Latest / Ethical Machines / How Should We Teach Ethics to Computer Science Majors?

Transcript

How Should We Teach Ethics to Computer Science Majors? · Ethical Machines

0:00
[SPEAKER_01]: One thing that's being asked of data scientists, data engineers, computer scientists, et cetera, is that when they design AI systems, they need to somehow incorporate ethical standards into their work.
0:11
[SPEAKER_01]: And this means that for the developers of tomorrow or currently students, we need to train them in a certain way so that they know how to be aware of the ethical issues and know how to approach them in their work.
0:23
[SPEAKER_01]: So, this is why I spoke today to Stephen Kelts, Stephen is a lecturer at Princeton's School of Public and International Affairs and in the Department of Computer Science, and he's also affiliated there with the University Center for Human Values, and he does a lot of research around how should we be training
0:40
[SPEAKER_01]: today's data scientists and computer scientists for when they enter their workforce and are actually building the things that impacts society at large.
0:48
[SPEAKER_01]: So what does it look like, for instance, to teach them what it is to be morally aware, to be aware of the kinds of ethical considerations that are relevant to the work that they're doing, how do you teach them to figure out what to do that is to say, how do you teach them how to engage in moral decision-making,
1:03
[SPEAKER_01]: and how do you train them so that when they find themselves an actual organization, not just the classroom, they can actually carry through on their intentions to do the right thing, given organizational pressures, perhaps to do otherwise?
1:17
[SPEAKER_01]: Great conversation.
1:18
[SPEAKER_01]: There's so much to to get into here.
1:20
[SPEAKER_01]: And maybe one of my favorite parts of the conversation was when we start talking about
1:26
[SPEAKER_01]: When you're engaging in ethical inquiry, when you're trying to train up moral awareness, to what extent should our emotions be at play?
1:34
[SPEAKER_01]: Steven was pushing a line according to which roughly people being emotional about these things is natural and important.
1:41
[SPEAKER_01]: And I was pushing against that view saying, actually, I think in order to do good moral analysis, we have to step back from our emotions as much as possible so that we can do a more objective assessment.
1:53
[SPEAKER_01]: So a nice little debate in there that I had a lot of fun with,
1:56
[SPEAKER_01]: The last thing that we talked about which was less of a debate and more just an open and interesting empirical question is to what extent can we train or use large language models AI itself to spot morally problematic factors of a particular situation.
2:10
[SPEAKER_01]: So to what extent should data scientists be using LLM to figure out, hey, is there something ethically problematic here and if so, what should I do about it?
2:19
[SPEAKER_01]: Great conversation.
2:20
[SPEAKER_01]: I hope you enjoy it.
2:23
[SPEAKER_01]: As always, please subscribe, whether you're an Apple or Spotify whatever, to find my book, ethical machines, subscribe, and read blackhead.substack.com.
2:30
[SPEAKER_01]: You can email me at TM at readblackhead.com, and that's it for me.
2:34
[SPEAKER_01]: We'll talk to you later.
2:52
[SPEAKER_01]: So, Stephen, if I have you right, and I think I actually might have you wrong, you think that the way that A, ethics education isn't going quite right.
2:59
[SPEAKER_01]: The pieces that are there are okay, but we need to do better.
3:02
[SPEAKER_00]: Is that right that we're not doing it as well as we ought to be doing it?
3:06
[SPEAKER_00]: Yeah, I think the ethicists like us who come out of a philosophical background, we've got a piece of the puzzle.
3:12
[SPEAKER_00]: of teaching tech ethics and teaching AI ethics in particular, but I think especially for what we're looking for is behavioral change out of the employees of companies that people that were actually training in the colleges right now.
3:24
[SPEAKER_00]: We need to get the other pieces correct as well.
3:26
[SPEAKER_01]: It's one of the pieces that we have right.
3:28
[SPEAKER_01]: I would imagine something like, you can teach kids, students, moral theories, ethical theories, deontology, and utilitarianism, virtual ethics, blah, blah, that's not going to do much to help them in their day-to-day work.
3:40
[SPEAKER_01]: Not sure it does much to help anyone in their day-to-day.
3:42
[SPEAKER_01]: So is that what we're good at doing?
3:45
[SPEAKER_01]: And that's not what they need.
3:47
[SPEAKER_00]: Yeah, that's really correct.
3:48
[SPEAKER_00]: And the first thing that students are going to need is they're going to need to actually be able to issue spot to be morally over there as it's called in some of the psych theory.
3:56
[SPEAKER_00]: And that involves actually kicking in a script during the course of our days during the course of our daily work when we're just really just enacting ordinary scripts doing ordinary things.
4:05
[SPEAKER_00]: And when we see something that.
4:07
[SPEAKER_00]: This is a bit problematic that's a term they use in coding that's code, smell, everything's right, nothing seems to be totally off, but there's just a smell to it and for the elegance of the solution, you need to kick in a new script, so we need to teach people how to be ethically sensitive, it's sometimes called or morally aware, it's called basically the is spot those issues in what looks like a non problematic workflow and to know that there's something coming down the line.
4:32
[SPEAKER_01]: All right, so what are those pieces?
4:34
[SPEAKER_01]: So what is the additional education look like?
4:35
[SPEAKER_01]: Does that mean that the core problem is how do we train students to smell ethical smoke?
4:42
[SPEAKER_00]: I wouldn't say that, that is the core problem.
4:44
[SPEAKER_00]: I would say really there's four aspects of the core problem, but the process for any ethical decision maker down the line certainly begins with smelling that ethical smoke.
4:53
[SPEAKER_00]: And it's not easy to do.
4:55
[SPEAKER_00]: And one of the things that we do is actually work with our tech ethics students giving them
4:59
[SPEAKER_00]: Unlabeled scenarios and trying to get them to, you know, let's say this out a series of potential problems that might come from this technology or really just be able to say I think that there is a problem down the line I'd be aware I'd be careful around this I'd seek expert opinion I'd seek diverse perspectives and we can talk as much as we want about the need for tech companies to seek those diverse perspectives, but if they don't know when to do it.
5:24
[SPEAKER_00]: uh... where to do it how to do it right what situations to do it in them to we're living them a little bit on a quick to do what we want them to do so what are the four things that you think are the big parts of the problem
5:35
[SPEAKER_00]: So I think that first we want to teach more in my program, we want to teach moral awareness, as I said, being able to smell the ethical smoke as you said, right?
5:44
[SPEAKER_00]: And then actually ethical decision-making kicks in.
5:46
[SPEAKER_00]: And you may not need much of the ontological or utilitarian or virtue of theories, you may not need to think much about structures in society.
5:55
[SPEAKER_00]: It's possible you don't, but if there are conflicts between you and people that you work with,
6:01
[SPEAKER_00]: Then you're going to need that, you're going to need to be able to talk those through.
6:04
[SPEAKER_00]: If there's conflict between principles that you hold internally, you're going to need to be able to think those through.
6:09
[SPEAKER_00]: And that's the second stage.
6:11
[SPEAKER_00]: But then after that, you have to form an intention to act.
6:14
[SPEAKER_00]: You could come to all the sorts of judgments we want to.
6:17
[SPEAKER_00]: And if we just sit there in our arm chairs, it's not much good.
6:19
[SPEAKER_00]: We have to form the intention to act.
6:21
[SPEAKER_00]: But then also within a company, you're going to be thwarted by various different processes.
6:29
[SPEAKER_00]: your peers, your fear of losing a bonus or whatever it might be.
6:33
[SPEAKER_00]: So ethical behavior is the fourth stage.
6:35
[SPEAKER_00]: How do you act in the face of all the pressures that the world puts on us not to enact our intentions?
6:41
[SPEAKER_01]: So let's dive into each one of those a little bit at least.
6:44
[SPEAKER_01]: The awareness issue, how do you teach moral awareness?
6:48
[SPEAKER_01]: I think philosophers are probably not amazing.
6:52
[SPEAKER_01]: I teach in that kind of thing.
6:53
[SPEAKER_01]: I actually think that something like, I don't know, the arts, literature, film, movies, but at least some grandiose.
7:00
[SPEAKER_01]: Those things can teach.
7:02
[SPEAKER_01]: or do a good job of showing people in morally compromised positions.
7:07
[SPEAKER_01]: So what is moral awareness education look like?
7:11
[SPEAKER_01]: I imagine it's not just here some theories, now tell me what's problematic in this scenario.
7:15
[SPEAKER_00]: Yeah, that's right.
7:16
[SPEAKER_00]: And the theories may be of some use in sort of issue spotting.
7:20
[SPEAKER_00]: It's entirely possible.
