The following is a conversation with Kate Darling, a researcher at MIT interested in
social robotics, robot ethics, and generally how technology intersects with society.
She explores the emotional connection between human beings and lifelike machines,
which for me is one of the most exciting topics in all of artificial intelligence.
As she writes in her bio, she is a caretaker of several domestic robots,
including her plio dinosaur robots named Yochai, Peter, and Mr. Spaghetti.
She is one of the funniest and brightest minds I've ever had the fortune to talk to.
This conversation was recorded recently, but before the outbreak of the pandemic.
For everyone feeling the burden of this crisis, I'm sending love your way.
This is the Artificial Intelligence Podcast. If you enjoy it, subscribe on YouTube,
review it with five stars on Apple Podcasts, support on Patreon,
or simply connect with me on Twitter at Lex Friedman spelled F-R-I-D-M-A-N.
As usual, I'll do a few minutes of ads now and never any ads in the middle that can break the
flow of the conversation. I hope that works for you and doesn't hurt the listening experience.
Quick summary of the ads. Two sponsors, Masterclass and ExpressVPN. Please consider supporting
the podcast by signing up to masterclass at masterclass.com slash lex and getting expressvpn
at expressvpn.com slash lex pod. This show is sponsored by Masterclass.
Sign up at masterclass.com slash lex to get a discount and to support this podcast.
When I first heard about Masterclass, I thought it was too good to be true.
For $180 a year, you get an all-access pass to watch courses from to list some of my favorites,
Chris Hadfield on Space Exploration, Neil deGrasse Tyson on Scientific Thinking and Communication,
Will Wright, creator of SimCity and Sims, Love Those Games on Game Design, Carlos Santana on Guitar,
Garry Kasparov on Chess, Daniel Nagrano on Poker, and many more.
Chris Hadfield explaining how rockets work and the experience of being launched into space alone
is worth the money. By the way, you can watch it on basically any device. Once again,
sign up on masterclass.com slash lex to get a discount and to support this podcast.
This show is sponsored by ExpressVPN. Get it at expressvpn.com slash lex pod
to get a discount and to support this podcast. I've been using ExpressVPN for many years.
I love it. It's easy to use, press the big power on button, and your privacy is protected.
And if you like, you can make it look like your location is anywhere else in the world.
I might be in Boston now, but it can make it look like I'm in New York, London, Paris, or
anywhere else. This has a large number of obvious benefits. Certainly, it allows you to access
international versions of streaming websites like the Japanese Netflix or the UK Hulu.
ExpressVPN works on any device you can imagine. I use it on Linux. Shout out to Ubuntu,
2004, Windows, Android, but it's available everywhere else too. Once again, get it at
expressvpn.com slash lex pod to get a discount and to support this podcast.
And now here's my conversation with Kate Darling.
You co-taught Robot Ethics at Harvard. What are some ethical issues that arise
in the world with robots? Yeah, that was a reading group that I did when I,
like at the very beginning, first became interested in this topic. So I think if I
taught that class today, it would look very, very different. Robot Ethics, it sounds very
science fiction-y, especially did back then. But I think that some of the issues that
people in Robot Ethics are concerned with are just around the ethical use of robotic
technology in general. So for example, responsibility for harm, automated weapons systems,
things like privacy and data security, things like automation and labor markets.
And then personally, I'm really interested in some of the social issues that come out of our
social relationships with robots. One-on-one relationship with robots. I think most of
the stuff we have to talk about is like one-on-one social stuff. That's what I love. I think that's
what you love as well and our expert in. But as a societal level, there's a presidential
candidate now, Andrew Yang, running. Concerned about automation and robots and AI in general,
taking away jobs. He has a proposal of UBI, universal basic income of everybody gets a
thousand bucks as a way to save you if you lose your job from automation to allow you time to
discover what it is that you would like to or even love to do. Yes. So I lived in Switzerland
for 20 years and universal basic income has been more of a topic there, separate from the whole
robots and jobs issue. So it's so interesting to me to see these Silicon Valley people latch
onto this concept that came from a very left-wing socialist, kind of a different place in Europe.
But on the automation and labor markets topic, I think that it's very... So sometimes in those
conversations, I think people overestimate where robotic technology is right now and we also have
this fallacy of constantly comparing robots to humans and thinking of this as a one-to-one
replacement of jobs. So even Bill Gates a few years ago said something about maybe we should
have a system that taxes robots for taking people's jobs. And it just... I mean, I'm sure
that was taken out of context. He's a really smart guy, but that sounds to me like kind of
viewing it as a one-to-one replacement versus viewing this technology as kind of a supplemental
tool that, of course, is going to shake up a lot of stuff. It's going to change the job landscape,
but I don't see robots taking all the jobs in the next 20 years. That's just not how it's going to
work. Right. So maybe drifting into the land of more personal relationships with robots and
interaction and so on. I got to warn you. I may ask some silly philosophical questions,
I apologize. Oh, please do. Okay. Do you think humans will abuse robots in their interaction?
So you've had a lot of... And what we'll talk about is sort of anthropomorphization and
this intricate dance, emotional dance between human and robot, but there seems to be also a
darker side where people, when they treat the other as servants especially, they can be a
little bit abusive or a lot abusive. Do you think about that? Do you worry about that?
Yeah, I do think about that. So I mean, one of my main interests is the fact that people
subconsciously treat robots like living things. And even though they know that they're interacting
with a machine and what it means in that context to behave violently. I don't know if you could say
abuse because you're not actually abusing the inner mind of the robot that robot doesn't have
any feelings. As far as you know. Well, yeah. It also depends on how we define feelings and
consciousness, but I think that's another area where people kind of overestimate where we currently
are with the technology. The robots are not even as smart as insects right now. And so I'm not worried
about abuse in that sense, but it is interesting to think about what does people's behavior towards
these things mean for our own behavior? Is it desensitizing the people to be verbally abusive
to a robot or even physically abusive? And we don't know. Right. It's a similar connection from
if you play violent video games. What connection does that have to desensitization to violence?
