525: Some Sort of Metal
Transcript from 525: Some Sort of Metal with Tom Williams, Christopher White, and Elecia White.
EW (00:00:07):
Welcome to Embedded. I am Elecia White, alongside Christopher White. Our guest this week is Dr. Tom Williams. We are going to talk about robotics, but maybe not in the way you think or I think. I am not sure what to expect, beyond robotics.
CW (00:00:24):
Hi, Tom. Thanks for joining us today.
TW (00:00:26):
Thank you for having me.
EW (00:00:28):
Could you tell us about yourself, as if you were visiting the Harvey Mudd College Alumni Reunion?
CW (00:00:36):
Why? <laugh>
TW (00:00:37):
Yes, absolutely. Yeah. Dear Alumni of Harvey Mudd, thank you so much for inviting me out to California. I am Dr. Tom Williams. I am a social roboticist at the Colorado School of Mines. That means I study how people interact with social robots, and how we can design robots to interact more effectively and ethically with humans. My background is I am a cognitive scientist, by training.
EW (00:01:06):
But you teach computer science?
TW (00:01:08):
I do. I do. My PhD is in computer science and cognitive science. So it is sort of the intersection of computer science, psychology, philosophy, linguistics. So I like to say this gives me a license to dabble.
(00:01:21):
I do teach in the computer science department. I teach computer science and robotic students. But the courses I teach are at the intersection of those areas and psychology, design, philosophy and sociology.
EW (00:01:37):
I did some education in computer science and cognitive psychology. You said, "Cognitive science." Are those the same?
TW (00:01:47):
Yeah. Cognitive science is sort of the blend or the intersection of those different areas. If you go to a cognitive science conference, it will be a lot of cognitive psychologists using computational methods. And a lot of computer scientists studying or taking insights from cognitive psychology, to inspire their algorithms.
EW (00:02:16):
Cool. Okay. Well, I have many more questions, but first we have lightning round, where we will ask you short questions and we want short answers. If we are behaving ourselves, we will not ask, "Why? How? Huh."
TW (00:02:27):
Brilliant.
EW (00:02:28):
I think we are going to start with a short game of robot or not. Are you ready?
TW (00:02:32):
Yes.
CW (00:02:33):
Is a microwave a robot?
TW (00:02:35):
No.
EW (00:02:36):
Self-driving car?
TW (00:02:37):
Yes.
CW (00:02:38):
A Roomba?
TW (00:02:39):
Yes.
EW (00:02:40):
Data from "Star Trek"?
TW (00:02:41):
Yes.
CW (00:02:42):
Those little doggies from like the '80s? They would be outside mall toy stores. They would run into something, and then turn around and then keep walking, and then run into something and turn around.
TW (00:02:51):
Not old enough.
CW (00:02:51):
<laugh>
EW (00:02:54):
Coffee maker?
TW (00:02:55):
No.
EW (00:02:56):
Automatic coffee maker?
TW (00:02:58):
No.
CW (00:02:58):
The little drinking birds that do the thermodynamic thing? They go into the water and then they heat up and then they go back out of the water, and then...
TW (00:03:06):
No.
EW (00:03:07):
He gave you the same look I did when you suggested that.
CW (00:03:10):
<laugh> You cannot tell.
EW (00:03:12):
How many nerve center mugs does it take to trap a Wayno?
TW (00:03:18):
Six.
CW (00:03:19):
A Waymo.
EW (00:03:21):
It says "Wayno" here.
CW (00:03:22):
<laugh>
TW (00:03:23):
Oh. Oh, I thought you meant the Wayno off-brand Waymo car. If it is a genuine Waymo, then it is going to be eight.
CW (00:03:30):
There is an off- I will ask later. Complete one project or start a dozen?
TW (00:03:34):
Start a dozen.
EW (00:03:36):
Would you rather read "Sex, Race and Robots: How to Be Human in the Age of AI" by Ayanna Howard, or "Embodied AI Safety: Reimagining Safety Engineering for Artificial Intelligence in Physical Systems" by Philip Koopman?
TW (00:03:51):
The first. Which is to say I have read the first, and I have not read the second one. I heartily recommend the first, Ayanna Howard's book.
EW (00:03:58):
It was good.
CW (00:03:59):
Okay. All about these robots. What is your favorite fictional robot?
TW (00:04:03):
Fresh Cut Grass from "Critical Role."
CW (00:04:07):
Okay. I have been meaning to start "Critical." I know very little about it, but can you give us a little teaser?
TW (00:04:12):
Yeah. Fresh Cut Grass is a character from season three of "Critical Role," which is a Dungeons & Dragons actual play. He is a sort of therapist robot with a dark past, that he is discovering and working to overcome. It is not my favorite season of "Critical Role," but it is my favorite robot character.
EW (00:04:35):
Would you rather read "A Psalm for the Wild-Built" by Becky Chalmers, or "Platform Decay," the latest of the "Murderbot Diaries," by Martha Wells?
TW (00:04:45):
Hmm. Well, I just read "Psalm for the Wild-Built" a couple days ago, and I have not read "Platform Decay."
EW (00:04:54):
Have you read the "Murderbot" books?
TW (00:04:56):
I have read the first two. This is a hard one for me, because I was not a fan of either of these.
EW (00:05:03):
Ooh! Okay.
TW (00:05:03):
I enjoyed "Psalm for the Wild-Built," but it left me lacking a little. Actually for both of these, I enjoyed them, but I wanted more.
EW (00:05:14):
Okay. Tell me what you wanted more.
