00:48.040 --> 0:00:51.000 more and more into the wild of human society, they 0:00:51.040 --> 0:00:54.760 are just inevitably going to be situations where AI and 0:00:55.000 --> 0:00:58.960 robots do wrong and cause harm to people. Now, of course, 0:00:59.000 --> 0:01:02.200 when a human is wrong and causes harm to another human, 0:01:02.240 --> 0:01:04.920 we have a legal system through which the victim can 0:01:05.000 --> 0:01:07.440 seek various kinds of remedies, And we talked in the 0:01:07.520 --> 0:01:10.800 last episode about the idea of remedies that the simple 0:01:10.880 --> 0:01:12.959 version of that is the remedy is what do I 0:01:13.000 --> 0:01:15.480 get when I win in court? So that can be 0:01:15.560 --> 0:01:18.960 things like monetary rewards. You know, I ran into your 0:01:18.959 --> 0:01:21.560 car with my car, so I pay you money, or 0:01:21.600 --> 0:01:24.520 it can be punishment, or it can be court orders 0:01:24.560 --> 0:01:28.560 like commanding or restricting the behavior of the perpetrator. And 0:01:28.600 --> 0:01:31.760 so we discussed the idea that as robots become more 0:01:31.840 --> 0:01:36.000 unpredictable and more like human agents, more sort of independent, 0:01:36.400 --> 0:01:40.399 and more integrated into society, it might make sense to 0:01:40.440 --> 0:01:43.960 have some kind of system of legal remedies for when 0:01:44.080 --> 0:01:48.200 robots cause harm or commit crimes. But also, as we 0:01:48.240 --> 0:01:51.800 talked about last time, this is much easier said than done. 0:01:51.840 --> 0:01:54.640 It's going to present tons of new problems because our 0:01:54.720 --> 0:01:59.400 legal system is in many ways not equipped to deal 0:01:59.600 --> 0:02:04.000 with with defendants and situations of this kind. And this 0:02:04.040 --> 0:02:06.400 may cause us to ask questions about how we already 0:02:06.440 --> 0:02:10.359 think about culpability and blame and and punishment in the 0:02:10.440 --> 0:02:13.240 legal system. And so in the last episode we talked 0:02:13.240 --> 0:02:16.000 about one big legal paper that we're going to continue 0:02:16.000 --> 0:02:18.480 to explore in this one. It's by Mark A. Limley 0:02:18.560 --> 0:02:22.000 and Brian Casey in the University of Chicago Law Review 0:02:22.040 --> 0:02:25.600 from twenty nineteen called remedies for robots, So I'll be 0:02:25.840 --> 0:02:27.560 referring back to that one a good bit throughout this 0:02:27.600 --> 0:02:30.560 episode too. Now. I think when we left off last time, 0:02:30.639 --> 0:02:33.560 we had mainly been talking about sort of trying to 0:02:33.600 --> 0:02:36.600 categorize the different sorts of harm that could be done 0:02:36.680 --> 0:02:40.399 by robots or AI intelligent machines, and so we talked 0:02:40.400 --> 0:02:44.680 about some things like unavoidable harms and deliberate least cost harms. 0:02:44.760 --> 0:02:47.200 These are sort of going to be unavoidable parts of 0:02:47.240 --> 0:02:50.560 having something like autonomous vehicles, right if you have cars 0:02:50.639 --> 0:02:53.280 driving around on the road, Like, even if they're really 0:02:53.360 --> 0:02:56.280 really good at minimizing harm, there's still going to be 0:02:56.360 --> 0:02:59.079 some cases where there's just no way harm could be 0:02:59.120 --> 0:03:02.760 avoided because their cars. Another would be defect driven harms. 0:03:02.760 --> 0:03:05.680 That's pretty straightforward, that's just where the machine malfunctions or 0:03:05.720 --> 0:03:09.160 breaks in some way. Another would be misuse harms. That's 0:03:09.160 --> 0:03:12.840 where the machine is used in a way that is harmful. 0:03:13.240 --> 0:03:16.000 And in those cases it can be usually pretty clear 0:03:16.040 --> 0:03:18.680 who's at fault. It's the person who misused the machine. 