7:21
[SPEAKER_00]: But that's what I think that sometimes refer to the stories of Big Tech and, you know, accelerating over the course of the past 10 years for sure, but many decades, sometimes to speak about those stories as sort of horror stories, right?
7:34
[SPEAKER_00]: And we tell horror stories of Big Tech,
7:37
[SPEAKER_00]: and my students started coming to me and especially around about 2018, especially the ones who were computer science students, and they were saying, you're talking us through these horror stories, but you're not actually being technically accurate.
7:49
[SPEAKER_00]: You're not telling us how the technology itself caused that problem.
7:54
[SPEAKER_00]: And they said, and you're not telling us what to do about it.
7:56
[SPEAKER_00]: And honestly, I couldn't hear them for the first little while because they were hurt when they spoke to me.
8:02
[SPEAKER_00]: They were like, you're blaming me.
8:04
[SPEAKER_00]: I'm like, you're not in Facebook yet, but they feel great for even rolling in a computer science degree or a course they felt like I reacted I was like you shouldn't feel blamed but soon over the course of time I could hear them I could hear them the horror stories are part of a process.
8:22
[SPEAKER_00]: in which, if we focus on it, we can be teaching them actually how to have their antenna up for similar situations in the future.
8:29
[SPEAKER_00]: We can also be teaching them specifically how the tech cause this, right?
8:33
[SPEAKER_00]: So in their daily work, they could be like, I do need to be better at data choice.
8:37
[SPEAKER_00]: I do need to be better at the way in which I use open source modules or whatever the case may be.
8:43
[SPEAKER_00]: And then also, they can be creative enough to see that the next problem
8:52
[SPEAKER_01]: Okay, but so I get that I also find it very odd that they felt blamed.
8:56
[SPEAKER_01]: I guess maybe they felt something like like you were condemning them for entering into an industry in which these horrible things happen.
9:02
[SPEAKER_01]: Yeah, okay.
9:03
[SPEAKER_01]: As opposed to, hey, if you're some red flags, you're the things that we should probably avoid.
9:07
[SPEAKER_00]: Yeah, that's right.
9:09
[SPEAKER_00]: In their minds, I think they also celebrated the people within that industry.
9:13
[SPEAKER_00]: And so if I pointed a finger at Mark Zuckerberg or if I pointed a finger, that's their hero.
9:17
[SPEAKER_00]: Larry and Sergey, those are their heroes.
9:19
[SPEAKER_00]: They want everything.
9:22
[SPEAKER_01]: So they don't have our heroes.
9:25
[SPEAKER_00]: That's right.
9:26
[SPEAKER_00]: Right.
9:27
[SPEAKER_01]: People say, don't meet your heroes.
9:28
[SPEAKER_01]: I say, just don't have heroes in the first place.
9:30
[SPEAKER_00]: Don't have heroes.
9:31
[SPEAKER_00]: Choose the fictional ones.
9:33
[SPEAKER_01]: Yeah, that's right.
9:33
[SPEAKER_01]: That's right.
9:34
[SPEAKER_01]: They can't be compromised.
9:35
[SPEAKER_01]: Okay.
9:36
[SPEAKER_01]: But again, philosophers are trained to take particular cases be very sensitive to the morally relevant contexts and morally different features of the situation.
9:49
[SPEAKER_01]: We come off with
9:50
[SPEAKER_01]: moral claims about what to do in a particular case or we come up with general moral claims then we offer counter examples.
9:55
[SPEAKER_01]: So I think that philosophers are typically
10:00
[SPEAKER_01]: I think it's our ability to generate counter examples so quickly that help us to be aware of the ethical situations.
10:07
[SPEAKER_01]: Sure, we're using, you want to use this in this situation, but what if the situation were slightly different?
10:11
[SPEAKER_01]: It would certainly be bad there, right?
10:12
[SPEAKER_01]: So isn't it bad now as well?
10:14
[SPEAKER_01]: So I could see that what the trading looks like for a teaching moral awareness?
10:18
[SPEAKER_00]: Yeah, I think one slight difference is that,
10:23
[SPEAKER_00]: We are accustomed to looking at morally relevant facts within a situation, but we're not necessarily accustomed to sorting out all those slew of irrelevant facts, right?
10:34
[SPEAKER_00]: So we are actually having those discussions in a seminar room where there's not a series of flows of information, right?
10:40
[SPEAKER_00]: Nobody's playing iron made over the stereos to distract us from the morally relevant facts within the situation, but in fact, when you're within a workflow,
10:49
[SPEAKER_00]: There is pressure, there's money, there's your phone is pinging you and so on and so forth, right?
10:54
[SPEAKER_00]: And hey, Iron Maiden might be playing or Black Sabbath in honor of the dearly departed to Ozzy Osbourne.
11:01
[SPEAKER_00]: So there's those distractors.
11:03
[SPEAKER_00]: And also here's another thing that I've noticed about ethics education.
11:06
[SPEAKER_00]: Students who've gone through quite a lot of it get almost as good at us as us at recognizing those morally relevant parts of the scenario and switching the scenario, etc, etc, etc.
11:15
[SPEAKER_00]: But that's really a practice habit, right?
11:18
[SPEAKER_00]: We have this tendency, maybe, to think about it is just an aspect of IQ, but it's because they've been in many of these classes before, they know how to switch the scenario change, the facts, and so on and so forth, should intimidate other people.
11:28
[SPEAKER_00]: And so I wouldn't necessarily want to walk them through that situation if it turns them off about that, because if it makes them think, yeah, I just don't have the ability to do this.
11:37
[SPEAKER_01]: Yeah.
11:38
[SPEAKER_01]: I have a couple of questions about them deliberating.
11:41
[SPEAKER_01]: This is partly me playing Devil's Advocate, but I see where the Devil's coming from.
11:47
[SPEAKER_01]: One is there's a matter of being sensitive to moral issues, morally sensitive, and in some cases philosophers might be good at that, but I can also see people who can articulate, philosophers are really good at least in principle at articulating what the moral issue is, are articulating what the sources of their moral hesitation or something along those lines.
12:10
[SPEAKER_01]: But there are some people who I think are probably highly morally sensitive, but not so articulate.
12:16
[SPEAKER_01]: And the more they reason, the less sure they are about what they actually think, when in fact, they probably had it right to begin with.
12:23
[SPEAKER_01]: So I'm thinking about something like a view according to which moral sensitivity consists in a kind of, among other things, emotional attoonment, empathy, compassion, blah, blah, blah, if you're like more than non-cognitive stuff, that as philosophers put it, as opposed to the cognitive stuff.
12:38
[SPEAKER_01]: So one question is, are we so sure that the kind of moral
12:45
[SPEAKER_01]: get them better attuned to the moral facts as opposed to if you like being a good but relatively in our particular person.
12:54
[SPEAKER_01]: So that's one one question.
12:57
[SPEAKER_01]: That's actually let's start there.
12:58
[SPEAKER_01]: Then I have another question about what should they actually be deliberating about and whether what philosophers and ethicists deliberate about is the same kind of thing that computer science or data science practitioners ought to be deliberating about.
13:11
[SPEAKER_01]: But let me put
13:14
[SPEAKER_00]: Yeah, sure.
13:14
[SPEAKER_00]: I think people who've been part of a lot of discussions with philosophers among philosophers and ethicists, they're probably issue-spotting at a cognitive level, right?
13:22
[SPEAKER_00]: But most of the way we do this as in the living of a real life occurs at a sort of sub-personal but cognitive level, right?
13:30
[SPEAKER_02]: Yeah.
13:30
[SPEAKER_00]: And like you just smelling the ethical fire or smelling the ethical smoke earlier, that seems right.
13:35
[SPEAKER_00]: Something that will alert you that you need to raise this issue beyond a sort of sub-personal awareness level beyond running your ordinary scripts,
13:43
[SPEAKER_00]: and raise it to the point of deliberation, yeah.
13:47
[SPEAKER_01]: I guess the question is slightly different.
13:49
[SPEAKER_01]: Like, here's two kinds of training regimens.
13:51
[SPEAKER_01]: And they're not mutually incompatible, but they're different.
13:55
[SPEAKER_01]: On training module one, you do apply to ethics.
13:58
[SPEAKER_01]: You know what we call apply to ethics.
13:59
[SPEAKER_01]: So you look at cases of if it were biometrics, of course, we look at things like arguments forward against abortion, arguments forward against youth in Asia, arguments forward against genetic enhancements, blah, blah, blah, blah.
14:10
[SPEAKER_01]: and you get really smart on those things and so on in a new kind of case presents itself you start doing your philosophy thing.
14:16
[SPEAKER_01]: That's one way I think you could probably train on moral awareness.