I haven't read literature on that. I wonder about that. Because everything I've heard,
people don't seem to any longer be so worried about violent video games. Correct. The research on it is
it's a difficult thing to research. So it's sort of inconclusive, but we seem to have gotten the
sense, at least as a society, that people can compartmentalize. When it's something on a screen
and you're shooting a bunch of characters or running over people with your car that doesn't
necessarily translate to you doing that in real life, we do, however, have some concerns about
children playing violent video games. And so we do restrict it there. I'm not sure that's based on
any real evidence either, but it's just the way that we've kind of decided we want to be a little
more cautious there. And the reason I think robots are a little bit different is because there is a
lot of research showing that we respond differently to something in our physical space than something
on a screen. We will treat it much more viscerally, much more like a physical actor. And so it's
totally possible that this is not a problem. And it's the same thing as violence and video games,
you know, maybe, you know, restrict it with kids to be safe, but adults can do what they want.
But we just need to ask the question again, because we don't have any evidence at all yet.
Maybe there's an intermediate place too. I did my research on Twitter. By research, I mean
scrolling through your Twitter feed. You mentioned that you were going at some point to an animal
law conference. So I have to ask, do you think there's something that we can learn
from animal rights that guides our thinking about robots?
Oh, I think there is so much to learn from that. I'm actually writing a book on it right now,
that's why I'm going to this conference. So I'm writing a book that looks at the history of animal
domestication and how we've used animals for work, for weaponry, for companionship. And, you know,
one of the things the book tries to do is move away from this fallacy that I talked about of
comparing robots and humans, because I don't think that's the right analogy. But I do think that on
a social level, even on a social level, there's so much that we can learn from looking at that
history, because throughout history, we've treated most animals like tools, like products. And then
some of them we've treated differently. And we're starting to see people treat robots in really
similar ways. So I think it's a really helpful predictor to how we're going to interact with
the robots. Do you think we'll look back at this time, like 100 years from now, and see what we
do to animals as like similar to the way we view like the Holocaust in World War II?
That's a great question. I mean, I hope so. I am not convinced that we will. But I often wonder,
you know, what are my grandkids going to view as, you know, abhorrent that my generation did,
that they would never do. And I'm like, well, what's the big deal? You know, it's a fun question
to ask yourself. It always seems that there's atrocities that we discover later. So the things
that at the time people didn't see as, you know, you look at everything from slavery
to any kinds of abuse throughout history, to the kind of insane wars that were happening,
to the way war was carried out and rape and the kind of violence that was happening during war.
And that we now, you know, we see as atrocities, but at the time, perhaps didn't as much. And so now
I have this intuition that I have this worry. Maybe you're going to probably criticize me,
but I do anthropomorphize robots. I have, I don't see a fundamental philosophical
difference in a robot and a human being in terms of once the capabilities are matched.
So the fact that we're really far away doesn't, in terms of capabilities in the net from,
from natural language processing, understanding a generation to just reasoning and all that stuff.
I think once you solve it, I see the, this is a very gray area. And I don't feel comfortable
with the kind of abuse that people throw at robots, subtle, but I can see it becoming,
I can see basically a civil rights movement for robots in the future. Do you think, let me put
it in the form of a question, do you think robots should have some kinds of rights?
Well, it's interesting because I came at this originally from your perspective. I was like,
you know what, there's no fundamental difference between technology and like human consciousness,
like we can probably recreate anything. We just don't know how yet. And so
there's no reason not to give machines the same rights that we have once, like you say,
they're kind of on an equivalent level. But I realized that that is kind of a far future question.
I still think we should talk about it because I think it's really interesting. But I realized that
it's actually, we might need to ask the robot rights question even sooner than that. While
the machines are still, quote unquote, really dumb and not on our level, because of the way
that we perceive them. And I think one of the lessons we learned from looking at the history
of animal rights and one of the reasons we may not get to a place in 100 years where we view it
as wrong to eat or otherwise use animals for our own purposes is because historically, we've always
protected those things that we relate to the most. So one example is whales. No one gave a
shit about the whales. Am I allowed to swear? Yeah. Yeah, no one gave a shit about the whales
until someone recorded them singing. And suddenly people were like, Oh, this is a beautiful creature.
And now we need to save the whales. And that started the whole Save the Whales movement in
the 70s. So as much as I, and I think a lot of people want to believe that we care about
consistent biological criteria, that's not historically how we formed our alliances.
Yeah. So why do we believe that all humans are created equal? Killing of a human being,
no matter who the human being is, that's what I meant by equality is bad. And then because I'm
connecting that to robots, and I'm wondering whether mortality, so the killing act is what
makes something, that's the fundamental first right. So I am currently allowed to take a shotgun
and shoot a Roomba. I think I'm not sure, but I'm pretty sure it's not considered murder,
right? Or even shutting them off. So that's where the line appears to be, right? Is this
mortality a critical thing here? I think here again, like the animal analogy is really useful
because you're also allowed to shoot your dog, but people won't be happy about it. So we do
give animals certain protections from like, you know, you're not allowed to torture your dog and
set it on fire, at least in most states and countries. But you're still allowed to treat
it like a piece of property in a lot of other ways. And so we draw these arbitrary lines all
the time. And there's a lot of philosophical thought on why viewing humans as something unique
is not, is just speciesism and not, you know, based on any criteria that would actually
justify making a difference between us and other species.
Do you think in general, most people are good? Do you think there's evil and good in all of us?
Is that's revealed through our circumstances and through our interactions?
I like to view myself as a person who like, believes that there's no absolute evil and good
and that everything is, you know, gray. But I do think it's an interesting question. Like,
when I see people being violent towards robotic objects, you said that bothers you because
the robots might someday, you know, be smart. And is that what?