0:03:19.040 --> 0:03:20.920 But then there are a couple of other categories that 0:03:20.960 --> 0:03:24.800 where things get really tricky, which are unforeseen harms and 0:03:24.919 --> 0:03:29.200 systemic harms and in the case of unforeseen harms. One 0:03:29.240 --> 0:03:31.720 example we talked about in the last episode was the 0:03:31.840 --> 0:03:35.560 drone that invented a wormhole. So, you know, people were 0:03:35.600 --> 0:03:39.080 trying to train a drone to move towards like an 0:03:39.080 --> 0:03:42.760 autonomous flying vehicle, to move towards the center of a 0:03:42.800 --> 0:03:45.920 circular area. But the drone started doing a thing where 0:03:45.920 --> 0:03:48.160 when it got sufficiently far away from the center of 0:03:48.160 --> 0:03:50.800 the circle, it would just fly out of the circle altogether. 0:03:51.240 --> 0:03:54.000 And so it seems kind of weird at first, like, Okay, 0:03:54.040 --> 0:03:56.440 why would it be doing that? But then what the 0:03:56.480 --> 0:03:59.360 researchers realized was that whenever it did that, they would 0:03:59.400 --> 0:04:01.280 turn it off off and then they would move it 0:04:01.320 --> 0:04:04.200 back into the circle to start it over again. So, 0:04:04.360 --> 0:04:07.320 from the machine learning point of view of the drone itself, 0:04:07.560 --> 0:04:10.360 it had discovered like a like a time space warp 0:04:10.680 --> 0:04:12.840 that you know. So so it was doing this thing 0:04:12.880 --> 0:04:15.240 that made no sense from a human perspective, but actually 0:04:15.360 --> 0:04:20.320 was it was following its programming exactly. Now for an example, uh, 0:04:20.800 --> 0:04:23.880 sort of a thought experiment of how this could become lethal. 0:04:24.160 --> 0:04:27.080 There's an example that is stuck in my head. I 0:04:27.120 --> 0:04:29.960 can't recall where I heard this who presented this idea? 0:04:30.080 --> 0:04:31.760 And I kind of had it in my head that 0:04:31.800 --> 0:04:34.479 it came from Max tag Mark. But I did some 0:04:34.520 --> 0:04:36.960 searching around in my notes and some searching around and 0:04:37.720 --> 0:04:40.039 one of his books, and I couldn't find it. Perhaps 0:04:40.080 --> 0:04:43.440 you can help refresh me. You remember maybe you remember this, Joe, 0:04:43.440 --> 0:04:46.280 But the idea of the the AI that is running 0:04:46.320 --> 0:04:49.200 deciding how much oxygen needs to be in a train station, 0:04:49.240 --> 0:04:52.799 it didn't given time. Oh this sounds familiar. I don't 0:04:52.800 --> 0:04:54.640 know the answer, but a lot of these thought experiments 0:04:54.640 --> 0:04:56.839 tend to trace back to Nick Bostrom, so I wouldn't 0:04:56.839 --> 0:05:00.719 be surprised if in there. But but go ahead, right, okay, 0:05:00.880 --> 0:05:03.320 as I remember it. The way it works is you have, um, 0:05:03.400 --> 0:05:05.719 you have this AI that's in charge of of making 0:05:05.720 --> 0:05:07.920 sure there's enough oxygen in the train station for when 0:05:08.080 --> 0:05:11.640 humans are there, and it seems to have learned this fine. 0:05:11.680 --> 0:05:13.520 And when humans are there to get on the train, 0:05:13.680 --> 0:05:19.760 everything goes goes well, everybody's breathing fine. And then one day, uh, 0:05:19.800 --> 0:05:21.800 the train arrives a little late or it leaves a 0:05:21.800 --> 0:05:24.440 little right late. You get which whatever it is, and 0:05:24.920 --> 0:05:28.160 there's not enough oxygen and people die and then it 0:05:28.200 --> 0:05:31.240 turns out that the train was not basing its decision 0:05:31.760 --> 0:05:34.080 on when people were there, but it was basing it 0:05:34.240 --> 0:05:37.679 on a clock in the train station, like what train 0:05:37.760 --> 0:05:40.960 it was? Um, And I may be mangling this horribly, 0:05:41.040 --> 0:05:43.200 but you know another way of illustrating the point that 0:05:43.240 --> 0:05:47.400 machine learning could end up, you know, latching onto shortcuts 0:05:47.480 --> 0:05:52.279 or heuristic devices. Uh. That would just seem completely insane 0:05:52.760 --> 0:05:56.400 to a quote unquote logical human mind, but might make 0:05:56.440 --> 0:05:59.760 sense within the framework of the