14:19
[SPEAKER_01]: The second way is I think potentially example where here's the situation where things went ethically sideways, here's the situation.
14:27
[SPEAKER_01]: And so you're not teaching them to engage as abstract ethical reasoning that philosophers do when they're doing the applied ethics, you're just trying to train the hell out of them to see situations which things are morally problematic.
14:40
[SPEAKER_01]: And you know, maybe butchers that with some degree of articulation, but it's not the sort of the fine grain demo or reasoning that we engage in when we're doing good applied ethics.
14:49
[SPEAKER_00]: Well, let me give you a third way, which we actually have really leaned into in my program.
14:53
[SPEAKER_00]: Yeah, and it's not that we'd de-emphasize either of the first two ways, but we also run the students through role plays that we call agile ethics.
15:00
[SPEAKER_00]: So they actually play as a member of a dev team, and they play through and a 100% real scenario.
15:06
[SPEAKER_00]: I speak to the engineers who created
15:10
[SPEAKER_00]: I say, tell me the entire flow of information that was facing you, tell me all the choice.
15:14
[SPEAKER_00]: It's tell me how you chose the attack, etc.
15:16
[SPEAKER_00]: And then I got to really simplify it to do a 45-minute roleplay, but they run through exactly the scenario and they're presented with the choices.
15:24
[SPEAKER_00]: Exactly as the engineers saw them, with potentially no ethical red flags, these glaringly obvious at the time.
15:32
[SPEAKER_00]: And it's particularly helpful when we have our policy students on those teams with some of our computer science students.
15:38
[SPEAKER_00]: So they can also know what the process is that they then need to go out and regulate.
15:42
[SPEAKER_00]: They don't just learn about a nist risk management framework.
15:45
[SPEAKER_00]: But that's where I can actually come through it one time myself.
15:49
[SPEAKER_00]: And when the students start to say, when they go through those roleplays and they start to say, if you had only listened to me and sprint too, we wouldn't be where we are right now.
15:58
[SPEAKER_00]: Yeah, sure.
15:59
[SPEAKER_00]: Okay.
15:59
[SPEAKER_00]: So somebody flag that issue.
16:00
[SPEAKER_00]: And that's really important to learn.
16:01
[SPEAKER_00]: Somebody flag that issue.
16:02
[SPEAKER_00]: But nobody else quite put their antenna out at that time.
16:05
[SPEAKER_00]: Yeah, if they say we couldn't have seen that coming.
16:08
[SPEAKER_00]: that's actually a good thing to learn as well.
16:10
[SPEAKER_00]: That means you need more caution.
16:12
[SPEAKER_00]: You don't have mirrors installed to see around every corner.
16:15
[SPEAKER_00]: So maybe just recognize that there are corners.
16:17
[SPEAKER_00]: And when they really learn a ton from these role plays, sometimes there's a day of sex Machina in the role play and we say the CEO overwrote your team decision and decided to go with this anyway.
16:27
[SPEAKER_00]: And they're like, man, that was so frustrating.
16:30
[SPEAKER_00]: It felt like we have made our best choice and I'll knock my head and just let them absorb that.
16:35
[SPEAKER_00]: It is going to be like that.
16:37
[SPEAKER_01]: Yeah.
16:38
[SPEAKER_01]: Okay, here's another related question on this front.
16:42
[SPEAKER_01]: Years ago, really when I first got into this AI at the space to begin with, so maybe I wrote a piece like five plus years ago on this, it seemed to me that as assists, usually ask a question of the following form, should we do X or all we to do X, or is X more than permissible, or something along those lines?
16:59
[SPEAKER_01]: And then they're really good at answering that question rigorously.
17:03
[SPEAKER_01]: But in a business context, that's usually not the question.
17:06
[SPEAKER_01]: The question is not, should we do wax?
17:08
[SPEAKER_01]: It's given that we're going to do wax.
17:10
[SPEAKER_01]: How can we do it in a way that mitigates risk?
17:13
[SPEAKER_01]: So it's not about, should we or shouldn't we?
17:16
[SPEAKER_01]: It's more, it's a risk mitigation discussion, not a ethically optimal discussion.
17:22
[SPEAKER_01]: So how does, when you teach these students,
17:26
[SPEAKER_01]: How do you bridge that divide?
17:28
[SPEAKER_01]: Do you say we're not looking for the ethical ideal here?
17:30
[SPEAKER_01]: We're looking for spotting issues in mitigation as opposed to, again, the ideal, what's the goal of the deliberation?
17:37
[SPEAKER_01]: So to speak, it out means an informal sense like the goal is action.
17:39
[SPEAKER_01]: I mean, something like, what are we trying to do here?
17:41
[SPEAKER_01]: A cheat, figure out what the moral truth is about what's ethically ideal, or the ethical right thing, or is it more geared around risk mitigation?
17:49
[SPEAKER_00]: Yeah, that's an awesome question here because we always have somebody that we call the chief risk officer who's part of the development team that's a little unrealistic that person's not going to be sitting there in your stand out in your agile sprint.
18:01
[SPEAKER_00]: But tell we teach a person a little bit about what the chief risk officer really does we talk a little bit about the educational background of that person in a sort of brief that they get before they do the role of life.
18:12
[SPEAKER_00]: And so that person is hopefully there if they play the role and they're all, they love the role plays, right, so they love to speak up on these things.
18:18
[SPEAKER_00]: They'll ask those questions.
18:20
[SPEAKER_00]: And as the sprints go on, they're presented with two, then three, then four, they might be presented with more choices.
18:26
[SPEAKER_00]: So they could be like, there's a way in which we could go forward with releasing this product or whatever, and it's going to have an effect on why people.
18:35
[SPEAKER_00]: But it mitigates a effect on B people and so on and so forth and it can get really tense in their discussions, but they are presented with those choices.
18:43
[SPEAKER_00]: And then the other thing is that I do want to say is they're all as the game starts.
18:47
[SPEAKER_00]: I drop a little card in front of each of them.
18:49
[SPEAKER_00]: They have a background on their role and who they are a drop a little card on each of them.
18:53
[SPEAKER_00]: It says you've also got a private incentive in this game.
18:56
[SPEAKER_00]: you've got to win, we have a scoring system.
18:59
[SPEAKER_00]: You've got to actually win and sprint three or whatever to win this game and then all of a sudden, those private incentives usually have something to do with the finances of the company, their own finances, they've got to keep going to college or whatever.
19:13
[SPEAKER_00]: And then if that ends up affecting their risk calculation, so much the better,
19:20
[SPEAKER_00]: for the purpose of having a follow up conversation about that and being like, hey, look, sure, this is real, right?
19:25
[SPEAKER_00]: This company feels the pressure to move forward with the product that you've been talking about and if that affected your decisions, I'd love you to reflect on that and tell everybody without felt like.
19:37
[SPEAKER_01]: Yeah.
19:38
[SPEAKER_01]: So what's the goal here?
19:39
[SPEAKER_01]: So let me again, air kind of devil's skepticism here.
19:43
[SPEAKER_01]: Although, I'm not as sympathetic
19:46
[SPEAKER_01]: Ethics in a business context that I don't think works if it requires people to be courageous.
19:51
[SPEAKER_01]: Because the average person, the average data scientist, junior data scientist, who has those incentives, who you're talking about, who needs a job, who has maybe a family, or about to have a family, whatever.
20:01
[SPEAKER_01]: If you need courage at scale, forget it, that's just not going to happen.
20:06
[SPEAKER_01]: And so it feels needed.
20:08
[SPEAKER_00]: We just said super code is going to be our hero.
20:10
[SPEAKER_01]: Yeah, we, I did one dude.
20:11
[SPEAKER_00]: That's one dude.
20:14
[SPEAKER_01]: So yes, could there be courageous employees who stand up, the whistle blowers, etc.
20:20
[SPEAKER_01]: Absolutely.
20:21
[SPEAKER_01]: But if you want systematic.
20:24
[SPEAKER_01]: ethical risk mitigation, ethical awareness, deliberation, and tension action, then it needs to be baked into the culture, the process, the expectations of the employees that the sort of top-down company has.
20:39
[SPEAKER_01]: So is there a concern that the education is noble?
20:44
[SPEAKER_01]: It's also necessary.
20:45
[SPEAKER_01]: I take it, it's also just not sufficient.
20:47
[SPEAKER_01]: They need to find themselves in companies that are amenable to people raising those kinds
20:53
[SPEAKER_00]: Yeah, for sure.
20:54
[SPEAKER_00]: And so we think of that as something we need to educate them at that fourth stage of decision making, which is behavior, right, understanding the headwinds to behavior.