Well, it bothers me because it reveals, so I personally believe, because I've
studied way too much, I'm Jewish, I studied the Holocaust and World War II exceptionally well,
I personally believe that most of us have evil in us. That what bothers me is the abusive robots
reveals the evil in human beings. Yeah. And I think it doesn't just bother me. I think it's
an opportunity for roboticists to make, help people find the better sides of the angels
of their nature, right? That abuse isn't just a fun side thing. That's you revealing a dark part
that you shouldn't, that should be hidden deep inside. Yeah, I mean, you laugh, but
some of our research does indicate that maybe people's behavior towards robots reveals something
about their tendencies for empathy generally, even using very simple robots that we have today that
like clearly don't feel anything. So, you know, Westworld is maybe, you know, not so far off and
it's like, you know, depicting the bad characters as willing to go around and shoot and rape the
robots and the good characters is not wanting to do that, even without assuming that the robots
have consciousness. So, there's an opportunity, it's interesting, there's an opportunity to almost
practice empathy. Robots is an opportunity to practice empathy. I agree with you. Some people
would say, why are we practicing empathy on robots instead of, you know, on our fellow humans or
on animals that are actually alive and experience the world? And I don't agree with them because I
don't think empathy is a zero sum game and I do think that it's a muscle that you can train and
that we should be doing that, but some people disagree. So, the interesting thing, you've heard,
you know, raising kids, sort of asking them or telling them to be nice to the smart speakers,
to Alexa and so on, saying please and so on during the requests. I don't know if I'm a huge fan of
that idea because, yeah, that's towards the idea of practicing empathy. I feel like politeness,
I'm always polite to all the systems that we build, especially anything that's speech interaction
based, like when we talk to the car, I always have a pretty good detector for please. I feel like
there should be a room for encouraging empathy in those interactions. Yeah. Okay, so I agree with
you, so I'm going to play devil's advocate. Sure. What is the devil's advocate argument there?
The devil's advocate argument is that if you are the type of person who has abusive tendencies or
needs to get some sort of behavior like that out, needs an outlet for it, that it's great to have
a robot that you can scream at so that you're not screaming at a person. And we just don't know
whether that's true, whether it's an outlet for people or whether it just kind of, as my friend
once said, trains their cruelty muscles and makes them more cruel in other situations.
Oh boy, yeah. And that expands to other topics, which I don't know. There's a
topic of sex, which is weird one that I tend to avoid from robotics perspective,
and mostly general public doesn't. They talk about sex robots and so on. Is that an area you've
touched at all research-wise? Because that's what people imagine any kind of interaction
between human and robot that shows any kind of compassion. They immediately think from a product
perspective in the near term is sort of expansion of what pornography is and all that kind of stuff.
Yeah. Do researchers touch this? That's kind of you to characterize it as though they're
thinking rationally about product. I feel like sex robots are just such a titillating news hook
for people that they become like the story. And it's really hard to not get fatigued by it when
you're in the space because you tell someone you do human-robot interaction. Of course,
the first thing they want to talk about is sex robots. Really? Yeah, it happens a lot. And
it's unfortunate that I'm so fatigued by it because I do think that there are some interesting
questions that become salient when you talk about sex with robots. See, what I think would happen
when people get sex robots, like let's talk guys, okay, guys get female sex robots, what I think
there's an opportunity for is an actual, like they'll actually interact. What I'm trying to say,
they won't, outside of the sex would be the most fulfilling part. Like the interaction,
it's like the folks who, there's movies in this, right, who pay a prostitute and then end up just
talking to her the whole time. So I feel like there's an opportunity, it's like most guys and
people in general joke about the sex act, but really people are just lonely inside and they're
looking for connection, many of them. And it'd be unfortunate if that connection is established
through the sex industry. I feel like it should go into the front door of like people are lonely
and they want a connection. Well, I also feel like we should kind of de-stigmatize the sex industry
because even prostitution, like there are prostitutes that specialize in disabled people
who don't have the same kind of opportunities to explore their sexuality. So I feel like we
should de-stigmatize all of that generally. But yeah, that connection and that loneliness is an
interesting topic that you bring up because while people are constantly worried about robots
replacing humans and oh, if people get sex robots and the sex is really good and they won't want
their partner or whatever, but we rarely talk about robots actually filling a hole where there's
nothing and what benefit that can provide to people. Yeah, I think that's an exciting,
there's a whole giant, there's a giant hole that's unfillable by humans. It's asking too much of
your friends and people you're in a relationship with in your family to fill that hole. Because
it's exploring the full complexity and richness of who you are. Like who are you really?
People, your family doesn't have enough patience to really sit there and listen to who are you
really. And I feel like there's an opportunity to really make that connection with robots.
I just feel like we're complex as humans and we're capable of lots of different types of
relationships. So whether that's with family members, with friends, with our pets, or with
robots, I feel like there's space for all of that and all of that can provide value in a different
way. Yeah, absolutely. So I'm jumping around. Currently, most of my work is in autonomous
vehicles. So the most popular topic among general public is the trolley problem. So most
roboticist kind of hate this question, but what do you think of this thought experiment?
What do you think we can learn from it outside of the silliness of the actual application of it to
the autonomous vehicle? I think it's still an interesting ethical question and that in itself,
just like much of the interaction with robots, has something to teach us. But from your perspective,
do you think there's anything there? Well, I think you're right that it does have something to
teach us. But I think what people are forgetting in all of these conversations is the origins of
the trolley problem and what it was meant to show us, which is that there is no right answer and
that sometimes our moral intuition that comes to us instinctively is not actually what we should
follow if we care about creating systematic rules that apply to everyone. So I think that
as a philosophical concept, it could teach us at least that. But that's not how people are
using it right now. And these are friends of mine and I love them dearly and their project
adds a lot of value. But if we're viewing the moral machine project as what we can learn from
the trolley problems, the moral machine is, I'm sure you're familiar, it's this website that you
can go to and it gives you different scenarios like, oh, you're in a car, you can decide to run over
these two people or this child. What do you choose? Do you choose the homeless person? Do you choose
the person who's jaywalking? And so it pits these like moral choices against each other and then
tries to crowdsource the quote unquote correct answer, which is really interesting and I think
valuable data. But I don't think that's what we should base our rules and autonomous vehicles on
because it is exactly what the trolley problem is trying to show, which is your first instinct might
not be the correct one if you look at rules that then have to apply to everyone and everything.
So how do we encode these ethical choices in interaction with robots? So for example,
with autonomous vehicles, there is a serious ethical question of, do I protect myself?