CW (00:05:15):
<laugh>
EW (00:05:15):
Sorry, I cannot do it. I know I am supposed to ask about all the other things.
CW (00:05:19):
Derailed the podcast talking about books.
TW (00:05:22):
I know. Yeah. This is lightning. Lightning. No, no. For a conversation about books, this is a lightning round for me. Yeah. I think that with the "Psalm for the Wild-Built," it had a very cozy feeling to it. It was delightful. But it felt like there was not as much depth to it as I was hoping for.
(00:05:44):
In contrast, I am currently reading- Immediately following reading her two books, I started reading "Death of the Author," from... Let me pull it up. From Nnedi Okorafor.
EW (00:06:07):
Oh yeah. Okay.
TW (00:06:09):
Which is about an author who writes a book very, very similar to the "Monk and Robot" books, about these robots in this sort of post-apocalyptic society. Because it is alternating between the science fiction about the robot, and then the much more human and more modern story about the author, I think there is a lot more emotional depth to the book.
(00:06:44):
So I think that the "Psalm for the Wild-Built" did what it accomplished, but I was hoping for a little bit more literary depth to it.
EW (00:06:52):
That is fair. The Nnedi Okorafor books are grittier, and "Psalm for the Wild-Built" was-
TW (00:07:01):
Is not.
EW (00:07:02):
Fairytale-esque.
TW (00:07:03):
Yes.
EW (00:07:04):
I actually was listening to it to go to sleep for a little while. It was not something I was worried about waking up in the middle of a murder scene or something.
TW (00:07:13):
A hundred percent.
EW (00:07:15):
Okay. Since I have derailed this, I do have a question about robot-themed musical theater improv?
TW (00:07:24):
Ooh, yes.
EW (00:07:26):
I do not know what my question is, so let us just go with, "Whah?"
TW (00:07:31):
Okay. Yeah. You said, "Robot-themed musical improv," and I think you are connecting together a couple of the different things I do.
(00:07:40):
Outside of my work as a computer scientist, I spend ten to 15 hours a week doing improv theater. This includes performing improvised musicals, where we get a suggestion from the audience of a choice they have made, and then perform an improvised musical about their life if they had made a different choice.
(00:08:03):
And then separately, I have been doing some robot-themed comedy nights. This includes "Robot Riot," which is a monthly robot comedy show, where we have comedy about robots and AI, and also involving robots and AI.
(00:08:19):
So for example, we will have improvisers performing on stage, while a robot interjects and gives them rules they have to follow, to showcase their talents. But also to highlight for the audience the ways that the rules we impose on robots, end up being rules that we are imposing on human autonomy as well.
(00:08:40):
And then I also am running an improvised "Black Mirror" show, where we get tech ethics students from Mines and from CU Boulder to write pitches for "Black Mirror" episodes, and then we enact them on stage.
EW (00:08:56):
Okay. I am going to go back to what the show is supposed to be about, because otherwise I am going to get totally lost.
CW (00:09:01):
<laugh>
EW (00:09:01):
We can come back to that later. I mentioned the Mudd Alumni thing because we recently went to our reunion. Every other person I talked to had something to say about the Colorado School of Mines. All I really knew about it was that there were explosions, which- Very attractive.
CW (00:09:20):
Which was the reason I almost applied there.
EW (00:09:22):
Right. Can you tell us a little about the Colorado School of Mines?
TW (00:09:28):
Yeah. The Colorado School of Mines is an engineering college that is just outside Denver, Colorado. It has been around for about 150 years. I think 151 this year. It started in the 1800s as a mining engineering school, to educate the people who were coming to Colorado as part of... I guess I do not know enough about Colorado history, but not the gold rush. I do not know. Whyever people were coming to Colorado to make their fortune in the mining industry in the 1800s.
CW (00:10:09):
Some sort of metal.
TW (00:10:10):
Some sort of metal. Yeah. So Mines as a university, it does have a very strong mining engineering department. It owns a mine out about 45 minutes from the school. It is an old silver mine. I have been down there. It is very cool. People run all sorts of experiments there. It is very cool. They have got a classroom that is hollowed out of the rock, that they hold some classes in. Yeah, it is wild.
(00:10:39):
But the university today, that is a small part of what we actually do. It is in general just a STEM university. So most of the departments are focused on engineering or applied science, with the exception of the economics and business department, which is still oriented around those types of science and engineering programs.
CW (00:11:10):
How big is it?
TW (00:11:12):
It is about 7,000 students.
CW (00:11:15):
Oh, wow.
TW (00:11:16):
So it is small-ish, but not tiny. It is about the size of if you took the engineering school from a larger university and airlifted it out and dropped it somewhere else. It is not huge, but that is because we do not have all of those other majors. We have within computer science and mechanical engineering, I think something like 70 faculty.
(00:11:51):
It has a pretty robust program in robotics. We have got robotics masters and PhD programs, for example, which are not common in the US. But we have no one in- We have no English majors, we have no philosophy majors. We have no psychology majors, which is interesting as a cognitive scientist.
CW (00:12:10):
But- Okay, so it is a graduate school as well. You award PhDs.
TW (00:12:15):
Yep.
CW (00:12:15):
Okay.
EW (00:12:16):
When I think about robotics, it is about the complex electrical and mechanical systems.
CW (00:12:26):
Because you are an implementer.
EW (00:12:27):
Because I am an engineer, and I develop these things. If you show me a BLE goldfish monitor, I am going to say, "No, that is not a robot."
(00:12:36):
But having worked on an autonomous water sampling system, with motors and sensors and enough intelligence to let it live in the wild for months on end, I am like, "Yeah. That is pretty much a robot, even though it does not walk around and say anything to people." Would that be a robot to you?