AI. Right. They worked 0:05:59.760 --> 0:06:02.680 in training cases, and it doesn't understand because it doesn't 0:06:02.680 --> 0:06:05.719 have common sense, it doesn't understand why they wouldn't work 0:06:05.720 --> 0:06:09.640 in another case. There was actually a real world case 0:06:09.720 --> 0:06:12.760 that we talked about in Part one where there was 0:06:13.120 --> 0:06:17.080 an attempt to do some machine learning on what risk 0:06:17.160 --> 0:06:20.960 factors would would make a pneumonia case admitted to the 0:06:21.000 --> 0:06:24.440 hospital have a higher or lower chance of survival. And 0:06:24.560 --> 0:06:27.640 one thing that a machine learning algorithm determined was that 0:06:27.760 --> 0:06:31.360 asthma meant that you were you were better off when 0:06:31.400 --> 0:06:34.520 you got pneumonia if you had asthma. But actually the 0:06:34.560 --> 0:06:37.000 reason for that that that isn't true. Actually, the reason 0:06:37.040 --> 0:06:38.680 for that is that if you have asthma, you're a 0:06:38.800 --> 0:06:43.000 higher risk case for pneumonia, so you've got more intensive 0:06:43.080 --> 0:06:46.440 treatment in the hospital and thus had better outcomes on 0:06:46.520 --> 0:06:48.760 the data set that the algorithm was trained on. But 0:06:48.760 --> 0:06:52.880 the algorithm came up with this completely backwards uh failure 0:06:52.920 --> 0:06:56.400 to understand the difference between correlation and causation. There it 0:06:56.480 --> 0:07:00.000 made it look like asthma was a superpower. Now, of course, 0:07:00.040 --> 0:07:02.800 if you take that kind of shortsighted algorithm and you 0:07:02.880 --> 0:07:04.680 make it god, then it will say, oh, I've just 0:07:04.720 --> 0:07:07.159 got to give everybody asthma, so so we'll have a 0:07:07.160 --> 0:07:09.680 better chance of surviving. The point is it can be 0:07:10.120 --> 0:07:13.880 hard to imagine in advance all the cases like this 0:07:13.960 --> 0:07:16.880 that would arise when you've got a world full of 0:07:17.840 --> 0:07:20.080 robots and and AI is running around in it that 0:07:20.120 --> 0:07:23.080 are trained on machine learning. Basically, they're just a number 0:07:23.120 --> 0:07:25.840 of less sort of soft sky nets that you couldn't 0:07:25.880 --> 0:07:29.120 possibly predict, you know, like the skynet scenario being sort 0:07:29.160 --> 0:07:31.680 of like a robots decide that one to end all war, 0:07:32.240 --> 0:07:36.000 and um, you know, humans causal war, therefore end all humans, 0:07:36.320 --> 0:07:39.520 that sort of thing. But there's so many different like 0:07:39.640 --> 0:07:42.720 lesser versions of it. It It could also be destructive or 0:07:42.800 --> 0:07:45.640 annoying or just get in the way of effectively using 0:07:45.680 --> 0:07:49.360 AI for whatever we turn to it for. Uh yeah, yeah. 0:07:49.400 --> 0:07:51.720 To come to an example that is definitely used by 0:07:51.840 --> 0:07:55.200 Nick Bostrom, the paper clip maximizer. You know, a robot 0:07:55.240 --> 0:07:57.720 that is designed to make as many paper clips as 0:07:57.760 --> 0:07:59.920 it can, and it just it looks at your body 0:08:00.000 --> 0:08:01.960 and says, hey, that's full of matter. Those could be 0:08:02.000 --> 0:08:07.640 paper clips. Yeah, yeah, yeah, that would be that would 0:08:07.680 --> 0:08:10.120 be quite an apocalypse. Now, before we get back into 0:08:10.200 --> 0:08:12.720 the main subject and talking about this limly in Casey 0:08:12.760 --> 0:08:16.600 paper with with robots as offenders, there was one thing 0:08:16.800 --> 0:08:18.600 that was interesting I came across. It was just a 0:08:18.600 --> 0:08:22.560 brief footnote in their paper, but about the question of 0:08:22.600 --> 0:08:26.480 what about if the robot is the plaintiff in a case? Uh. 0:08:26.760 --> 0:08:29.320 They said, it's it is possible to imagine a robot 0:08:29.360 --> 0:08:32.880 as a plaintiff in a court case because of course robots, 0:08:32.920 --> 0:08:36.160 you know, can be injured by