Does my life have higher priority than the life of another human being? Because that changes
certain controlled decisions that you make. So if your life matters more than other human beings,
then you'd be more likely to swerve out of your current lane. So currently automated emergency
braking systems that just break, they don't ever swerve. Right. So swerving into oncoming traffic
or or no, just in a different lane can cause significant harm to others, but it's possible
that it causes less harm to you. So that's a difficult ethical question. Do you have a hope
that like the trolley problem is not supposed to have a right answer, right? Do you hope that
when we have robots at the table, we'll be able to discover the right answer for some of these
questions? Well, what's happening right now, I think is this question that we're facing of,
what ethical rules should we be programming into the machines is revealing to us that
our ethical rules are much less programmable than we probably thought before. And so that's a really
valuable insight, I think that these issues are very complicated. And that in a lot of these
cases, you can't really make that call, not even as a legislator. And so what's going to happen in
reality, I think is that car manufacturers are just going to try and avoid the problem and avoid
liability in any way possible. Or they're going to always protect the driver because who's going
to buy a car if it's programmed to kill someone, kill you instead of someone else. So that's what's
going to happen in reality. But what did you mean by like once we have robots at the table,
like, do you mean when they can help us figure out what to do? No, I mean, when robots are part
of the ethical decisions. So no, no, not they help us. Well,
Oh, you mean when it's like, should I run over a robot or a person?
Right, that kind of thing. So when you it's exactly what you said, which is when you have to
encode the ethics into an algorithm, you start to try to really understand what are the fundamentals
of the decision making process, you make this make certain decisions. Should you,
like capital punishment, should you take a person's life or not to punish them for a certain crime?
Sort of, you can use, you can develop an algorithm to make that decision, right? And
the hope is that the act of making that algorithm, however you make it, so there's a few approaches,
will help us actually get to the core of what is right and what is wrong under our current
societal standards. But isn't that what's happening right now? And we're realizing that we don't have
a consensus on what's right and wrong. You mean in politics in general? Well, like when we're thinking
about these trolley problems and autonomous vehicles and how to program ethics into machines
and how to make AI algorithms fair and equitable, we're realizing that this is so complicated.
And it's complicated in part because there doesn't seem to be a one right answer in any of these cases.
Do you have a hope for, like one of the ideas of the moral machine is that crowd sourcing can help us
converge towards democracy, can help us converge towards the right answer? Do you have a hope for
crowd sourcing? Well, yes and no. So I think that in general, I have a legal background and policy
making is often about trying to suss out what rules does this particular society agree on and then
trying to codify that. So the law makes these choices all the time and then tries to adapt
according to change in culture. But in the case of the moral machine project, I don't think that
people's choices on that website necessarily reflect what laws they would want in place.
I think you would have to ask them a series of different questions in order to get up what their
consensus is. I agree. But that has to do more with the artificial nature of, I mean, they're
showing some cute icons on a screen. That's almost, so if you, for example, we would do a lot of work
in virtual reality. And so if you put those same people into virtual reality where they have to
make that decision, their decision would be very different, I think. I agree with that. That's one
aspect. And the other aspect is it's a different question to ask someone, would you run over the
homeless person or the doctor in this scene? Or do you want cars to always run over the homeless
people? I think, yeah. So let's talk about anthropomorphism. To me, anthropomorphism,
if I can pronounce it correctly, is one of the most fascinating phenomena from both the engineering
perspective and the psychology perspective, machine learning perspective in robotics in general.
Can you step back and define anthropomorphism, how you see it in general terms in your work?
Sure. So anthropomorphism is this tendency that we have to project human-like traits and behaviors
and qualities onto nonhumans. And we often see it with animals, like we'll project emotions on
animals that may or may not actually be there. We often see that we're trying to interpret things
according to our own behavior when we get it wrong. But we do it with more than just animals.
We do it with objects, teddy bears. We see faces in the headlights of cars. And we do it with robots,
very, very extremely. Do you think that can be engineered? Can that be used to enrich an interaction
between an AI system and a human? Oh, yeah, for sure. And do you see it being used that way often?
I haven't seen, whether it's Alexa or any of the smart speaker systems, often trying to optimize
for the anthropomorphization. You said you haven't seen? I haven't seen. They keep moving away from
that. I think they're afraid of that. I only recently found out, but did you know that Amazon
has a whole team of people who are just there to work on Alexa's personality?
So I know, it depends on you, my personality. I didn't know that exact thing. But I do know that
how the voice is perceived is worked on a lot, whether if it's a pleasant feeling about the
voice, but that has to do more with the texture of the sound and the audio and so on. But personality
is more like... It's like what's her favorite beer when you ask her. And the personality
team is different for every country too. There's a different personality for German Alexa than
there is for American Alexa. That said, I think it's very difficult to use the, or really, really
harness the anthropomorphism with these voice assistants because the voice interface is still
very primitive. And I think that in order to get people to really suspend their disbelief and treat
a robot like it's alive, less is sometimes more. You want them to project onto the robot and you
want the robot to not disappoint their expectations for how it's going to answer or behave in order
for them to have this kind of illusion. And with Alexa, I don't think we're there yet or Siri,
that they're just not good at that. But if you look at some of the more animal-like robots,
like the baby seal that they use with the dementia patients, it's a much more simple design. It
doesn't try to talk to you. You can't disappoint you in that way. It just makes little movements
and sounds and people stroke it and it responds to their touch. And that is a very effective way
to harness people's tendency to treat the robot like a living thing. Yeah. So you bring up some
interesting ideas in your paper chapter, I guess, anthropomorphic framing, human-robot interaction
that I read the last time we scheduled this. That was a long time ago.
What are some good and bad cases of anthropomorphism in your perspective?