TW (00:12:55):
I think I would need to see it, but it is complicated-
EW (00:13:02):
So complicated.
TW (00:13:02):
Because there is not a clear definition of a robot.There are engineering definitions that people have created. For example, saying that a robot is an embodied artifact that can sense its environment, make decisions, and take actions in the world. But that definition does apply to a lot of the things you talked about in the lightning round. It applies to-
EW (00:13:25):
The coffee maker.
TW (00:13:26):
The coffee maker, to thermostats, to elevators, to dishwashers. Part of the reason why this is complicated is because robots are a science fiction concept. The word "robot" comes from science fiction. The word "robotics" comes from science fiction.
(00:13:41):
So to a large extent, the definition of a "robot," in terms of the most accurate and I think compelling definition of a "robot," is just if you saw it in a movie, would your parents describe it as a robot?
EW (00:13:58):
<laugh>
TW (00:13:58):
And then this is slightly complicated because things like autonomous cars now do not fit that mold of how we typically think about robots. But they are sort of on the border and people do, I think, understand them or think of them as robots. Or more as robots than they would their dishwasher, because they are autonomously moving or semi-autonomously moving through the environment.
(00:14:22):
But then there are all sorts of other things that people in robotics work on, or they use the tools and techniques from robotics to work on, that the general public would look at and be like, "Well, that is not a robot."
(00:14:34):
So to the roboticist to sort of claim that what they are doing is robotics, is it a robot? Yes. But in the larger cultural sense, does our society consider it a robot? No. There was a great paper that came out at the Human-Robot Interaction Conference two months ago, from Waki Kamino from Cornell University, on the robot umpires in baseball.
CW (00:15:03):
Right, right. They have started using automation to detect balls and strikes, right?
TW (00:15:08):
Exactly. Exactly. There are these automated ball and strike cameras. I think she writes in her paper that the Major League Baseball really tried to emphasize that it is not a robot. But the public very much picked up on this robot framing for it, and it has very consistently been referring to it as a robot.
(00:15:32):
She points out in her paper the ways that the definition of "robot" is really fluid. It depends on what it is doing, and who it is made by, and where it is deployed, but in ways that are really hard to pin down. So it is a fluid technology that does not really have a very firm definition, where you can clearly and easily draw a line in the sand between robots and not robots.
EW (00:16:04):
As a professor, if a student comes to you, sophomore junior, says, "I want to be a roboticist. I want to work on robots," what classes do you recommend they take?
TW (00:16:17):
There it does sort of break down into the classic robotics definition. So if we think about robots again, as things that perceive the world, make decisions, take actions, then the students need some type of classes on perception, like computer vision or mapping.
(00:16:39):
They need some type of classes on cognition, whether that is machine learning or AI or other types of more specific planning methodologies. They need some training on the action side of how to actually take actions in the world. That tends to be the more mechanical engineering classes, like mechatronics or robot control.
(00:17:07):
And then from my perspective, students also need to understand something about robots and society, and robots and people. It is not helpful to be able to know about the mechanical and software construction of the robot, if you do not know anything about how to talk to people and figure out what their needs are.
(00:17:31):
If you have no idea about what the implications of that robot are going to be in society once it is deployed. Because of this, I have gotten it so that all of our robotic students at Mines have to take either my human-robot interaction class, which is sort of the intersection between robots and psychology and design. Or they need to take my robot ethics class, which is the intersection between robotics and philosophy and sociology and history.
EW (00:18:01):
You wrote a book. Was it for one of these classes?
TW (00:18:04):
The book, "Degrees of Freedom," is not for one of these classes, but it is tightly interconnected with the classes. The book is very much on the history, philosophy and sociology of robotics. So in that sense, it is very tightly connected with my robotics class.
(00:18:23):
But it also makes recommendations for how we can move forward as a field. What types of methods we might take for designing robots in ways that are more responsible. And in that way it connects with the human-robot interaction class, because the focus is not just on what are the problems, but how can we use concrete design processes to do better?
EW (00:18:50):
I remember- I think it was in your introduction- When we talked to Ayanna Howard on the show, she mentioned having this study where she had a robot. The robot was supposed to lead people around. The robot in a quasi fake emergency situation led people in an entirely incorrect place. People still followed it!
(00:19:13):
Even after the robot had done things that indicated it was in a bad state. People trust robots. I wonder about- You just mentioned this umpire thing. I wonder if folks who watch baseball like the idea of a robotic umpire because it is super fair. Even though I can think of many ways to make it not fair <laugh>.
TW (00:19:39):
I think that is really interesting. With the study that Ayanna was talking about, it was not just that they had seen the robot performing poorly. It was that they had seen the robot performing poorly, and then the building begins to fill with smoke.
(00:19:54):
The robot just drives off and says, "Follow me." It is in this setting where it is like, "Well, there might actually be an emergency here." Every single participant follows the robot, instead of leaving the building and following the exit signs.
EW (00:20:09):
Instead of following the lit exit signs.
TW (00:20:12):
Yes! Yes!
EW (00:20:14):
That is wrong!
TW (00:20:15):
Yes. It is crazy, right? People have a tendency to dramatically overtrust robots. I think in the case with the baseball umpire, it is very interesting where one of the things that Waki talks about in her paper, is about the importance of theatricality. Where the purpose of the baseball umpire is not just to call the balls and strike successfully, but it needs to do so in a way that does not violate the sort of theatrical spirit of baseball.