humans. And they cited 0:08:36.200 --> 0:08:38.960 a bunch of examples of news stories of humans just 0:08:39.080 --> 0:08:43.800 intentionally like torturing and being cruel two robots like that. 0:08:43.920 --> 0:08:48.760 They cited one news article fromen about people just aggressively 0:08:48.960 --> 0:08:54.040 kicking food delivery robots, and then they share another story actually, 0:08:54.080 --> 0:08:57.880 remember this one from the news from about a Silicon 0:08:58.040 --> 0:09:01.480 Valley security robot that was just violently attacked by a 0:09:01.559 --> 0:09:04.600 drunk man in a parking garage. I don't remember this one, 0:09:04.640 --> 0:09:07.600 but I can imagine how it went down. Yeah, exactly. 0:09:07.640 --> 0:09:10.480 So they say that in a case like this, this 0:09:10.559 --> 0:09:13.640 is actually pretty straightforward as a property crime. I mean, 0:09:13.760 --> 0:09:17.400 unless we start getting into a scenario where we're really 0:09:17.520 --> 0:09:20.440 seeing robots as like human beings with their own like 0:09:20.520 --> 0:09:25.520 consciousness and interests and all that, the attacks against robots 0:09:25.520 --> 0:09:28.720 are really probably just property crimes against the owner of 0:09:28.720 --> 0:09:31.280 the robot. It's like, you know, attacking somebody's computer or 0:09:31.320 --> 0:09:35.120 their car or something potentially. But that we'll get into 0:09:35.120 --> 0:09:37.560 some stuff a little later that I think shows some 0:09:37.640 --> 0:09:40.640 other directions that could go in as well, you know, um, 0:09:40.679 --> 0:09:45.520 you know, especially considered the awesome possibility of robots owning themselves. Yeah, 0:09:45.559 --> 0:09:48.040 and and that's obviously a very different world, I mean, 0:09:48.080 --> 0:09:50.000 where you get into the idea like does a robot 0:09:50.000 --> 0:09:53.680 actually have rights? Um, which is not. That's sort of 0:09:53.720 --> 0:09:57.200 beyond the horizon of what's explored in in this paper itself. 0:09:57.200 --> 0:09:59.719 This paper is more focused on like the kinds of 0:09:59.800 --> 0:10:02.640 ro bots that you can practically imagine within the next 0:10:02.640 --> 0:10:06.160 few decades. And and in those cases, it seems like 0:10:06.200 --> 0:10:09.120 all of the really thorny stuff would probably be in 0:10:09.600 --> 0:10:14.439 robots as offenders rather than robots as victims of crimes. Right. 0:10:14.480 --> 0:10:17.600 But to your point, like the the initial crimes against 0:10:17.679 --> 0:10:19.360 robot that we can in the robots that we can 0:10:19.400 --> 0:10:22.640 imagine would be stuff like drunk people pushing them over 0:10:22.960 --> 0:10:25.400 things like that. Yeah, or just like a human driver 0:10:25.520 --> 0:10:28.599 and a human powered vehicle hitting an autonomous vehicle. You know, 0:10:29.040 --> 0:10:31.280 right now, As I mentioned in the last episode, this 0:10:31.400 --> 0:10:33.200 is a very big paper and we're not gonna have 0:10:33.280 --> 0:10:35.880 time to get into every avenue they go down in it. 0:10:35.920 --> 0:10:38.120 But I just wanted to go through uh and and 0:10:38.200 --> 0:10:40.880 mention some ideas that stuck out to me is interesting 0:10:41.520 --> 0:10:45.480 that they discuss. And one thing that really fascinated me 0:10:45.520 --> 0:10:50.560 about this was that the idea of robots as possible 0:10:50.600 --> 0:10:54.840 agents in in a legal context UH brings to the 0:10:55.000 --> 0:10:59.360 for a philosophical argument that has existed in the realm 0:10:59.400 --> 0:11:01.880 of substance of law for a while. Uh. And I'll 0:11:01.880 --> 0:11:03.760 try not to be too dry about this, but I 0:11:03.840 --> 0:11:07.439 think it actually does get to some really interesting philosophical territory. Uh. 0:11:07.480 --> 0:11:10.400 And this is the distinction between what Limbly and Casey 0:11:10.480 --> 0:11:16.120 call the normative versus economic interpretations of substantive law. Again, 0:11:16.240 --> 