Like, when is it good, when is it bad? Well, I should start by saying that while design can
really enhance the anthropomorphism, it doesn't take a lot to get people to treat a robot like
it's alive. Over 85% of Roombas have a name, which I don't know the numbers for your regular
type of vacuum cleaner, but they're not that high, right? So people will feel bad for the Roomba
when it gets stuck. They'll send it in for repair and want to get the same one back. And that one
is not even designed to make you do that. So I think that some of the cases where it's maybe
a little bit concerning that anthropomorphism is happening is when you have something that's
supposed to function like a tool and people are using it in the wrong way. And one of the concerns
is military robots. Early 2000s, which is a long time ago, iRobot, the Roomba company,
made this robot called the PacBot that was deployed in Iraq and Afghanistan with the
bomb disposal units that were there. And the soldiers became very emotionally attached to
the robots. And that's fine until a soldier risks his life to save a robot, which you
really don't want. But they were treating them like pets, like they would name them,
they would give them funerals with gun salutes, they would get really upset and traumatized
when the robot got broken. So in situations where you want a robot to be a tool, in particular,
when it's supposed to do a dangerous job that you don't want a person doing,
it can be hard when people get emotionally attached to it. That's maybe something that
you would want to discourage. Another case for concern is maybe when companies try to
leverage the emotional attachment to exploit people. So if it's something that's not in the
consumer's interest, trying to sell them products or services or exploit an emotional connection
to keep them paying for a cloud service for a social robot or something like that,
might be, I think that's a little bit concerning as well.
Yeah, the emotional manipulation, which probably happens behind the scenes now
with some social networks and so on, but making it more explicit. What's your favorite robot?
Fictional or real?
No, real. Real robot, which you have felt a connection with, or not anthropomorphic
connection, but I mean, you sit back and say, damn, this is an impressive system.
Wow, so two different robots. So the Pleo baby dinosaur robot that is no longer sold that came
out in 2007, that one I was very impressed with. But from an anthropomorphic perspective,
I was impressed with how much I bonded with it, how much I wanted to believe that it had this
inner life. Can you describe Pleo? Can you describe what it is? How big is it? What can it actually
do? Yeah, Pleo is about the size of a small cat. It had a lot of motors that gave it this kind of
lifelike movement. It had things like touch sensors and an infrared camera. So it had
all these cool little technical features, even though it was a toy. And the thing that really
struck me about it was that it could mimic pain and distress really well. So if you held it up
by the tail, it had a tilt sensor that told it what direction it was facing and it would start to
squirm and cry out. If you hit it too hard, it would start to cry. So it was very impressive
in design. And what's the second robot that you said there might have been two that you liked?
Yeah, so the Boston Dynamics robots are just impressive feats of engineering.
Have you met them in person? Yeah, I recently got a chance to go visit. And I was always one
of those people who watched the videos and was like, this is super cool, but also it's a product
video. I don't know how many times that they had to shoot this to get it right. But visiting them,
I'm pretty sure that I was very impressed. Let's put it that way.
Yeah. And in terms of the control, I think that was a transformational moment for me
when I met Spotmini in person. Yeah. Because, okay, maybe this is a psychology experiment,
but I anthropomorphized the crap out of it. So I immediately, it was like my best friend.
Right? I think it's really hard for anyone to watch Spotmove and not feel like it has agency.
Yeah, this movement, especially the arm on Spotmini, really obviously looks like a head.
Yeah. And they say, no, wouldn't mean it that way. But it obviously, it looks exactly like that.
And so it's almost impossible to not think of it as almost like the baby dinosaur, but slightly
larger. And this movement of the, of course, the intelligence is their whole idea is that
it's not supposed to be intelligent. It's a platform on which you build
higher intelligence. It's actually really, really dumb. It's just a basic movement platform.
Yeah. But even dumb robots can, like we can immediately respond to them in this visceral way.
What are your thoughts about Sophia, the robot and this kind of mix of some basic natural
language processing and basically an art experiment? Yeah. An art experiment is a good
way to characterize it. I'm much less impressed with Sophia than I am with Boston Dynamics.
She said she likes you. She said she admires you. Yeah, she followed me on Twitter at some
point. Yeah. She tweets about how much she likes you. So what does that mean? I have to be nice?
No, I don't know. See, I was emotionally manipulating. No, how do you think of
the whole thing that happened with Sophia is quite a large number of people kind of immediately
had a connection and thought that maybe we're far more advanced with robotics than we are,
or actually didn't even think much. I was surprised how little people cared
that they kind of assumed that, well, of course AI can do this. Yeah. And then they, if they assume
that, I felt they should be more impressed. Well, people really overestimate where we are.
And so when something, I don't even think Sophia was very impressive or is very impressive. I
think she's kind of a puppet, to be honest. But yeah, I think people are a little bit influenced
by science fiction and pop culture to think that we should be further along than we are.
So what's your favorite robots in movies and fiction? Wally. Wally. What do you like about
Wally? The humor, the cuteness, the perception control systems operating on Wally that makes it
over? Just in general? The design of Wally the robot, I think that animators figured out,
starting in the 1940s, how to create characters that don't look real, but look like something
that's even better than real, that we really respond to and think is really cute. They figured
out how to make them move and look in the right way. And Wally is just such a great example of that.
You think eyes, big eyes or big something that's kind of eye-ish. So it's always playing on some
aspect of the human face, right? Often, yeah. So big eyes. Well, I think one of the first
animations to really play with this was Bambi. And they weren't originally going to do that.
They were originally trying to make the deer look as lifelike as possible. Like they brought deer
into the studio and had a little zoo there so that the animators could work with them.
And then at some point, they were like, hmm, if we make really big eyes and like a small
nose and like big cheeks, kind of more like a baby face, then people like it even better than
if it looks real. Do you think the future of things like Alexa in the home has possibly to
take advantage of that, to build on that, to create these systems that are better than real,
that create a close human connection? I can pretty much guarantee you without having any
knowledge that those companies are working on that, on that design behind the scenes.
Like, I'm pretty sure. I totally disagree with you. Really? So that's what I'm interested in.
I'd like to build such a company. I know a lot of those folks and they're afraid of that
because you don't, how do you make money off of it? Well, but even just like making Alexa look
a little bit more interesting than just like a cylinder would do so much. It's an interesting
thought, but I don't think people are, from Amazon perspective, are looking for that kind
of connection. They want you to be addicted to the services provided by Alexa, not to the device.