(00:20:48):
It needs to do so in a way that allows the drama and tension of baseball to be maintained, to the extent that baseball has drama intention. I enjoy baseball, but it is a slow sport. But they pointed out that they have to make sure that the amount of time it takes to challenge a human umpire and get a call from the automated system, is not so long that it diffuses the tension.
(00:21:21):
And also they wanted it to be accurate. But to have a level of accuracy, where the challenges would go with equal likelihood towards the players or towards the umpires. So that it was not that every time a challenge was made, the automated system was just siding with the ump. Instead if it is like a fifty-fifty chance of siding with the ump, versus the player who made the challenge, then there is more drama because you really do not know how it is going to turn out.
(00:22:04):
That is interesting, right? Where it is not just about being accurate. It is even in that setting where it seems like it is really clean. It is like, "Well, is it a ball, or is it a strike?" The human factors, and the cultural and societal factors, are really, really important for quantifying whether or not you consider it to be successful or not.
EW (00:22:24):
It should just be a math problem. <laugh> It has got inputs. It is a bounding box. That is what a strike zone is. It should just be a number.
CW (00:22:36):
Right. Well, but if you want to do that, you should just put the game into a simulation. It is a game, right? It is a human event, so you are trying to put technology into something that is traditionally done by humans. You have to be very careful.
EW (00:22:46):
Yeah, I get it.
TW (00:22:48):
Well, it is also- I think this is a good point, where it is not just about the internal operations of the system, it is also about how it is deployed. Where it is not just how does the system make a decision? It is also what is the larger sort of choice architecture that leads to it being triggered?
(00:23:08):
How many challenges do you allow to players during the game? What is the procedure for making a challenge? Is this something that automatically happens, or you need to make a challenge? Those types of decisions about how the robot is deployed are going to also shape these questions around the drama and narrative of the game.
EW (00:23:29):
Why would you not do it automatically?
CW (00:23:30):
<laugh>
EW (00:23:30):
Okay, sorry.
CW (00:23:36):
There have been umpires for 150 years, 200 years. People are used to it. You cannot just get rid of them.
EW (00:23:43):
And it is kind of fun to yell at them.
TW (00:23:46):
Uh-huh. Well, it is also interesting in the paper- I know we are going on and on about this specific paper, but it was a really cool paper. But they also point out the ways that this is not completely new. They have a picture from, I do not know, in the 1940s or something, of an automated ball strike reader camera being used or prototyped at a game.
(00:24:12):
Obviously it did not catch on, because the tech was not there yet. But this is something that people have been trying to do in Major League Baseball for like 80 years.
EW (00:24:22):
Yeah. Let us go back to your book. It looks from the table of contents like either it has a lot of really great puns or it is about robotics. There is "control systems," there is "end effectors," there is "bounding boxes." But you do not have a chapter called "kinematics," or "the math you wish you remembered from school," or any of the chapters I need in my books these days.
TW (00:24:51):
Well, there is no chapter on the math you needed from school, because there is no math in the book. And the titles-
EW (00:24:56):
Which seemed like a win when I started it. But then it made me think too much.
CW (00:25:01):
Oh no! Not that.
EW (00:25:03):
<laugh>
TW (00:25:04):
<laugh> Oh no, I am sorry I made you think. Yeah. The names of the chapters are very much puns, that I have to explain in footnotes. So for "bounding box" for example, the chapter is talking about computer vision systems and robotics. When we talk about a bounding box, typically in computer vision we are talking about the sort of rectangle that is drawn over the image at the area where something is detected.
(00:25:34):
But in the chapter, I am talking about it in terms of the ways that algorithms are used to sort people into categories of race and gender. In ways that then go on to reinforce and legitimize those specific ways of reasoning about the world. So it is about computer vision, but the bounding box is about the sort of sociological category that people are being forced into.
EW (00:25:59):
What about "control system"?
TW (00:26:01):
The control system chapter is talking about the ways that robots' moral reasoning algorithms serve to control people. This goes back to what we were talking about with the comedy show, and some of the things I try to highlight in that particular act in the "Robot Riot" show.
(00:26:19):
We have this tendency in robotics to think about moral decision making by robots as a matter of compliance with ethical rules. I think this comes from two different places. One is it comes from Isaac Asimov, from his Laws of Robotics. The second is it comes from the dominant philosophical theory that ethicists have drawn on in robotics, which is deontology.
(00:26:48):
Both of these are kind of messed up if you get down to them. For Asimov, Asimov's Laws of Robotics, Asimov actually credits to his editor, John Campbell. Which if you know anything about the history of science fiction, they had to rename the John Campbell Award because John Campbell was an apologist for slavery.
(00:27:13):
This sort of helps to explain why Asimov, in his stories, his characters refer to his Laws of Robotics as giving the robots good healthy slave complexes. So they are really these slave codes for keeping these autonomous slaves in line.
(00:27:33):
And then deontology also has this similar worrying history, where the philosopher Charles Mills points out that at the same time that Immanuel Kant was creating his moral and political philosophies, he was also creating his race hierarchy and was one of the first race scientists.
(00:27:53):
So the folks for whom Kant is saying, "Well, we all agree that these are the rules that need to be imposed," are the people at the top of his hierarchy of race and gender. And the people on whom the rules tend to be imposed are everyone else.
(00:28:10):
This is not to say that we need to completely throw out rules, or throw out Kant completely. But I think it does suggest that maybe we should be aware of this history, and think about other types of perspectives we might use to approach what it means for a robot to behave ethically.
CW (00:28:27):
So glad I paid attention during "The Good Place."