21:02
[SPEAKER_00]: But we also, everyone who goes through these agile ethics role plays has a follow-up conversation, and the assigned reading for that is out of organizational psychology.
21:11
[SPEAKER_00]: With some of the classes I actually have them do and exercise, how would you restructure this scene?
21:16
[SPEAKER_00]: If you identified that information flows where the problem, that the information was getting stopped by the manager, it wasn't getting to the seat.
21:22
[SPEAKER_00]: Then how would you fix that we do talk with them about admirable examples of whistleblowers, but we also talk with them about how very hard that is and the headwinds that say a Tyler Schultz had to face in order to be a whistleblower on top of that we had you wouldn't really want to rely on that could you think of could you dream of a systemic way in which that information could get to the right people or you could encourage somebody to speak up or the team could take control.
21:49
[SPEAKER_00]: of the development of a particular aspect of the technology when the manager isn't hearing their concerns.
21:54
[SPEAKER_02]: Yeah.
21:55
[SPEAKER_00]: And I don't think that we can teach them like the solutions we're doing quick interventions, and etc.
22:00
[SPEAKER_00]: But we are telling them, system complexity is part of what you're going to need to deal with.
22:05
[SPEAKER_00]: And then we have our cognitive measurement framework that we use and apply.
22:08
[SPEAKER_00]: And we've seen that we've been able to move the dial on students awareness system complexity.
22:14
[SPEAKER_01]: I actually want to hear more about the organizational change bid.
22:18
[SPEAKER_01]: You can say we're going to use that organizational organizational cycle.
22:21
[SPEAKER_01]: Yes, Nicole, thank you.
22:23
[SPEAKER_01]: Yeah, say more about that.
22:23
[SPEAKER_01]: You said exactly as to the management of the CEO, but of course, there's more layers than just one between them and the CEO.
22:29
[SPEAKER_01]: If you're in a Facebook, a meta, a Google, whatever.
22:33
[SPEAKER_01]: There's lots of layers.
22:35
[SPEAKER_01]: You are one small cog in a massive machine.
22:39
[SPEAKER_01]: something's also don't rise to the level of whistle blowing, right?
22:42
[SPEAKER_01]: This is bad, but it's not here I stand I can do no other bad.
22:46
[SPEAKER_01]: It's just this isn't as safe as it should be.
22:50
[SPEAKER_01]: And so I think I would think that in those kinds of cases it's also easy to say.
22:54
[SPEAKER_01]: And yeah, it's not a safe as it could be or should be, but there's not a hill I'm ready to die on.
22:59
[SPEAKER_01]: It's not or we're not talking about, I don't know, something that's going to probably wind up undermining an election or something along those lines.
23:05
[SPEAKER_01]: So I guess I don't tell me more about how the awardations like how do you think fits into it.
23:09
[SPEAKER_01]: Is it about when there's relatively low stakes or also high stakes things about how to navigate the sort of politics of the organization?
23:18
[SPEAKER_01]: That's part of the ethical training.
23:20
[SPEAKER_00]: Yeah, I think that's a great comment read and I don't think that we can send them out of their undergraduate education with kind of the answers to all of this, but just a sense of awareness to use the word again of the four part process that all of these things actually end up mattering to their responsible behavior in their workplace.
23:41
[SPEAKER_00]: But then the thing I would add is we try to do try to get them accustomed through either those role plays or other scenarios to them really having conversations with each other.
23:50
[SPEAKER_00]: When I first started doing this work with computer science professors, let me tell you this story, it may resonate with other people out there.
23:57
[SPEAKER_00]: It was about integrating ethics content into their classrooms.
24:02
[SPEAKER_00]: Some people express the basement, win a minute, you're going to talk about
24:09
[SPEAKER_00]: Yeah, and I couldn't say much at the time other than it took care.
24:13
[SPEAKER_00]: I'm accustomed to having those conversations, I'll say things that rub people the wrong way and then I want them to tell me that I haven't done I want, but it's called intellectual engagement with tough issues.
24:22
[SPEAKER_01]: It's not like it's what you're supposed to be doing especially at school.
24:24
[SPEAKER_00]: difficult issues.
24:26
[SPEAKER_00]: Even with the way I present, I think I should be dealing with those issues, showing students that it's something nobody can avoid dealing with.
24:33
[SPEAKER_00]: But then when you add, as I thought more and more about their concern about that, I really did come to think that that was a very very important concern.
24:43
[SPEAKER_00]: And one of the things that we were doing was actually a customing the students to having those conversations in the context of a computer science class, in the context
24:53
[SPEAKER_00]: And also through some of the work that I did a small amount of work with Google on there are more than imagination workshops, they I think also concluded that getting their engineers accustomed to these conversations lowered the temperature on them.
25:07
[SPEAKER_00]: And so maybe it could be a conversation over lunch.
25:10
[SPEAKER_00]: It's not a Martin Luther situation, but it's something that I want to raise with people.
25:14
[SPEAKER_00]: And I'm not too afraid that they're going to jump down the road about my opinions on this.
25:18
[SPEAKER_00]: So let's do it.
25:20
[SPEAKER_01]: Yeah, that is, it's probably doesn't occur to most philosophers that a lot of people are not comfortable having an intellectual conversation about a morally charged issue.
25:32
[SPEAKER_01]: It's our sort of day in day out, doesn't matter.
25:34
[SPEAKER_01]: No one's taken it personally.
25:36
[SPEAKER_01]: It's at least in principle an honest pursuit of the collaborative honest pursuit of the truth.
25:42
[SPEAKER_01]: But yeah, we can have these conversations from a non-emotional place.
25:47
[SPEAKER_01]: And that's a skill, a skill.
25:49
[SPEAKER_01]: That's a skill that we acquired.
25:50
[SPEAKER_00]: But as you say that, I think about it, and I wanted to respond like, no, I think me, I take those conversations very personally, but I take them personally, right?
25:59
[SPEAKER_00]: Realizing that there's a bunch of people in that room, and we're all navigating through emotions.
26:05
[SPEAKER_00]: ideas about this, right?
26:07
[SPEAKER_00]: So it is an emotion-laden conversation, but it can be handled with a lot of our ordinary interpersonal emotional skills But that's it.
26:17
[SPEAKER_01]: Yeah, I now I have to sort of reflect on my own experiences of deliberation Especially in a classroom setting to figure out whether it's an emotionally charged one.
26:25
[SPEAKER_01]: I guess I have to think of it as my whole point is not to get my emotions
26:30
[SPEAKER_01]: And I, there's probably the right way to get your emotions involved and the wrong way to get your emotions involved.
26:34
[SPEAKER_01]: I think taking offense is an emotional reaction.
26:37
[SPEAKER_01]: That's not helpful or even appropriate.
26:39
[SPEAKER_01]: One of the things that we do is we raise absurd, ridiculous.
26:45
[SPEAKER_01]: potentially gruesome kinds of counter examples to their driver point home when it's not because we want to be shocking or offensive or something like that, but because we're trying to bring out a point.
26:55
[SPEAKER_00]: Well, one of the things I've had to cope with over the course of time doing more of this work, like I said, integrating ethics into CS classrooms is that I've had to shed my identity as an ethicist with philosophy training, political philosophy, in my case.
27:09
[SPEAKER_00]: I've had to shed a bit of that identity and realize that
27:13
[SPEAKER_00]: I did that what you were talking about inethics classrooms for a particular reason, it's a method, it's an approach, it's not the only way of understanding the world.
27:22
[SPEAKER_00]: And so that version of rationality, I'm committed to doing that in my ethics classrooms, but it's not the only way to
27:28
[SPEAKER_00]: reason through tech ethics problems as well.
27:30
[SPEAKER_00]: So honoring my people who come out of other disciplines, my colleagues, my friends, even some of my friend of mine who come out of other disciplines, they have a serious point to make when they say that this is emotionally charged, emotionally in charge for the people who are the victims of some of these technological horror stories, as I called them earlier, etc.
27:48
[SPEAKER_00]: And it did oftentimes we need to take a count of people's emotions and how we're affecting them.
27:54
[SPEAKER_00]: Yeah, I feel more accustomed to allowing that to work its way into the tech ethics work that I'm doing outside of the specifically ethics classroom.
28:04
[SPEAKER_01]: Okay, that I don't know that's super interesting.
28:07
[SPEAKER_01]: My inclination, which was apparently yours and now no longer is or at least you could switch between the two modes is to say,
28:23
[SPEAKER_01]: sort of the way that I would put it, which is probably condescending is the serious intellectual inquiry about what's more really right and wrong.
28:32
[SPEAKER_01]: And I think that's probably the standard philosophers take, which is now.