So the device itself, it's felt that you can lose a lot because if you create a connection
and then it does, it creates more opportunity for frustration, for negative stuff than it
does for positive stuff, is I think the way they think about it.
That's interesting. Like, I agree that there is, it's very difficult to get right and you have
to get it exactly right. Otherwise, you wind up with Microsoft's Clippy. Okay, easy now.
What's your problem with Clippy? You like Clippy's Clippy, your friend?
Yeah, I was just, I just, I just talked to, we just had this argument. They said Microsoft CTO
and they said, he said he's not bringing Clippy back. They're not bringing Clippy back and that's
very disappointing. I think it was, Clippy was the greatest assistance we've ever built.
It was a horrible attempt, of course, but it's the best we've ever done because it was a real
attempt to have an actual personality. I mean, it was obviously technology was way
not there at the time of being able to be a recommender system for assisting you in anything
and typing in Word or any kind of other application, but still was an attempt of personality. That
was legitimate. That's true. I thought was brave. Yes, I'll get, yes. Okay. You know,
you've convinced me I'll be slightly less hard on Clippy. And I know I have like an army of people
behind me who also miss Clippy, so. Really? I want to meet these people. Who are these people?
It's the people who like to hate stuff when it's there and miss it when it's gone.
So everyone. It's everyone. Exactly. All right. So Anki and Gibo, the two companies,
two amazing companies, social robotics companies that have recently been closed down. Yes.
Why do you think it's so hard to create a personal robotics company? So making a business out of
essentially something that people would anthropomorphize, have a deep connection with,
why is it so hard to make it work? Is the business case not there or what is it?
I think it's a number of different things. I don't think it's going to be this way forever.
I think at this current point in time, it takes so much work to build something that only barely
meets people's minimal expectations because of science fiction and pop culture giving people
this idea that we should be further than we already are. When people think about a robot
assistant in the home, they think about Rosie from the Jetsons or something like that. And
Anki and Gibo did such a beautiful job with the design and getting that interaction just right.
But I think people just wanted more. They wanted more functionality. I think you're also right
that the business case isn't really there because there hasn't been a killer application
that's useful enough to get people to adopt the technology in great numbers.
I think what we did see from the people who did get Gibo is a lot of them became very
emotionally attached to it. But it's kind of like the palm pilot back in the day. Most people are
like, why do I need this? Why would I? They don't see how they would benefit from it until
they have it or some other company comes in and makes it a little better.
How far away are we? Do you think? How hard is this problem?
It's a good question. And I think it has a lot to do with people's expectations.
Those keep shifting depending on what science fiction that is popular.
But also, it's two things. It's people's expectation and people's need for an emotional
connection. I believe the need is pretty high. Yes. But I don't think we're aware of it.
That's right. I really think this is like the life as we know it. So we've just kind of gotten
used to it. I hate to be dark because I have close friends. But we've gotten used to really never
weren't being close to anyone. And we're deeply, I believe, okay, this is hypothesis,
I think we're deeply lonely, all of us, even those in deep fulfilling relationships.
In fact, what makes those relationships fulfilling, I think, is that they at least
tap into that deep loneliness a little bit. But I feel like there's more opportunity
to explore that, that doesn't interfere with the human relationships you have.
It expands more on the, yeah, the rich, deep, unexplored complexity that's all of us,
weird apes. Okay. I think you're right. Do you think it's possible to fall in love with a robot?
Oh, yeah, totally. Do you think it's possible to have a long-term committed
monogamous relationship with a robot? Well, yeah, there are lots of different types of
long-term committed monogamous relationships. I think monogamous implies, like,
you're not going to see other humans sexually or like, you basically on Facebook have to say,
I'm in a relationship with this person, this robot. I just don't, like, again, I think this
is comparing robots to humans. When I would rather compare them to pets, like you get a robot,
but it fulfills, you know, this loneliness that you have in a, maybe not the same way as a pet,
maybe in a different way that is even, you know, supplemental in a different way.
But, you know, I'm not saying that people won't, like, do this, be like, oh, I want to marry my
robot or I want to have like a, you know, sexual relation, monogamous relationship with my robot.
But I don't think that that's the main use case for them.
But you think that there's still a gap between human and pet.
So between husband and pet. There's a different relationship.
It's an engineering. So that's a gap that can be closed through.
I think it could be closed someday, but why would we close that? Like, I think it's so
boring to think about recreating things that we already have when we could create something that's
different. I know you're thinking about the people who like don't have a husband and like,
what could we give them? Yeah, but let's, I guess what I'm getting at is maybe not.
So like the movie her. Yeah. Right. So a better husband. Well, maybe better in some ways.
Like it's, I do think that robots are going to continue to be a different type of relationship,
even if we get them like very human looking or when, you know, the voice interactions we have
with them feel very like natural and human like, I think there's still going to be differences.
And there were in that movie too, like towards the end, it goes off the rails.
But it's just a movie. So your intuition is that, because you kind of said two things, right?
So one is, why would you want to basically replicate the husband? Yeah. Right. And the
other is kind of implying that it's kind of hard to do. So like anytime you try, you might build
something very impressive, but it'll be different. I guess my question is about human nature.
It's like, how hard is it to satisfy that role of the husband? So removing any of the sexual
stuff aside is the, it's more like the mystery, the tension, the dance of relationships.
Do you think with robots that's difficult to build? What's your intuition? I think that
well, it also depends on how we talk about robots now in 50 years in like indefinite amount of time
where I'm thinking like five or 10 years, five or 10 years, I think that robots at best will be
like a more similar to the relationship we have with our pets than relationship that we have
with other people. I got it. So what do you think it takes to build a system that exhibits great
and greater levels of intelligence? Like it impresses us with its intelligence, you know,
a Roomba. So you talked about anthropomorphization that doesn't, I think intelligence is not
required. In fact, intelligence probably gets in the way sometimes, like you mentioned. But
what do you think it takes to create a system where we sense that it has a human level
intelligence is something that, probably something conversational, human level,
intelligent. How hard do you think that problem is? It'd be interesting to sort of hear your
perspective, not just purely, I talked to a lot of people, how hard is the conversational agents?