0:11:19.360 complicated philosophical and legal distinction. I'll try to do my 0:11:19.400 --> 0:11:22.880 best to sum it up simply. So. The normative perspective 0:11:22.960 --> 0:11:27.880 on substantive law says that the law is a prohibition 0:11:28.000 --> 0:11:31.760 against doing something bad. So when something is against the law, 0:11:32.040 --> 0:11:35.880 that means you shouldn't do it, And we would stop 0:11:35.920 --> 0:11:38.680 the offender from doing the thing that's against the law 0:11:38.720 --> 0:11:41.600 if we could. But since we usually can't stop them 0:11:41.640 --> 0:11:44.800 from doing it, often because it already happened, the remedy 0:11:44.920 --> 0:11:48.439 that exists, You know that maybe paying damages to the 0:11:48.520 --> 0:11:51.520 victim or something like that is a is a an 0:11:51.559 --> 0:11:54.760 attempt to right the wrong, in other words, to do 0:11:55.000 --> 0:11:58.000 the next best thing to undoing the harm in the 0:11:58.040 --> 0:12:02.360 first place. So, basically, getting into the idea of negative reinforcement, 0:12:02.720 --> 0:12:05.800 somebody or something did something bad, we can't we couldn't 0:12:05.800 --> 0:12:08.319 stop them from doing something bad, but we can try 0:12:08.760 --> 0:12:12.959 and and give them stimulus that would make them not 0:12:13.080 --> 0:12:16.920 do it again. Be that economic or otherwise. Well, yes, 0:12:17.000 --> 0:12:19.199 but I think what you're saying uh could actually apply 0:12:19.240 --> 0:12:21.760 to both of these conditions I'm going to talk about. 0:12:21.800 --> 0:12:25.600 So I think maybe the distinction comes in about whether 0:12:25.760 --> 0:12:28.960 the whether there is such a thing as an inherent prohibition. 0:12:29.120 --> 0:12:32.280 So the thing that's uh operative in the normative view 0:12:32.360 --> 0:12:35.000 is that the thing that's against the law is a 0:12:35.040 --> 0:12:39.680 thing that should not be done, and thus the remedy 0:12:39.840 --> 0:12:42.080 is an attempt to try to fix the fact that 0:12:42.120 --> 0:12:44.880 it was done in the first place. The the economic 0:12:45.040 --> 0:12:47.959 view is the alternative here, and the way they sum 0:12:48.040 --> 0:12:52.960 that up is there is no such thing as forbidden conduct. Rather, 0:12:53.120 --> 0:12:56.920 a substantive law tells you what the cost of the 0:12:57.000 --> 0:13:01.360 conduct is. Does that distinction make any more sense? Yes, yes, 0:13:01.400 --> 0:13:06.480 So it's basically the first version is doing crimes is bad. Um. 0:13:06.640 --> 0:13:09.640 The second one is doing crimes is expensive. So it's 0:13:09.720 --> 0:13:12.760 the first is crimes should not be done, and the 0:13:12.840 --> 0:13:15.280 second one is crimes can be done if you can 0:13:15.320 --> 0:13:18.839 afford it. Yes, exactly so. In Limley, in cases words, 0:13:18.880 --> 0:13:22.480 quote damages on this view the economic view are simply 0:13:22.520 --> 0:13:26.800 a cost of doing business. One we want defendants to internalize, 0:13:27.080 --> 0:13:31.440 but not necessarily to avoid the conduct altogether. And now 0:13:31.559 --> 0:13:33.840 you might look at this and think, oh, okay, well, 0:13:33.880 --> 0:13:37.040 so the economic view is just like a psychopathic way 0:13:37.080 --> 0:13:39.640 of looking at things, And in a certain sense, you 0:13:39.640 --> 0:13:41.520 could look at it that as like if you're calculating 0:13:41.520 --> 0:13:45.400 what's the economic cost of murder? Then yeah, okay, that 0:13:45.520 --> 0:13:49.040 does just like that's evil, that's like psychopathic. But they're 0:13:49.040 --> 0:13:51.959 actually all kinds of cases we're thinking about. The economic 0:13:52.080 --> 0:13:55.160 view makes more sense of the way we actually behave 0:13:55.640 --> 0:13:59.280 And they use the example of stopping at a traffic light. Yes, 0:13:59.640 --> 0:14:02.480 so read from Limely in casey here quote. Under the 0:14:02.640 --> 0:14:06.320 normative view, a red light stands as a prohibition against 0:14:06.400 --> 0:14:09.480 traveling through an intersection, with the remedy being a ticket 0:14:09.600 --> 0:14:12.559 or a fine against those who are caught breaking the prohibition. 