TW (00:28:29):
<laugh>
EW (00:28:32):
You talk about roboticists enforcing white supremacy, capitalism, heteropatriarchy through their design choices. When you say "roboticists" there, do you mean the companies, the software engineers, mechanical engineers, industrial designers?
CW (00:28:49):
Sci-fi authors?
EW (00:28:50):
Product design? The funders? Who is responsible here?
TW (00:28:56):
All of the above, which is complicated. When we talk about this in my robot ethics class, we talk about how difficult this is. Because if you are talking about responsibility, everyone is responsible. It is not just that there is a single agent which is responsible, which makes it difficult then to assign accountability when things go wrong. But ultimately everyone is responsible in different ways.
(00:29:27):
In the book though, I am mostly talking to roboticists. It is a slightly generalist book, and the intended use is in tech ethics classes. So I am not expecting that people who are reading the book have robotics knowledge. But I am sort of talking to the next generation of robotics students, and robotics adjacent students.
(00:29:56):
So in that sense, I am talking to them as potential designers, potential engineers. But yes, of course the entire history of robotics has been shaped by and determined by choices made by science fiction authors. And of course it is the companies at a higher level that are steering robot design more fundamentally today, as opposed to the specific decisions of an individual engineer.
(00:30:27):
I think this is also somewhere where there is a- We have a collaborator and friend of the Lab, Katie Winkle, who is a roboticist currently in Sweden. Who has this awesome paper on feminist human-robot interaction, where she points out that the field of human-robot interaction is often pretty narrowly focused on the interaction itself.
(00:30:50):
Whereas we do need to consider this broader network of other people who are interacting with the robots, or are bystanders and otherwise affected by the robots. The places where the robots are deployed, the institutions, the funding agencies if those still exist. The researchers themselves, their participants. And all of the power relations between all these different groups of stakeholders, that go on to shape robot design.
EW (00:31:18):
It is hard. I spend a lot of my time working on control systems and trying to make the robot smart enough to do the things they need to do, in order to get whatever my goal is.
(00:31:30):
For like the water robot- You have examples where robots have a wasp shape, so that they are more feminized. It was a box. It was a clear box in spots, but mostly it was a box. To me it was very robotic, because there were so many moving parts and the electromechanical was so hard.
CW (00:31:54):
But that was not really something that interacted with humans. You put it on the bottom of a river and it interacted with fish.
EW (00:32:01):
And other creatures. Now I know about the ring-tailed thingies. Those were very cute. It interacted with people in a limited way. It did not interact with the public. It interacted only with scientists.
(00:32:14):
But even that, it was a lot of control systems. It was a lot of state machines. It was a lot of planning and thinking about how things are going to be used. But it was never... There was the ethics hurdle of, is this a good project? But as soon as it was like, "Okay, environmental DNA in the middle of wilds," I am like, "That sounds like fun." I did not think about this.
TW (00:32:41):
Yeah. I think it is complicated. Where on the one hand I would say that in the book, I am pretty focused as a social roboticist on interactive and humanoid robots. So robots that look like people and/or that are interacting with people.
(00:32:58):
But I think that even in the types of cases you are talking about, while the issues of how a robot is going to be gendered are maybe less salient, I think you still need to be thinking about the broader ways that that technology is part of a human sociotechnical system.
(00:33:24):
Yes, maybe it is only interacting directly with the scientist, but the existence of that robot may change the ways that certain industries operate. It may serve to legitimize certain forms of science, or certain ways of knowing. It could lead to greater funding in one area, at the expense of another.
(00:33:55):
All these broader systemic issues, which are maybe much harder to reason about, or are less immediately visible. So it is interacting with human society, and it is going to have real impact on people. But yeah, some of the issues around race and gender, probably less salient.
EW (00:34:16):
But when we talk about self-driving cars, which I think we did agree are robots, those issues do come up again sometimes.
TW (00:34:27):
Yeah. I think with self-driving cars, it is really- I think that is actually a great example, for two reasons. The first is that we can think about the broader transportation systems and who they serve.
(00:34:39):
This is something where, when I was doing my PhD in Boston, I loved having a public transit system to take around. I used the subway and the buses. I was a much bigger fan of the subway system because I was like, "I know where it is. I know how to get there. I can conceptualize it easier in my head."
(00:35:02):
But other- I cannot remember the scholar's name, but I know that people have written about the ways that there are gendered differences between uses of subways versus buses, for example. Where the subways tend to be used to get from external places in the city, into the middle of the city to do work in offices, and then to leave for example.
(00:35:28):
Whereas buses have higher ridership among women, who might need to be making multiple stops across the city, in different parts of the city over the course of the day, in ways that are more easily navigated through those bus lines. Which is something I had never thought about before.
(00:35:46):
I am guessing that there would be similar patterns of gendered use with autonomous vehicles. So that is maybe something interesting to think about.
(00:35:56):
But then the second point is that we talk about autonomous cars as, well, it is right there in the name, "autonomous." But these autonomous vehicles are rarely actually fully autonomous. There is pretty much always somebody behind the scenes who is either directly driving that robot, or more commonly it is when the robot encounters a problem, the human is able to hop in and take control.
(00:36:27):
I was just at an event in Washington DC yesterday where Missy Cummings, who is a professor at George Mason and one of the first female fighter pilots, gave a talk all about this.
(00:36:40):
She pointed out some of the problems here, where because these robots are controlled from the Philippines, there is sufficiently high latency that at least from her perspective, they are really not safe at over ten miles an hour. Which says something when you have got Waymos driving 60 miles an hour on the highway, or what have you.