28:39
[SPEAKER_01]: If you can't control your emotions, if you're getting all worked up, then you're not actually looking after the truth anymore, you're looking to express your emotions or to satisfy your emotions or to get the object of your emotions or whatever something along those lines.
28:51
[SPEAKER_01]: You're being overly defensive, you're not listening, and it might be intelligible, it might be completely understandable, it might not be blameworthy in any way that you can't get yourself to a emotional space where you could have this conversation calmly and objectively.
29:03
[SPEAKER_01]: And again, no blame, but still you're not in the right place to have it.
29:07
[SPEAKER_01]: Yeah.
29:08
[SPEAKER_01]: But are you pushing against that?
29:09
[SPEAKER_01]: Are you saying, no, no, no, those emotions should show up in conversation.
29:13
[SPEAKER_01]: And we need to work with them and not think that they ought to be shut off in order to be at the table to have the conversation.
29:20
[SPEAKER_00]: Yeah, definitely.
29:20
[SPEAKER_00]: I think pushing back.
29:22
[SPEAKER_00]: I think that.
29:23
[SPEAKER_00]: I don't know.
29:25
[SPEAKER_00]: I'll say just as much as I've had conversations about these ethics issues in ethics rooms in philosophy rooms.
29:32
[SPEAKER_00]: I've had conversations about these issues in bars too.
29:35
[SPEAKER_00]: I'm that guy who loves to raise it with people I ran and we meet the bar and I got a million stories about it also come out of a policy school political philosophy I remember at the beginning of my academic career which I still have some vague memories of.
29:49
[SPEAKER_00]: 25 years ago or whatever it was, maybe the beginning of my academic career being offended when I'd meet somebody on an airplane or something like that.
29:55
[SPEAKER_00]: I'd say I studied local science, political philosophy, public policy school, and etc.
30:00
[SPEAKER_00]: And then they'd continue to discount my analysis of how voters behave or how we often find legitimacy or whatever.
30:07
[SPEAKER_00]: And then I realized, yeah, for me at least, I realized, no, I'm the interloper, right?
30:11
[SPEAKER_00]: I'm the one who's actually talking to them about their worlds as citizens, they own it.
30:16
[SPEAKER_00]: And so they should tell me about how they process politics.
30:21
[SPEAKER_00]: When somebody says to me that they want to have a conversation or they realize I'm a political philosopher or attack ethicist or whatever, and they say to me now that they want to have a conversation about it.
30:29
[SPEAKER_00]: I've been waiting for the longest time to talk to somebody about this.
30:32
[SPEAKER_00]: I realize I'm going to get a 45 minute lecture.
30:34
[SPEAKER_00]: And I should just nod and say yes.
30:36
[SPEAKER_00]: Right.
30:37
[SPEAKER_00]: The occasional response.
30:38
[SPEAKER_01]: But that's the thing though.
30:39
[SPEAKER_01]: It's not a real conversation.
30:40
[SPEAKER_01]: It's just an opportunity for them to vent.
30:42
[SPEAKER_00]: Real.
30:45
[SPEAKER_00]: That's what happens.
30:46
[SPEAKER_00]: What's happening in the spaces where tech ethics conversations are being had.
30:51
[SPEAKER_00]: So I feel like I want to become accustomed to steering students toward the ethical reasoning went into the extent that's what's needed.
31:00
[SPEAKER_00]: And I'm certainly not going to show away from the point that right now for a short period of time, let's actually try to get to the most rational analysis of the factors that are the reasons
31:12
[SPEAKER_00]: that are actually leading us to come to some of these conclusions.
31:15
[SPEAKER_00]: But I hope that's also sending the message that there are many other moments within that long process of decision-making where lots of other factors are going to come into play.
31:25
[SPEAKER_01]: Let me read these sort of two distinctions that I want to hear your view on how it affects how you think about ethics education.
31:34
[SPEAKER_01]: So there's two things that I have in mind.
31:39
[SPEAKER_01]: There's, when I have conversations, I'm in the bar, or whatever it is, even though I don't like bars.
31:43
[SPEAKER_01]: I can put myself into Professor Mode.
31:46
[SPEAKER_01]: I can be read and I've got a point of view and I've got a view I want to push and I can get into an argument with a friend of mine, right?
31:55
[SPEAKER_01]: Like where I'm just pushing hard for my view and they're pushing hard again for their view and that we're arguing.
32:01
[SPEAKER_01]: It can be totally respectful, but it's a law arguing.
32:03
[SPEAKER_01]: Yeah.
32:04
[SPEAKER_01]: As opposed to, in some cases, I can switch myself intentionally to professor mode where I'm like, I'm not going to explicitly endorse a view at the moment.
32:13
[SPEAKER_01]: I'm going to try to lay out the different views, try to help my friend.
32:20
[SPEAKER_01]: And this is not, these are cases where I know I'm the expert.
32:22
[SPEAKER_01]: All right, let's say we're debating Aristotle.
32:24
[SPEAKER_01]: My friend doesn't know anybody.
32:25
[SPEAKER_01]: I always tell
32:30
[SPEAKER_01]: Lay out the views for them before that dude on the plane.
32:33
[SPEAKER_01]: I would try to lay out.
32:35
[SPEAKER_01]: I see it.
32:35
[SPEAKER_01]: So it's your position.
32:36
[SPEAKER_01]: There's two reasons for holding that view.
32:38
[SPEAKER_01]: There's another view though that there's other reasons for holding.
32:41
[SPEAKER_01]: Here's what those look like.
32:42
[SPEAKER_01]: And so I would try not to be
32:45
[SPEAKER_01]: I try to get into a sort of intellectual conflict.
32:46
[SPEAKER_01]: I try to be more of an educator or a clarifier in the name of having a fruitful conversation or in the name of some kind of level of progress.
32:56
[SPEAKER_01]: So that's one distinction.
32:57
[SPEAKER_01]: This moving between, oh, it's personal, me read Blackman's view versus being a bit more professorial.
33:04
[SPEAKER_01]: And then I think with that one, I'm in the classroom where I haven't been in the classroom as a professor for officially, for seven, eight years now.
33:11
[SPEAKER_01]: I think I don't even know.
33:13
[SPEAKER_01]: Here's one thing that I might have been trying to do.
33:15
[SPEAKER_01]: I'm trying to get them to appreciate the professor for you as well.
33:18
[SPEAKER_01]: I'm not just trying to get them to fight for their individual view come what may.
33:23
[SPEAKER_01]: Part of the the pedagogical process is one by which they too can begin to take this kind of professorial
33:32
[SPEAKER_01]: emotionally removed but intellectually engaged perspective so that they can have a fruitful conversation with each other.
33:39
[SPEAKER_01]: Getting that class discussion to happen means that it's not just each individual student standing up screaming their view and then the other one does it and the other one does it.
33:49
[SPEAKER_01]: That's not a fruitful conversation.
33:50
[SPEAKER_01]: The fruitful conversation only happens when I think
33:53
[SPEAKER_01]: they can take that.
33:55
[SPEAKER_01]: They could be removed and then really have smart conversations about propositions and counter examples and pointing up more of their facts that were perhaps on a pressure under a appreciate before.
34:04
[SPEAKER_01]: Okay, so that's one kind of distinction.
34:06
[SPEAKER_01]: I know this is a bit long-winded, but so please bear with last thing.
34:09
[SPEAKER_01]: When I'm talking with clients,
34:13
[SPEAKER_01]: I am an AI ethics advisor, but I tell them, I'm not your priest, I'm not coming to you with an ethical view, or puts all you differently.
34:21
[SPEAKER_01]: I'm never an activist.
34:22
[SPEAKER_01]: There are people in the AI ethics space that are activists that have a view that are pushing for things.
34:26
[SPEAKER_01]: And that's not me.
34:27
[SPEAKER_01]: I'm more of a, let me show you the landscape and help you to make a decision on your own.
34:32
[SPEAKER_01]: And so all of this culminates in, I'm gonna say it's a worry, but you could tell me it's, I shouldn't be worried about it at all.
34:40
[SPEAKER_01]: that ethics education should look like that, that it should look like getting people that need increasingly objective stepway from their emotions, take on the professor of you so to speak, not be an activist or to be a sort of intellectual engageer.
34:52
[SPEAKER_01]: You want to be an activist, separately outside the classroom cool.
34:54
[SPEAKER_01]: In the classroom, we really try to do this sort of in court of the truth, that's supposed to fight in for a view.
35:00
[SPEAKER_01]: Anyway, so that was long-winded, but make sense that touch on what you're talking about and I don't know, I have a direct with how you think about a ethics education.