Yeah. How hard is it to pass a touring test? But my sense is it's easier than just solving,
it's easier than solving the pure natural language processing problem,
because I feel like you can cheat. Yeah. Yeah. So how hard is it to pass a
touring test in your view? Well, I think, again, it's all about expectation management. If you
set up people's expectations to think that they're communicating with what was it a 13-year-old boy
from the Ukraine, then they're not going to expect perfect English. They're not going to
expect perfect understanding of concepts or even like being on the same wavelength in
terms of like conversation flow. So it's much easier to pass in that case.
Do you think, you kind of alluded this to with audio, do you think it needs to have a body?
I think that we definitely have, so we treat physical things with more social agency,
because we're very physical creatures. I think a body can be useful.
Does it get in the way? Is there a negative aspects like...
Yeah, there can be. So if you're trying to create a body that's too similar to something that people
are familiar with, like I have this robot cat at home that Hasbro makes. And it's very disturbing
to watch, because I'm constantly assuming that it's going to move like a real cat, and it doesn't,
because it's like a $100 piece of technology. So it's very disappointing, and it's very hard
to treat it like it's alive. So you can get a lot wrong with the body too, but you can also use
tricks same as the expectation management of the 13-year-old boy from the Ukraine. If you
pick an animal that people aren't intimately familiar with, like the baby dinosaur, like the
baby seal that people have never actually held in their arms, you can get away with much more,
because they don't have these preformed expectations. Yeah, I remember you were thinking
at a TED Talk or something that clicked for me that nobody actually knows what a dinosaur looks
like. So you can actually get away with a lot more. That was great. So what do you think about
consciousness and mortality being displayed in a robot? So not actually having consciousness,
but having these kind of human elements that are much more than just the interaction, much more
than just, like you mentioned, with a dinosaur moving kind of in interesting ways, but really
being worried about its own death and really acting as if it's aware and self-aware and identity.
Have you seen that done in robotics? What do you think about doing that? Is that a powerful good
thing? Well, I think it can be a design tool that you can use for different purposes. So I
can't say whether it's inherently good or bad, but I do think it can be a powerful tool. The fact
that the pleo mimics distress when you, quote unquote, hurt it is a really powerful tool to get
people to engage with it in a certain way. I had a research partner that I did some of the empathy
work with named Palashnandi and he had built a robot for himself that had a lifespan and that
would stop working after a certain amount of time just because he was interested in whether he
himself would treat it differently. And we know from Tamagotchi's, those little games that we used
to have that were extremely primitive, that people respond to this idea of mortality,
and you can get people to do a lot with little design tricks like that. Now,
whether it's a good thing depends on what you're trying to get them to do.
Have a deeper relationship. Have a deeper connection, sign a relationship.
If it's for their own benefit, that sounds great. Okay. You could do that for a lot of other reasons.
I see. So what kind of stuff are you worried about? So is it mostly about manipulation of your
emotions for like advertisements and so on, things like that? Yeah, or data collection or, I mean,
you could think of governments misusing this to extract information from people. It's,
you know, just like any other technological tool just raises a lot of questions.
What's, if you look at Facebook, if you look at Twitter and social networks,
there's a lot of concern of data collection now. What's from legal perspective or in general,
how do we prevent the violation of sort of these companies crossing a line? It's a great area,
but crossing a line, they shouldn't in terms of manipulating, like we're talking about manipulating
our emotion, manipulating our behavior using tactics that are not so savory.
Yeah, it's really difficult because we are starting to create technology that relies on data
collection to provide functionality. And there's not a lot of incentive even on the consumer side
to curb that because the other problem is that the harms aren't tangible. They're not really
apparent to a lot of people because they kind of trickle down on a societal level and then
suddenly we're living in like 1984, which sounds extreme, but that book was very prescient. And
I'm not worried about these systems. I have Amazon's Echo at home and Telelexa
all sorts of stuff and it helps me because Alexa knows what brand of diaper we use and so I can
just easily order it again. So I don't have any incentive to ask a lawmaker to curb that. But
when I think about that data then being used against low-income people to target them for
scammy loans or education programs, that's then a societal effect that I think is very severe.
And legislators should be thinking about. But yeah, the gray area is
the removing ourselves from consideration of explicitly defining objectives and more saying,
well, we want to maximize engagement in our social network.
Because you're not actually doing a bad thing, it makes sense. You want people to
to keep a conversation going, to have more conversations, to keep coming back
again and again to have conversations. And whatever happens after that, you're kind of
not exactly directly responsible. You're only indirectly responsible. So I think it's a really
hard problem. Are you optimistic about us ever being able to solve it?
You mean the problem of capitalism? Because the problem is that the companies
are acting in the company's interests and not in people's interests. And when those interests
are aligned, that's great. But the completely free market doesn't seem to work because of
this information asymmetry. So say you were trying to do the right thing. I guess what
I'm trying to say is it's not obvious for these companies what the good thing for society is
to do. Like, I don't think they sit there with, I don't know, with a glass of wine and a cat,
like petting a cat, evil cat. And there's two decisions. And one of them is good for society.
One is good for the for the profit. And they choose the profit. I think they actually,
there's a lot of money to be made by doing the right thing for society. Like that,
because Google, Facebook have so much cash that they actually was, especially Facebook,
was significantly benefit for making decisions that are good for society. It's good for their
brand, right? So but I don't know if they know what's good for society. That's the,
I don't think we know what's good for society in terms of how, yeah, how we manage the conversation
on Twitter, or how we design, we're talking about robots, like, should we emotionally manipulate
you into having a deep connection with Alexa or not? Yeah. Yeah. Do you have optimism that
we'll be able to solve some of these questions? Well, I'm going to say something that's
controversial, like in my circles, which is that I don't think that companies who are reaching out
to ethicists and trying to create interdisciplinary ethics boards, I don't think that that's totally
just trying to whitewash the problem and, and so that they look like they've done something.