(00:37:02):
But also pointing out that when San Francisco experienced the blackout in December and all of their Waymos were stranded and could not move anymore, the ways that this emphasized the importance of the humans behind the scenes. Where without those humans behind the scenes, the robots could not operate anymore.
(00:37:22):
That is something that is of course also going to be gendered and racialized in particular ways. I will also just briefly point at a parallel. We are talking about these being controlled by people in the Philippines. There is a case I talk about in the book of this robot, EngKey, which is a South Korean English language teaching robot.
(00:37:48):
The kids in the English class are told it is autonomous, this robot with this white woman's animated face on it. But the robot is not autonomous. It is teleoperated and primarily teleoperated by Filipino women, whose facial expressions are mapped onto this white woman's face, in a way that erases their labor and erases their identity.
(00:38:07):
This happens in this very specific racialized and gendered ways, due to the history of American imperialism in the Philippines and in South Korea. Which dictated who was in a position to be performing that labor cheaply, and who was stereotypically the correct instructor for English.
(00:38:29):
So we do not have those exact types of race and gender issues with the autonomous cars. But I think we should still be thinking about who is performing the labor, and why, and from where. And how does that tie into these global systems of race and gender, and capitalism and imperialism.
EW (00:38:48):
How do we balance this consideration of ethics and philosophy in our jobs, when we need to have jobs?
TW (00:38:58):
I think that is complicated. I think there is a part of me that would say, "Well, yes, we need to have jobs. But you do not need to work in a domain where you are causing harm."
(00:39:14):
I think that I would be more willing to apply that type of hardline logic, when it comes to companies like Palantir. Where it is like, "Yes, you need a job. You do not need to have a job where you are working at Palantir." I think that for other types of robotics companies, it can be hazier.
(00:39:41):
Or even you can think of academia. For me as an academic, in some ways the system of academia serves to reinforce power boundaries along race and gender and class. But what I am doing in the classroom, where I am asking students to think about these issues, is hopefully subverting that in some way. So I think it is complicated.
(00:40:14):
I do not think I would tell students, "Do not go into robotics, because there is too high of a chance that you are going to encounter ethical issues." Instead, I would say, "Do think about where you are choosing to take a job, and what type of work that company is specifically doing. And what the impacts of the technology are going to be on the communities around you. But then once you are at that company, wherever you choose to work, try to keep these issues in mind."
(00:40:49):
And I am not under the illusion that I have the ability to control what students do, once they go out into the workforce. But I am hoping that if they have been forced to think and write about these issues at least once in their life, then maybe it will lead to them making some more ethical choices once they get to these companies.
EW (00:41:19):
In your book, you prioritize the ethics of white supremacy above others. But for me, for technology, environmental impact is a harder thing to watch. I mean, I made children's toys, they were made of plastic, I am proud they taught kids to read. I think it was worth it. And yet I still look at the plastic and think, "Wow, I wish that was not there."
(00:41:49):
How do you prioritize ethics?
TW (00:41:54):
Yeah. Well, so, in the book I had to take a particular frame, I had to choose a particular scope. So I emphasize in the book that while I am talking primarily about issues of race and gender, because those issues, especially race, have been overlooked, I think, within the robotics community, of course there are broader issues of capitalism that tie through all of the above.
(00:42:25):
And similarly, you are right, issues of environmental impact are really important. I think that it is complicated in two ways. The first is that if we think about water use, one of the big environmental topics people talk about around AI is water use.
(00:42:48):
But I think that some of these concerns around water use are either overblown, or not talked about in the right way. Where some of the statistics that are used by people to talk about the amount of water that AI is using are incorrect. It is due to a misreading of the paper that they are coming from.
(00:43:13):
And then also the notion of what it means to use water is complicated. Hank Green, famous YouTuber, has a great video he made, I think like two months ago, on this where he talks about the complications around what it means to use water in AI for like an hour.
(00:43:34):
How there are different types of water. There are different types of data centers that are using water in different ways. Then even if you are using water in a way where it is evaporating out into the atmosphere, even in those situations, the water- It is not like it is going away, it is just going back into the water system.
(00:44:00):
The problem is more about where it is being used. If you are drawing a lot of water use in places that do not have a lot of water to spare, and in places where that water is not just going to go back into the reservoirs and direct it back in a usable form into the water system in a reasonable amount of time, then it is like in those very spatially contextualized ways that we need to be concerned about water use.
(00:44:38):
So I do think that water use with data centers is a big problem, but in ways that are maybe more nuanced than people talk about.
EW (00:44:47):
I totally agree. I have thought of water use as a synecdoche, as something that represents the whole problem.
CW (00:44:58):
Yeahh. Yeah.
EW (00:45:01):
Water use is inextricably related to power use. But it is easier to see a bottle of water, than it is to see a unit of power. In many ways they are convertible, although it is not something that is easy to think about. So I get that water- Yes, we say water use is bad, AI uses lots of water. But we are also saying AI uses lots of power, with no obvious results.
CW (00:45:37):
Okay. I have a larger question for you, Tom, about this. First, let me preface this by saying, there is lots of farming that uses water, that absolutely dwarfs AI. I am not a fan of AI to start with, but let us just start with that.
EW (00:45:48):
He is such not a fan of AI.
CW (00:45:49):
But I think the water use argument leads me to a larger question about criticism. My concern with water use, as you say, is it is perhaps overblown. The statistics are misread, it is dwarfed by other industries that nobody is talking about, all these kinds of things.
(00:46:03):
But I also feel like it is something that the rug can be pulled out from under the critics. "Oh, we have solved the water problem by changing cooling methods." Or, "We have made an advance and now we do not need data centers anymore."