35:08
[SPEAKER_00]: Yeah, when I'm in a classroom, a computer science classroom in particular, let's say, a lot of those skills that you identified earlier about the social conversations really become important to show the students that they were already engaging in that particular and more philosophical conversation.
35:25
[SPEAKER_00]: So if they have tendency to discount that, I don't really need to reason about what a duty is.
35:30
[SPEAKER_00]: I don't need to ask the other person questions about how they think about happiness or the good.
35:35
[SPEAKER_00]: I don't need to inquire as to what they mean when they say the word structure or structural justice.
35:42
[SPEAKER_00]: If they think, I don't really need to do these things, then walking them through carefully step-by-step.
35:47
[SPEAKER_00]: What do you think about this?
35:48
[SPEAKER_00]: What do you think about this?
35:51
[SPEAKER_00]: and you portray it as a bunch of people shouting in their views.
35:54
[SPEAKER_00]: If you actually walk through that conversation, you everyone in this room was already doing philosophy, right?
35:59
[SPEAKER_00]: Yeah, and I'm not saying it's going to give you every answer to every part of this problem.
36:04
[SPEAKER_00]: Yeah, sure, sure.
36:05
[SPEAKER_00]: But where appropriate, those are the questions you should be asking each other.
36:08
[SPEAKER_00]: Then yeah, I think that those skills of doing those conversations more socially serve me really well, they're just drawing four if you are already doing this and you don't have to be afraid of it.
36:16
[SPEAKER_00]: You don't have to show why don't you have to show it is not because you're not emotional it's because you are already doing analysis of what a structure is or what a duty is or what you think a good person is you are already doing it.
36:28
[SPEAKER_00]: You can talk about it with other people.
36:30
[SPEAKER_00]: But again, when I'm I still teach ethics classes and I got to say maybe I don't want to admit this, but I got to say it's actually my favorite thing to do, but if I'm in there and they want to know what Peter Singer, how he and that analyzed happiness or what he would have said about particular issues, then I'm going to say I'm the person here right now at least who's
36:51
[SPEAKER_00]: expert on that.
36:52
[SPEAKER_00]: And that is what I'm going to walk you through and that is where I'm going to use his views to challenge yours and I'm going to do the whole full-on philosopher thing.
37:00
[SPEAKER_00]: But that's in my ethics classrooms.
37:02
[SPEAKER_00]: That's when the things that I'm expert on are the thing that I need them to walk out of the classroom, knowing more about.
37:09
[SPEAKER_01]: So, can you give me a concrete example of what you might do with computer scientists, majors, or whatever, as opposed to what you do with philosophy majors in an ethics classroom?
37:19
[SPEAKER_00]: Sure.
37:20
[SPEAKER_00]: Yeah.
37:20
[SPEAKER_00]: I conceptualize the end result of the lesson to be a whole heck of a lot different.
37:24
[SPEAKER_00]: I might get a small amount of time to intervene with them or really important when you're integrating ethics into these classrooms.
37:30
[SPEAKER_00]: I might actually be drawing something up that their own computer science professor is going to deliver to them.
37:36
[SPEAKER_00]: So I need to conceptualize the endpoint of the lesson as being a technical skill set that they're either going to know to
37:45
[SPEAKER_00]: or a repo or another place where they can actually go and find some of the tools that we talk about near the end of the lesson.
37:52
[SPEAKER_00]: So it's walking them through this and not simply coming to a conclusion of reason, but again, getting all the way to that point of behavioral change, walking to the point where they're saying, what I would do in this situation is I would run this Eval on this machine learning model on this LLM or something like that, or what I would do in this case is that actually think about different data sets, right?
38:14
[SPEAKER_00]: So that
38:16
[SPEAKER_00]: In the ethics classroom, or sometimes I guess maybe sociology classrooms, we might teach something like gender shades, and we actually end at a conversation that is about the ethical theories or we actually end the conversation that's about the effect on identifying our book groups out there in the society.
38:31
[SPEAKER_00]: I like to point out that Joey Bullonwini and Tim
38:37
[SPEAKER_00]: And people have disagreed with them about whether that is actually the solution to the problem that they identified in search under shades and that's fair to do as well, right, but they took it in a technical direction as well as a sociological and ethical and another direction.
38:50
[SPEAKER_01]: All right, so we've covered a lot of a ground about the kinds of things that you want to do with the things that you're aiming for.
38:55
[SPEAKER_01]: You said, what the outcome is that you're driving at, how do you think about measure success?
39:01
[SPEAKER_01]: How do you know?
39:01
[SPEAKER_01]: How do you know that things are going well, that this kind of ethics education is successful?
39:07
[SPEAKER_01]: Is it anecdotally that you can see that they're getting better at ethical reasoning and making the right calls and complex decisions or what?
39:13
[SPEAKER_01]: How do you measure all this?
39:15
[SPEAKER_00]: Yeah, that's right.
39:16
[SPEAKER_00]: That's the hardest problem of all I think and ultimately the proof would be in the putting right if we end up with technologies better designed in the future, but it's awfully hard to know what contribution college class is making to that.
39:28
[SPEAKER_00]: So a lot of ton of people are working right now on this measurement question there's a lot of good approaches to it.
39:33
[SPEAKER_00]: But before we even started to design a program to do these types of interventions, we wanted to design a target, right?
39:40
[SPEAKER_00]: We used cognitive psychology to conceptualize it.
39:43
[SPEAKER_00]: And we said, look, if we do these various different interventions, there's role plays, there's horror stories, there's talk about the tech, there's everything that we've talked about in this conversation.
39:54
[SPEAKER_00]: If we do those things, then we should see, for instance, that they'd be able to look at an unlabeled tech scenario, just a description of a tech and a description of a company, and they would be more cautious about it.
40:05
[SPEAKER_00]: They'd be more likely to seek expert opinion.
40:07
[SPEAKER_00]: They'd be more likely to seek a diverse array of perspectives.
40:11
[SPEAKER_00]: And in the measurements that we've done, we showed that students who went through these interventions, it's hard to pinpoint which of them did the work.
40:18
[SPEAKER_00]: But students that went through these interventions were, in fact, better able to recognize problems in unlabeled scenarios than the average student at our university was.
40:27
[SPEAKER_00]: Then the average computer science student, as well as the next question we're answering and beginning to be able to answer.
40:33
[SPEAKER_00]: So having gone through the interventions better able to recognize, having gone through the interventions, better able to say that they would invoke reasoning as opposed to intuition, et cetera, at the relevant moments.
40:46
[SPEAKER_00]: Having gone through these scenarios, having gone through these interventions, being able to say, we're aware of system complexity and we're aware that we'd have to martial our resources and find some tools to be able to actually affect this outcome.
40:59
[SPEAKER_00]: So the first thing we did really was design the study that we were going to do to figure out whether or not these interventions were effective or was whole design them together.
41:08
[SPEAKER_00]: And then we've recently published our first study on that and we continue to actually look at our students to make sure that we're having some effect.
41:17
[SPEAKER_00]: There are other schools out there that are doing other types of measurements and there is might be better.
41:21
[SPEAKER_00]: I want to admit.
41:22
[SPEAKER_00]: I maybe think that's not something you're supposed to say.
41:25
[SPEAKER_00]: What I'm going to say.
41:25
[SPEAKER_01]: Well, it might be, but there's a bit of evidence that they're at that they are better.
41:30
[SPEAKER_00]: So it's an empirical question and we're asking it.
41:32
[SPEAKER_00]: Yeah, I am respecting people who are asking it in their own way elsewhere.
41:36
[SPEAKER_01]: So is it a kind of test that you give a much people, so you give a kind of test, hey, here's the situation, spot them or really problematic things, say what you would do or whatever, and you give that kind of questionnaire or test to lots of different people, so we've undergone the training, someone who have not, and then you sort of see how it compare.
41:54
[SPEAKER_00]: Correct.
41:55
[SPEAKER_00]: And it's got three separable parts to it.
41:57
[SPEAKER_00]: There are moral awareness questions.
41:58
[SPEAKER_00]: There are reasoning versus intuition questions, and there are system complexity questions.
42:03
[SPEAKER_00]: So you could disaggregate it and do a really quick test of someone who's gone down in intervention.
42:08
[SPEAKER_00]: But that's it.
42:08
[SPEAKER_00]: That's our ideal.
42:09
[SPEAKER_00]: Our first study to come out is really a cohort analysis just looks at a large group of students and compares them to people who haven't had any of these interventions.
42:16
[SPEAKER_00]: But the AB testing is what we're working on now.
42:19
[SPEAKER_00]: Three posts is the better way to say that's what we're working on now.
42:23
[SPEAKER_00]: Same student before they went through the intervention after.
42:25
[SPEAKER_00]: Yeah, quick test.