0:14:13.160 --> 0:14:15.839 We would stop you from running the red light if 0:14:15.840 --> 0:14:19.160 we could, but because policing every intersection in the country 0:14:19.200 --> 0:14:22.800 would be impossible, we instead punish those we do catch 0:14:22.920 --> 0:14:26.120 in hopes of deterring others. So in this first case, 0:14:26.240 --> 0:14:29.120 you running a red light is bad, you should not 0:14:29.240 --> 0:14:31.480 do it, and the cost of doing it, you know, 0:14:31.520 --> 0:14:34.120 the punishment you face for doing it is an attempt 0:14:34.160 --> 0:14:37.240 to right that wrong. But then they say, under the 0:14:37.280 --> 0:14:41.040 economic view, however, and absolute prohibition against running red lights 0:14:41.080 --> 0:14:44.520 was never the intention. Rather, the red light merely signals 0:14:44.560 --> 0:14:47.800 a consequence for those who do, in fact choose to 0:14:47.840 --> 0:14:51.120 travel through the intersection. As in the first instance, the 0:14:51.200 --> 0:14:54.080 remedy available is a fine or a ticket. But under 0:14:54.120 --> 0:14:56.280 this view, the choice of whether or not to violate 0:14:56.320 --> 0:14:59.120 the law depends on the willingness of the lawbreaker to 0:14:59.240 --> 0:15:03.040 accept the penalty. So in the case of a red light, 0:15:03.080 --> 0:15:05.640 well that that might make more sense if you're like 0:15:06.040 --> 0:15:08.360 sitting at a red light and you look around and 0:15:08.400 --> 0:15:11.320 there are no other cars anywhere near you, and you've 0:15:11.360 --> 0:15:14.320 you've got a clear view of the entire intersection and 0:15:14.360 --> 0:15:17.200 the red lights not changing, and you think maybe it's broken, 0:15:17.600 --> 0:15:19.400 and you're just like, Okay, I'm I'm just going to 0:15:19.520 --> 0:15:21.840 drive through. Well, if if you reach that point where 0:15:21.880 --> 0:15:24.200 you're like, I think it's broken, that I feel like 0:15:24.200 --> 0:15:26.520 that's a slightly different case. But if you're just like, 0:15:26.600 --> 0:15:30.560 nobody's watching, I'm gonna do it, um and and and 0:15:30.720 --> 0:15:33.480 the light isn't taking an absurd amount of time or 0:15:33.640 --> 0:15:36.480 longer than you're you're accustomed to, Yeah, I don't know 0:15:36.600 --> 0:15:39.440 how the the belief that the light is broken would 0:15:39.480 --> 0:15:42.280 factor into that, but yeah it is. I mean one 0:15:42.320 --> 0:15:44.720 thing that I think is clear that in it's that 0:15:44.800 --> 0:15:49.320 in many cases there are people, especially I think companies 0:15:49.360 --> 0:15:53.080 and corporations that operate on the economic view. And it 0:15:53.240 --> 0:15:56.840 is something that I think people generally look at and say, Okay, 0:15:56.880 --> 0:15:59.280 that that's kind of grimy. Like it like a company 0:15:59.320 --> 0:16:02.840 that says, Okay, there is a fine for not obeying 0:16:02.960 --> 0:16:06.440 this environmental regulation, and we're going to make more money 0:16:06.480 --> 0:16:09.720 by violating the regulation than we would pay in the fine. Anybody, 0:16:09.760 --> 0:16:11.800 So we're just gonna pay it. Yeah, you hear about 0:16:11.800 --> 0:16:14.760 that with factories, for instance, where where there'll, yeah, there'll 0:16:14.760 --> 0:16:17.400 be some situation where that the fine is not significant 0:16:17.560 --> 0:16:20.240 enough to really be at a turrent. It's just a 0:16:20.680 --> 0:16:24.840 fit for them breaking that that mandate being called on it. Occasionally. 