(00:46:15):
But all the other stuff that makes AI problematic still exists. People are not talking about that, because they are spending all the time talking about water, which I agree is a very easy thing to talk about.
EW (00:46:29):
Tom is talking about the white supremacy, and race and gender and-
CW (00:46:31):
Right. No, I am not talking about Tom. I am not talking about Tom. I am talking about the discourse. I see a lot of discourse about these kinds of easy number things. Where it is like, "Oh power. Oh water. Oh data center expend." All of these things, when I see less about psychological impacts, or the impact on employment. These kinds-
EW (00:46:53):
Capitalism.
CW (00:46:53):
Those come up. But I feel like- The question I have is, is this a common thing in criticism for industries and discourse, where people pick an easy target and the target can be pulled out from under them. And then it is easy for evangelists to discredit the critics by saying, "Oh, see, that was not really an issue. Therefore everything else you are complaining about is also not an issue."
TW (00:47:22):
Yeah. A hundred percent. I think this goes back to what we are talking about, of the network of who is responsible being just enormous and hard to grapple with. It is just hard for people to think about systems.
(00:47:36):
There is a fantastic book, "Atlas of AI," where Kate Crawford talks about some of these environmental impacts in terms of these larger systems. She has got this great graphic that goes along with it, that shows how all these different things fit together. In terms of all these different disparate impacts, how they intersect with the history, and the larger systems they fit into.
(00:48:13):
Unfortunately, the graphic is so large that if you tried to print it out, you would need a quite long hallway in order to display it. In fact, I think she had an exhibit on it, at I think The Met in New York City, where they had indeed a huge room to put this graphic up.
(00:48:38):
That is part of the problem. That all these impacts and the ways that they fit into these systems, are so vast and interconnected that you can shine the spotlight on one area, and in doing so you are necessarily excluding the other areas. But I think it is an intractable problem. It is just a limit of human psychology.
CW (00:49:03):
Right, right. Yeah. We end up- <laugh> To bring up "The Good Place" again-
EW (00:49:08):
<laugh>
CW (00:49:08):
I fear sometimes we end up like Chidi with his almond milk in his coffee, losing the bigger picture. Like, "Oh, this is what has damned me, is my almond milk in my coffee." No, it is that you have paralyzed yourself.
TW (00:49:23):
Yes. "The Good Place," a show that when I first saw the trailer for it, I was like, "This looks awful." Then I started watching it for some reason and I was like, "This is amazing." But it is easy for me to say that, as a professor of ethics.
CW (00:49:37):
Yes. Right. <laugh>
EW (00:49:41):
Okay. To that point, do not the students, the computer science students, the robotic students, who are learning control theory and robotics and SLAM localization algorithms and dealing with cameras- I could design a whole major. Do they really need to spend time learning the history, philosophy and ethics of everything they might possibly do? Can they not just go back to-
CW (00:50:14):
Look, if Mudd made me take a class on Kant, then everyone else should have to.
EW (00:50:20):
<laugh> You chose that. I took the history of science.
TW (00:50:23):
Yeah. No. I think absolutely that they do. I think that asking students to take a semester is really the bare minimum amount of time that students should be spending on this. Really, I think what we also need is what people refer to as ethics across the curriculum. Where students encounter this type of ethical deliberation and societal context in every class they take, even if it is just for a little bit.
(00:50:59):
Otherwise, what happens, and I have seen this at Mines, is that students their first year get exposure to some of these issues. They then do not encounter having to think about people at all for several years. And then it is their senior year, when it comes back to senior design or when they are taking an elective like one of my classes, that they are asked to think about it again.
(00:51:25):
But in the meantime, the sense that they are given is, "Well, maybe it does not matter that much, because we are going to put that on hold and not talk about that at all this semester." I think this is particularly important today with computer science, because of the rate at which computer science is able to move.
(00:51:48):
So you have got these students who are graduating, and especially in the age of AI where they are able to vibe code a product in 20 minutes. We have given students enough knowledge to be dangerous, to be extremely dangerous, in order to very, very quickly create a product that has the ability to potentially immediately impact millions of people.
(00:52:16):
So in the same way that professional engineers have licensing requirements- And have to have knowledge of the safety of how they are building their bridges, and know that they are responsible or liable if something goes wrong, if their bridge collapses.
(00:52:36):
We need to be instilling that same sense of responsibility and awareness into our computer science students, in ways that currently we just do not do at all.
EW (00:52:52):
I agree. Although it is a little difficult to have that awareness of responsibility without the feeling of power to change or authority. It gives a powerlessness to it.
TW (00:53:08):
Yeah. I think this is something where, for this reason, even in my robot ethics class, I make sure that we spend a week talking about design. So that it is not just all doom and gloom. It is also, "Well, here are specific paradigms and tools you could actually use, that might mitigate your risk of falling into these traps."
(00:53:29):
I think that this is also somewhere where when people talk about tech ethics, there tends to be an over-focusing on moral philosophy. And not as much, not only on these broader system level issues, but also specifically on design.
(00:53:51):
The philosopher, Peter-Paul Verbeek, talks in his writing about the ways that design is a way of doing ethics by other means. I think that if we move towards more practical, hands-on ways of thinking about ethics, then that might help to counter some of this doom and gloom and depression that you otherwise get in an ethics class.
EW (00:54:28):
Is that like thinking about the accessibility features as you are doing the design of your system? Or what do you have for that section? What are the design features that would make this not terrible?