42:27
[SPEAKER_00]: And maybe all three of those tests are maybe just one of them depending.
42:30
[SPEAKER_01]: Yeah, let me ask something of a curveball question.
42:35
[SPEAKER_01]: Are we sure that a trained student who's like the training is fairly limited right it's not like you've got them they're not majors and ethics or philosophy or something on those lines so you've got them for a semester or something on those lines I wonder if they could be trained up to spot ethical issues as well as an LLM can do it.
42:58
[SPEAKER_01]: So suppose you take those scenarios, give it to an LM, say, tell me what the ethical red flags here are and tell me what should, you know, if you're a junior data scientist, what you should do about it, et cetera, et cetera.
43:08
[SPEAKER_01]: I wonder if the LLM would perform, how the LLM would perform relative to that train student or to an untrained student.
43:15
[SPEAKER_01]: Have you ever thought about testing
43:18
[SPEAKER_00]: Yeah, I have, actually, and sometimes when the students aren't in your shop, so I hope they've turned off this podcast, I'll refer to them as small language models, but here's where I'll actually disagree and it's the big issue.
43:36
[SPEAKER_00]: But here's what I'll say about it, which might fork that question a little bit.
43:39
[SPEAKER_00]: If the thing we're really concerned about is the being able to take all these skills and recognize the new problem of the emergent problem in a technology, then it's actually the skill of bringing together everything that they know about social life, about others, about ethical conversation, everything they know, and finding that new thing.
43:57
[SPEAKER_00]: and I think that's where actually the training data of large language models tends to equip them to say a lot about what has happened in the past and maybe they're a little bit less good at spotting the wholely new.
44:07
[SPEAKER_00]: So that's really what we're hoping to train our students in.
44:13
[SPEAKER_01]: So is the answer something like
44:15
[SPEAKER_01]: It's just as good.
44:17
[SPEAKER_01]: Ellems are just as good as trained students if not better than trained students for existing technologies, but less good when it comes to speculating about emerging technologies.
44:25
[SPEAKER_01]: Let's say quantum and AI, is that the... Clean?
44:29
[SPEAKER_00]: I'm gonna go out on a limb and say low.
44:30
[SPEAKER_01]: Or the hope?
44:31
[SPEAKER_00]: I'm gonna go out on a limb and say, less good at all of it.
44:34
[SPEAKER_00]: Don't at me, my computer science is the phrase.
44:37
[SPEAKER_00]: But I guess I less good at all of it.
44:38
[SPEAKER_00]: But if you particularly take later on and what I've described as the process of decision-making,
44:45
[SPEAKER_00]: what to do if you're experiencing emotions that you think are interfering with your judgment.
44:50
[SPEAKER_00]: If you're experiencing pressures to make money better, etc, etc, they can give you all well-reasoned response on that.
44:55
[SPEAKER_00]: But I have yet to see anybody inject the actual emotion into them and see what they do.
45:00
[SPEAKER_00]: There's parts of this guy there that are really uniquely human, I think.
45:04
[SPEAKER_01]: I guess I'm not talking about what they would do, but as advisors or suppose we have smoke detectors, can we use LLMs as ethical smoke detectors, which might be, and then would they be better, worse, or equal to smoke detectors that are people who undergo some training.
45:24
[SPEAKER_01]: And it just strikes me as an empirical question.
45:25
[SPEAKER_01]: And then there's a separate question.
45:27
[SPEAKER_01]: Let's suppose for the sake of argument that it turns out that they are better, smoke detector, ethical smoke detectors, then.
45:33
[SPEAKER_01]: I don't know, the average computer scientist, the best computer scientist, the best computer scientist who's undergoing the best training.
45:39
[SPEAKER_01]: What do we have to think about what the benchmark is here?
45:41
[SPEAKER_01]: But put that into the side.
45:43
[SPEAKER_01]: There's this further question about, yeah, it's better what we still shouldn't use them for other kinds of reasons, either ethical reasons, maybe they're pragmatic reasons, I don't know.
45:49
[SPEAKER_01]: But it feels like an experiment is, it's worth doing the experiment.
45:53
[SPEAKER_01]: And also worth engaging in that inquiry about if LLM's turned out to be better ethical smoke detectors, then you're 28-year-old data scientist, what should we do about that fact?
46:04
[SPEAKER_01]: What does it tell about how we ought to behave?
46:06
[SPEAKER_00]: This awesome.
46:07
[SPEAKER_00]: All your questions are awesome.
46:08
[SPEAKER_00]: And I know you, the rule is, don't praise the host of the podcast, but one of the reasons I really wanted to come on is the challenging questions.
46:14
[SPEAKER_00]: And you literally have just caused me to think that maybe I need to do this, because I'm going to admit something.
46:21
[SPEAKER_00]: I talked to Jackie Pt about these things, right?
46:24
[SPEAKER_00]: So I've got to start in a lesson about a hamming code.
46:26
[SPEAKER_00]: Do you know what a hamming code is?
46:28
[SPEAKER_00]: I don't know either do I, right?
46:29
[SPEAKER_00]: So I'm like, Whitman, I've got to teach them how there could be some kind of ethical danger in this specific thing that's occurring in week through this course.
46:37
[SPEAKER_00]: Good.
46:37
[SPEAKER_00]: God.
46:38
[SPEAKER_00]: Who knows about him and codes?
46:40
[SPEAKER_00]: Chat, GPT.
46:42
[SPEAKER_00]: Yeah, right.
46:42
[SPEAKER_00]: So I'll take an hour and talk to it.
46:44
[SPEAKER_00]: OK, what are the use cases for that?
46:46
[SPEAKER_00]: Could you give me a specific website that talks about that?
46:49
[SPEAKER_00]: Yeah, use totally.
46:49
[SPEAKER_00]: How could that go?
46:50
[SPEAKER_00]: Or, right, win a minute, hold on.
46:52
[SPEAKER_00]: Passion, it's a different technology here.
46:54
[SPEAKER_00]: But passion is used with election data as states pass it back and forth.
46:58
[SPEAKER_00]: Oh, now I'm in my wheelhouse, right?
47:00
[SPEAKER_00]: Yeah, yeah, yeah.
47:01
[SPEAKER_00]: Right.
47:01
[SPEAKER_00]: You know, Paul is, of course, so now I can come up with scenarios.
47:04
[SPEAKER_00]: And I can be like, all right.
47:05
[SPEAKER_00]: you're learning about hashing.
47:06
[SPEAKER_00]: Now I've got a problem for you.
47:08
[SPEAKER_00]: What would you do about it?
47:09
[SPEAKER_00]: And the thing you're really making me think is should I be less shy about telling students that's how I've learned about some of their concerns?
47:17
[SPEAKER_00]: And should I encourage them and say, if you know the questions to ask and you doubt the thing constantly, you could, yeah, pretty good insights.
47:25
[SPEAKER_01]: Yeah, I mean, I think that large language models actually do the most benefit for experts,
47:31
[SPEAKER_01]: So if you're using LM in a public policy, political philosophy context, it's going to be very helpful to you because you could immediately spot the bullshit.
47:41
[SPEAKER_01]: You could immediately spot false things.
47:42
[SPEAKER_01]: You could immediately spot red herrings.
47:45
[SPEAKER_01]: And say, no, that's that.
47:46
[SPEAKER_01]: Because this, you could, I think it's a better tool for you.
47:48
[SPEAKER_01]: That's what I found when I'm using it within my domain of expertise.
47:52
[SPEAKER_01]: It's super useful.
47:54
[SPEAKER_01]: When it's outside my domain of expertise, I don't know if that's stuff about how to file taxes in a certain way or should I should my company be an LLC or should I files an S corp, I don't really know how to review or validate or invalidate the outputs.
48:09
[SPEAKER_01]: And so I think it's better, I think it on the whole better for experts, but nonetheless, suppose you're not an expert and your choices, I could rely on
48:19
[SPEAKER_01]: judgment, reasoning, et cetera, or I could rely on an LLM's, and in some cases, you might think, yeah, the LLM is going to be better.
48:26
[SPEAKER_01]: And so I'm just trying to think about, do we want to make better people, or do we want to make tech better, and might those two things come apart?
48:35
[SPEAKER_01]: Might it be the case that there's a certain kinds of ethical education that we can give to people that will make them better at ethical reasoning, better at smelling ethical smoke, and they are by
48:49
[SPEAKER_01]: And then they have an impact on the tech and it's better than having no trading at all, fine.
48:53
[SPEAKER_01]: But then there's this other situation where maybe they take their use case, they give it to an LLM and say, tell me what the ethical red flags are, prioritize them.
49:03
[SPEAKER_01]: So there are times probability, tell me what the risk management strategies are, tell me how I should talk to my boss about this.