0:16:24.840 --> 0:16:27.960 It's just the cost of doing business. Right. Uh. So 0:16:28.040 --> 0:16:30.320 there's a funny way to describe this point of view 0:16:30.320 --> 0:16:32.080 that the authors bring up here that they call it 0:16:32.120 --> 0:16:36.680 the bad man theory. And this comes from Justice Oliver 0:16:36.760 --> 0:16:40.040 Wendell Holmes, who is a U. S. Supreme Court justice. Uh. 0:16:40.120 --> 0:16:43.640 And he's talking about the economic view of substantive law. Uh. 0:16:43.680 --> 0:16:46.160 And Holmes wrote, quote, if you want to know the 0:16:46.240 --> 0:16:48.400 law and nothing else, you must look at it as 0:16:48.480 --> 0:16:52.440 a bad man who cares only for the material consequences 0:16:52.440 --> 0:16:55.520 which such knowledge enables him to predict, not as a 0:16:55.560 --> 0:16:59.280 good one who finds his reasons for conduct, whether inside 0:16:59.280 --> 0:17:02.160 the law or outside of it, in the vaguer sanctions 0:17:02.200 --> 0:17:05.240 of conscience. Uh. And so they write, the measure of 0:17:05.280 --> 0:17:07.520 the substantive law, in other words, is not to be 0:17:07.560 --> 0:17:11.239 mixed up with moral qualms, but is simply coextensive with 0:17:11.320 --> 0:17:14.360 its remedy. No more and no less. It just is 0:17:14.440 --> 0:17:17.280 what the remedy is. It's the cost of doing business. Now. 0:17:17.280 --> 0:17:19.879 Of course, there are plenty of legal scholars and philosophers 0:17:19.880 --> 0:17:22.800 who would dispute how Holmes thinks of this. But the 0:17:22.840 --> 0:17:26.280 interesting question is how does this apply to robots? If 0:17:26.320 --> 0:17:30.960 you're programming a robot to behave well, you actually don't 0:17:31.040 --> 0:17:34.280 get to just sort of like jump over this distinction 0:17:34.359 --> 0:17:36.840 the way humans do when they think about their own 0:17:36.880 --> 0:17:39.640 moral conduct. Right, Like, you're not sitting when you're trying 0:17:39.640 --> 0:17:41.879 to think what's a good way to be a good person. 0:17:42.400 --> 0:17:44.840 You're not sitting around thinking about well, am I going 0:17:44.880 --> 0:17:48.080 by the normative view of morality or the economic view 0:17:48.119 --> 0:17:51.160 of morality? You know? Um, you just sort of act 0:17:51.200 --> 0:17:53.119 a certain way whatever it seems to you the right 0:17:53.160 --> 0:17:55.199 way to do. But if you're trying to program a 0:17:55.280 --> 0:17:57.639 robot to behave well, you have to make a choice 0:17:57.720 --> 0:18:00.960 whether to embrace the normative view or the economic view. 0:18:01.119 --> 0:18:04.439 Does a robot view a red light, say, as a 0:18:04.600 --> 0:18:08.200 firm prohibition against forward movement, it's just a bad thing 0:18:08.280 --> 0:18:10.360 and you shouldn't do it to drive through a red light? 0:18:10.960 --> 0:18:14.800 Or does it just view it as a substantial discouragement 0:18:14.880 --> 0:18:18.040 against forward motion that has a certain cost, and if 0:18:18.119 --> 0:18:22.119 you were to overcome that cost, then you drive on through. Yeah, 0:18:22.160 --> 0:18:24.280 this is a great, great question because I feel like 0:18:24.280 --> 0:18:27.080 with humans, we're probably mixing a match and all the time, 0:18:27.680 --> 0:18:30.960 you know, perhaps you even done the same law breaking behavior. 0:18:31.040 --> 0:18:32.920 You know, we may do both on on one thing, 0:18:33.200 --> 0:18:34.919 and we do one on another thing, and then the 0:18:34.960 --> 0:18:37.159 other one on as still a third thing. But with 0:18:37.200 --> 0:18:39.639 the robot, it seems like you're gonna deal more or 0:18:39.720 --> 0:18:44.480 less with kind of an absolute direction. Either they're going 0:18:44.520 --> 0:18:48.000 to be um either the law is is to be 0:18:48.040 --> 0:18:50.760 obeyed or the law is to be taken into your 0:18:50.760 --> 0:18:53.040 cost analysis. Well, yeah, so they talk about how the 0:18:53.080 --> 0:18:57.120 normative view is actually very much like uh, Isaac Asimov's 0:18:57.200 --> 0:19:02.000 Laws of Robotics, inviolable rules, and the the Asimov story 0:19:02.080 --> 0:19:06.160 is doing a very good job of demonstrating why inviolable 0:19:06.240 --> 0:19:10.200 rules are really difficult to implement in the real world. 