TW (00:54:40):
Yeah. So we really hammer home on methods like participatory design, where you work with communities directly to build the systems that meet their needs. We talk about design justice, which is a framework where you say, "Instead of designing this technology paternalistically for this community, we will give them the tools that they need so that they can design it for themselves."
(00:55:10):
So instead of us saying, "Well, these are the rules that the robot needs to follow," maybe you give the communities an interface that they can use to instill their systems, values or norms or role ethics or what have you, into the system on their own.
(00:55:29):
Now, I will also say that this is complicated as well. Because I think the further you go towards empowering communities, the more you potentially abdicate your own moral responsibilities. If you say, "Well, we will just let the community decide for themselves how they are going to use this technology," then you are just avoiding making any ethical decisions yourselves, and just hoping that your users choose for the best.
(00:55:58):
This has been a problem with Apple, for example. They got some pushback on how the Siri voice is this white woman's voice. They started creating more voices for Siri. I think they ultimately made it so that the user has to choose, so that there is not a default voice.
(00:56:24):
I cannot remember if this is Apple or Amazon, but for one of these companies, where there is not a default voice and the user has to choose. To some extent that does decrease the likelihood of the system by default reinforcing these associations between femininity and these types of like administrative labor.
(00:56:47):
But on the other hand, it just pushes the decision to the users. You can expect that if those stereotypes are the reasons why the system was designed that way, then users are probably most likely going to choose the voice that plays into those stereotypes, for the same reasons.
CW (00:57:04):
Yeah. Why would we assume that the user community is a separate domain of humanity from the people who made it.
TW (00:57:11):
So it really depends. I think if you are working with- Like my lab does work with unhoused survivors of domestic abuse, for example. That is a space in which we want to make sure that we are listening to that group, and working with them to design the technology.
(00:57:30):
This is something where things come back to improv, where we have been developing new improvisational role play based design methods. That allow the community members to participate in robot interaction design in different ways.
(00:57:44):
Either by sharing stories of ways things have gone wrong in the past, that actors then act out, to give a sense of what a design might look like. Or the community members giving feedback, and directing those role play activities. Or them jumping in and role playing in these role played interactions between a hypothetical user and a robot product.
(00:58:15):
This is somewhere where we do want to have tight connection with those communities. And where I think we would feel more confidence in giving more agency over to those specific community members.
(00:58:29):
But obviously in other domains, like I do not know, if you were- I have written a lot, and there is a whole chapter in the book, about robots and policing. I do not think we should be developing robots for the police.
(00:58:39):
But if you were to design a police robotic system and say, "Well, we will just give it over to the police, and allow them to decide for themselves when and how the robot should be surveilling people," then for obvious reasons that might not turn out the way the roboticists would hope.
EW (00:58:56):
I mentioned Philip Koopman's "Embodied AI Safety" book, which is more technical, less ethical. More directive, implementation directive.
CW (00:59:07):
Prescriptive?
EW (00:59:07):
Prescriptive? Yeah. One of the ideas that he brought to me was moral crumple zones. The idea that, "Sure, I am not responsible. I told you not to run it into a wall." Or, "I told you not to use the AI."
CW (00:59:27):
"It is on page 50 of the manual, Appendix C."
EW (00:59:30):
Yes. I really like that as a idea, at least for myself, that you can tell people- You set people up for failure, and that is not okay either.
TW (00:59:47):
Yeah. I think that this ties in to these things we have been talking about, about who to hold responsible. Some of the types of moral crumple zones that we see in other places are with pilot in airlines.
(01:00:00):
Where when there is a plane crash, people tend to blame the pilot, even in cases where it is really not their fault. The airlines might sometimes design things in ways where it makes it easier cognitively to blame the pilot, in ways that avoid liability then for themselves.
(01:00:26):
For robot design, I think that it depends on the type of robot, who is likely to fall into that crumple zone. If it is the person who is using the robot, if it is the person in the Philippines who is remotely controlling it, et cetera. I think in all of these cases, again, yeah we need to be figuring out ways that we can reasonably hold the companies responsible.
(01:00:51):
But we also need to be identifying ways that we can inform the general public of this broader network of stakeholders that are involved, and of how the technology really works. So that they are not just by default putting people into the crosshairs that do not deserve it.
EW (01:01:11):
Tom, it has been wonderful to talk to you. Do you have any thoughts you would like to leave us with?
TW (01:01:16):
Yeah. I think that as a closing thought, the high level point I would make is that we cannot afford to think about robots purely as technical devices. Whether we are talking about a Android, or a cute fuzzy home robot pet, or if we are talking about a glass box that is exploring the bottom of a river, we always need to be thinking about the humans that are involved.
EW (01:01:46):
Our guest has been Dr. Tom Williams, Professor of Computer Science at the Colorado School of Mines, and author of "Degrees of Freedom: On Robotics and Social Justice."
CW (01:01:56):
Thanks, Tom.
TW (01:01:58):
Thank you.
EW (01:01:58):
Thank you to Christopher for producing and co-hosting. Thank you to David Goldberg for the introduction. And thank you for listening. You can always contact us at show@embedded.fm or hit the contact link on embedded.fm.
(01:02:11):
Now a quote to leave you with. "With outright horror, he refuses any responsibility for the thought that machines could take the place of people, or that anything like life, love, or rebellion could ever awaken in their cogwheels. He would regard this somber vision as an unforgivable over-evaluation of mechanics or as a severe insult to life."
(01:02:37):
That is the inventor of the word "robot," Karel Capek, in an explanation of how everyone got his idea of robots wrong. It is an IEEE Spectrum, as a translation. I will link that in the show notes.