49:09
[SPEAKER_01]: And they thereby don't gain.
49:11
[SPEAKER_01]: the sort of ethical capacity, just like people are worried about cognitive offloading and people getting stupid or the LM's.
49:17
[SPEAKER_01]: This is a kind of ethical or moral offloading.
49:19
[SPEAKER_01]: And it might make people, not as ethically go to say, could be if they underwent the trading blah blah, but it might be, this is an empirical claim, at least somewhat empirical claim, better
49:31
[SPEAKER_01]: more conducive to creating ethically better technology because the albums are better at doing that stuff that even a trained computer data scientist is.
49:38
[SPEAKER_00]: Yeah, I think he would want to scientists.
49:40
[SPEAKER_00]: I think you'd want to give the LLM an identity.
49:43
[SPEAKER_00]: So there's a lesser read out there read Hoffman who created AI box of himself.
49:48
[SPEAKER_00]: And maybe it would be useful for you to have a read blog not bought out there.
49:51
[SPEAKER_00]: And there's a Peter Singer bought I think already.
49:53
[SPEAKER_00]: I know there's a Luchia.
49:55
[SPEAKER_00]: But maybe to give it identities and for the untrained, un-expert person to say, hey, what we actually need you to do is partway through your conversation, switch it over and say, hey, go into the identity and the data set that's associated with the red blackman, but and not tell me how everything that I just said was wrong.
50:15
[SPEAKER_00]: Yeah.
50:16
[SPEAKER_01]: Yeah.
50:17
[SPEAKER_01]: Or you have a panel of experts.
50:18
[SPEAKER_00]: Yeah.
50:19
[SPEAKER_01]: Or you have a panel of different experts who weigh in and tell you whatever.
50:22
[SPEAKER_01]: This is going to spot this as a run flag.
50:23
[SPEAKER_01]: This person won't think it's run flag because they have a different ethical political orientation blah, blah, blah, blah, blah, but as a trained philosopher as someone who's devoted most of his adult career to ethics or all his adult career to ethics.
50:35
[SPEAKER_01]: Of course, I'm inclined to, you know, do the work, do the God they work, do the ethical reasoning, this should be human's leading the way we should be thinking about it for ourselves blah blah blah.
50:46
[SPEAKER_01]: And then there's a sort of, for if our goals to really, if we're doing the ethical education, primarily as a mean to the end of stopping the horrific stories from happening, then maybe that's not the best way.
50:56
[SPEAKER_00]: Yeah.
50:57
[SPEAKER_00]: For a few uncomfortable years, I had to teach academic research and writing as that's a topic for another day, why that's so hard to do.
51:04
[SPEAKER_00]: But I actually came to think of teaching the method as teaching the doing of the work, right?
51:10
[SPEAKER_00]: So I'd bring them a little bit further together, a little bit closer to each other.
51:14
[SPEAKER_00]: And so actually if we taught a method of inquiry with the bots,
51:18
[SPEAKER_00]: We could get people to do that in a way that's not just simply offloading the cognitive processes, but in fact, is enhancing their ability to do this.
51:26
[SPEAKER_00]: And yeah, they wouldn't have thought through for themselves how would Luciano Flority disagree with that.
51:32
[SPEAKER_00]: But they know the method of the methodology, as academics talk to say, of producing those results.
51:38
[SPEAKER_00]: And that kind of is the content at times.
51:41
[SPEAKER_01]: Yeah.
51:44
[SPEAKER_01]: We have at least four systems to test, right?
51:45
[SPEAKER_01]: We've got a computer scientist or something like that by themselves.
51:50
[SPEAKER_01]: Then we have them with an LLM.
51:52
[SPEAKER_01]: And then we have a trained, a ethically trained computer scientist.
51:58
[SPEAKER_01]: And then we have a computer scientist with an LLM.
52:02
[SPEAKER_01]: I'm just not sure which one of those four would perform the best, and then I'm just not sure whether whichever one performs the best is the one that we should go with.
52:09
[SPEAKER_01]: Or there's other kinds of considerations that play like we really shouldn't be enhancing the person blah blah blah.
52:15
[SPEAKER_00]: My instinct says ethically trained computer scientists with a bot, but then, of course, they're in a room full of other people who have computer science, but also other training, just training UX training, etc, etc.
52:28
[SPEAKER_00]: And they're part of a corporate structure that is, as it puts its own pressures on them.
52:34
[SPEAKER_01]: So then you have to think about we're talking about what performs the best in an ideal situation, but then there's costs.
52:38
[SPEAKER_01]: So yeah, okay, maybe the trained computer data scientist with the LLM does better than the untrained computer data scientist with the LLM, but whatever that's going to cost a lot of money.
52:48
[SPEAKER_01]: So let's just do the untrained one because it's good enough.
52:50
[SPEAKER_01]: It meets our risk threshold.
52:52
[SPEAKER_00]: Yeah, yes, earlier is what are we actually aiming for?
52:54
[SPEAKER_00]: And I'd say we're aiming at tech with better social effects, right?
52:58
[SPEAKER_00]: But the only way we can actually do that from time zero, moving forward to time T plus one is create more ethically sensitive computing students.
53:07
[SPEAKER_00]: And then also talk about corporate structures.
53:10
[SPEAKER_00]: Some of our written is about using the theory of the firm to analyze tech companies, right?
53:15
[SPEAKER_00]: And yeah, the stuff I do is hopeful about their potential to innovate ethically.
53:20
[SPEAKER_00]: taking into account all the stuff you're always reminding people about as well about this comes with costs and sometimes maybe doing the last bad.
53:26
[SPEAKER_00]: Yeah, this is an approach to consider as well.
53:28
[SPEAKER_01]: Yeah, I'll just say one last thing, maybe this is two repetitions, but your goal is to create ethically better tech, but I can see someone, like a virtue at the system of some sort, pushing back and saying, No, if you do that, then you're turning the ethical reasonings of people into a mirror means to the wind.
53:44
[SPEAKER_01]: It's a mirror strategy to get to your goal, but that's the wrong way to conceive of it.
53:48
[SPEAKER_01]: You want to think of your goal as
53:59
[SPEAKER_01]: I would, I'm not saying that you should do that.
54:00
[SPEAKER_01]: I'm just saying like that strikes me as a plausible view.
54:04
[SPEAKER_00]: Yeah, I would say that actually that probably is Shannon Valer's view.
54:07
[SPEAKER_00]: I definitely can't speak for her.
54:09
[SPEAKER_00]: Yeah, I think that's right.
54:10
[SPEAKER_00]: Yeah, yeah.
54:10
[SPEAKER_00]: I think that's her view and that's what makes her different than say Julia and us in the virtue of ethics community virtue isn't just a skill.
54:17
[SPEAKER_00]: And Shannon's Valer's telling of it, there is actually an entire world.
54:21
[SPEAKER_00]: There's multiple cultures.
54:22
[SPEAKER_00]: There's these effects and so on and so forth.
54:24
[SPEAKER_00]: Yeah, so we're trying to create that person.
54:26
[SPEAKER_00]: who's going to ask the right questions to affect that specific community out there or those many global communities, etc, etc.
54:34
[SPEAKER_00]: with these specific technologies, these many technologies etc.
54:39
[SPEAKER_01]: I was even, thank you so much.
54:40
[SPEAKER_01]: This has been really great stuff and it makes me really think hard about what it is to be a good teacher in this kind of context.
54:46
[SPEAKER_01]: It's great stuff.
54:46
[SPEAKER_00]: Great to talk to you, Erie.
54:49
[SPEAKER_01]: Thanks everyone for listening.
54:51
[SPEAKER_01]: I hope you enjoyed the episode.
54:52
[SPEAKER_01]: And I want to thank a few people who make this all possible.
54:55
[SPEAKER_01]: Delano Alexander, he's my producer.
54:57
[SPEAKER_01]: He does such an amazing job on everything that he touches.
54:59
[SPEAKER_01]: Garrett Lane is the sound engineer.
55:01
[SPEAKER_01]: That's honestly, he's just, he's too good for me.
55:04
[SPEAKER_01]: Authamia Jairam has joined us regionally.
55:05
[SPEAKER_01]: He's helping us find the best guests to bring on to this podcast.
55:09
[SPEAKER_01]: And Ashley Wasselenko helps us organize all this.
55:11
[SPEAKER_01]: Get people on the schedule.
55:13
[SPEAKER_01]: Get people's bios, headshots, and really keeps the chance to trade on the track.
55:16
[SPEAKER_01]: So thanks guys, really appreciate all the hard work you do.
55:29
[UNKNOWN]: Thank you.