0:19:10.280 --> 0:19:14.240 Like you know that they Asimov explored this brilliantly, and 0:19:14.320 --> 0:19:16.560 along these lines, the authors here argued that there there 0:19:16.560 --> 0:19:19.080 are major reasons to think it will just not make 0:19:19.160 --> 0:19:23.200 any practical sense to program robots with a normative view 0:19:23.200 --> 0:19:26.560 of legal remedies. That probably when people make AI s 0:19:26.600 --> 0:19:28.679 and robots that that have to take these kind of 0:19:28.720 --> 0:19:32.560 things into account, they're almost definitely going to program them 0:19:32.600 --> 0:19:35.920 according to the the economic view, right. Uh, They say 0:19:35.920 --> 0:19:38.520 that quote encoding the rule don't run a red light 0:19:38.560 --> 0:19:42.480 as an absolute prohibition, for example, might sometimes conflict with 0:19:42.520 --> 0:19:45.560 the more compelling goal of not letting your driver die 0:19:45.760 --> 0:19:50.160 by being hit by an oncoming truck. So the robots 0:19:50.160 --> 0:19:54.400 are probably going to have to be economically economically motivated 0:19:54.440 --> 0:19:57.560 to an extent like this. Um. But then they talk 0:19:57.600 --> 0:20:00.800 about how you know, this gets very complicated because robots 0:20:00.840 --> 0:20:05.479 will calculate the risks of reward and punishment with different 0:20:05.560 --> 0:20:08.800 biases than humans, or maybe even without the biases that 0:20:08.880 --> 0:20:11.840 humans have that the legal system relies on in order 0:20:11.880 --> 0:20:16.120 to keep us obedient. Humans are highly motivated, usually by 0:20:16.119 --> 0:20:19.320 certain types of punishments that like, you know, humans like 0:20:19.440 --> 0:20:23.239 really don't want to spend a month in jail, you know, 0:20:23.280 --> 0:20:26.000 most of the time. And you can't just rely on 0:20:26.160 --> 0:20:29.480 a robot to be incredibly motivated by something like this, 0:20:29.760 --> 0:20:31.520 first of all, because like it wouldn't even make sense 0:20:31.560 --> 0:20:34.360 to send the robot itself to jail. So you need 0:20:34.440 --> 0:20:39.360 some kind of organized system for making a robot understand 0:20:39.440 --> 0:20:42.760 the cost of bad behavior in a systematized way that 0:20:42.920 --> 0:20:46.880 made sense to the robot as a as a demotivating incentive. Yeah, 0:20:47.000 --> 0:20:49.760 Like shame comes to mind as another aspect of of 0:20:49.840 --> 0:20:52.240 all this, Like how do you shame a robot? Do 0:20:52.240 --> 0:20:55.200 you have to program a robot to feel shamee and 0:20:55.320 --> 0:20:57.880 being uh, you know, made to give a public apology 0:20:57.960 --> 0:21:00.280 or something? Yeah? Uh so, so they are you that 0:21:00.280 --> 0:21:02.520 that it really only makes sense for robots to look 0:21:02.520 --> 0:21:05.560 at legal remedies in an economic way, and then they 0:21:05.640 --> 0:21:09.399 write quote it thus appears that Justice Holmes, archetypical bad man, 0:21:09.440 --> 0:21:13.159 will finally be brought to corporeal form, though ironically not 0:21:13.280 --> 0:21:16.119 as a man at all. And if Justice Holmes metaphorical 0:21:16.160 --> 0:21:21.080 subject is truly morally impoverished and analytically deficient, as some accused, 0:21:21.119 --> 0:21:25.440 it will have significant ramifications for robots. But yeah, thinking 0:21:25.440 --> 0:21:28.439 about these incentives, it gets more and more difficult, Like 0:21:28.520 --> 0:21:31.760 the more you try to imagine the particulars humans have 0:21:32.119 --> 0:21:35.399 self motivations, you know, pre existing motivations that can just 0:21:35.480 --> 0:21:38.560 be assumed. In most cases, humans don't want to pay 0:21:38.600 --> 0:21:41.320 out money, Humans don't want to go to jail. How 0:21:41.320 --> 0:21:45.560 would these costs be instantiated as motivating for robots? You 0:21:45.560 --> 0:21:48.840 would have to you would have to basically force some 0:21:49.040 --> 0:21:52.960