I think that a lot of companies actually do, like you say, care about what the right answer is,
they don't know what that is, and they're trying to find people to help them find them. Not in
every case, but I think I, you know, it's much too easy to just vilify the companies as like you
said, sitting there with their cat going, one million dollars. That's not what happens. A lot
of people are well meaning even within companies. I think that what we do absolutely need is more
interdisciplinarity, both within companies, but also within the policymaking space, because we're,
you know, we've hurtled into the world where technological progress is much faster. It seems
much faster than it was and things are getting very complex. And you need people who understand the
technology, but also people who understand what the societal implications are, and people who are
thinking about this in a more systematic way to be talking to each other. There's no other solution,
I think. You've also done work on intellectual property. So if you look at the algorithms that
these companies are using, like YouTube, Twitter, Facebook, so on, I mean, that's kind of,
those are mostly secretive. The, the recommender systems behind, behind these algorithms. Do you,
do you think about an IP and the transparency about algorithms like this? Like what is the
responsibility of these companies to open source the algorithms or at least reveal to the public
what's how these algorithms work? So I personally don't work on that. There are a lot of people
who do though, and there are a lot of people calling for transparency. In fact, Europe's even
trying to legislate transparency, maybe they even have at this point, where like if, if an algorithmic
system makes some sort of decision that affects someone's life, that you need to be able to see
how that decision was made. I, you know, it's, it's a, it's a tricky balance because obviously
companies need to have, you know, some sort of competitive advantage and you can't take all
of that away or you stifle innovation. But yeah, for some of the ways that these systems are already
being used, I think it, it is pretty important that people understand how they work.
What are your thoughts in general on intellectual property in this weird age of software, AI,
robotics? Oh, that it's broken. I mean, the system is just broken. So can you describe,
I actually, I don't even know what intellectual property is in the space of software, what it
means to, I mean, I think, so I believe I have a patent on a piece of software from my PhD.
You believe? You don't know? No, we went through a whole process. Yeah, I do.
You get the spam emails like, we'll frame your patent for you.
Yeah, it's much like a thesis. So,
but that's useless, right? Or not? Where does IP stand in this age? What, what is,
what's the right way to do it? What's the right way to protect and own ideas in
this, when, when it's just code and, and this mishmash of something that feels much softer than
a piece of machinery. Yeah, I mean, it's hard because, you know, there are different types
of intellectual property and they're kind of these blunt instruments. They're, they're like,
it's like patent law is like a wrench, like it works really well for an industry like the
pharmaceutical industry. But when you try and apply it to something else, it's like,
I don't know, I'll just like hit this thing with a wrench and hope it works.
So software, you know, software, you have a couple of different options.
Software, like any code that's written down in some tangible form is automatically copyrighted.
So you have that protection, but that doesn't do much because if someone takes the basic
idea that the code is executing and just does it in a slightly different way,
they can get around the copyright. So that's not a lot of protection. Then you can patent
software, but that's kind of, I mean, getting a patent costs, I don't know if you remember
what yours cost or like, was it through an institution? Yeah, just through a university.
That's why they, it was insane. There were so many lawyers, so many meetings. It made me feel
like it must have been hundreds of thousands of dollars. It must have been something crazy.
It's, it's insane the cost of getting a patent. And so this idea of like protecting the like
inventor in their own garage, like came up with a great idea is kind of, that's the thing of
the past. It's all just companies trying to protect things and it costs a lot of money.
And then with code, it's oftentimes like, you know, by the time the patent is issued,
which can take like five years, you know, probably your code is obsolete at that point.
So it's, it's a very, again, a very blunt instrument that doesn't work well for that
industry. And so, you know, at this point, we should really have something better, but we don't.
You like open source? Yeah, it's open source, good for society. You think all of us should open
source code? Well, so at the Media Lab at MIT, we have an open source default because what we've
noticed is that people will come in, they'll like, write some code and they'll be like,
how do I protect this? And we're like, like, that's not your problem right now. Your problem
isn't that someone's going to steal your project. Your problem is getting people to use it at all
right. Like, there's so much stuff out there. Like, we don't even know if you're going to get
traction for your work. And so open sourcing can sometimes help, you know, get people's work
out there, but ensure that they get attribution for it for the work that they've done. So like,
I'm a fan of it in a lot of contexts. Obviously, it's not like a one size fits all solution.
So what I gleaned from your Twitter is your mom. I saw a quote, a reference to babybot.
What have you learned about robotics and AI from raising a human baby bot?
Well, I think that my child has just made it more apparent to me that the systems we're currently
creating aren't like human intelligence. There's not a lot to compare there. He has learned and
developed in such a different way than a lot of the AI systems we're creating that that's not really
interesting to me to compare. But what is interesting to me is how these systems are going to shape
the world that he grows up in. And so I'm even more concerned about kind of the societal effects
of developing systems that, you know, rely on massive amounts of data collection, for example.
So is he going to be allowed to use like Facebook or Facebook is over? Kids don't use that
anymore. Snapchat. What do they use Instagram over to? I don't know. I just heard that TikTok is over,
which I've never even seen. So I don't know. No, we're old. We don't know. Twitter and you just
I'm going to start gaming and streaming my my gameplay. So what do you see is the future of
personal robotics, social robotics, interaction with our robots? Like, what are you excited about
if you were to sort of philosophize about what might happen the next five, 10 years?
That would be cool to see. Oh, I really hope that we get kind of a home robot that makes it. That's
a social robot and not just Alexa. Like, it's, you know, I really love the Anki products. I thought
Jibo was had some really great aspects. So I'm hoping that a company cracks that. Me too.
So, Kate, it was a wonderful talking to you today. Likewise. Thank you so much. It's fun.
Thanks for listening to this conversation with Kate Darling. And thank you to our sponsors,
ExpressVPN and Masterclass. Please consider supporting the podcast by signing up to Masterclass
and masterclass.com slash Lex and getting ExpressVPN at expressvpn.com slash Lex pod.
If you enjoy this podcast, subscribe on YouTube, review it with five stars on Apple podcast,
support on Patreon or simply connect with me on Twitter and Lex Friedman.
And now let me leave you with some tweets from Kate Darling. First tweet is,
the pandemic has fundamentally changed who I am. I now drink the leftover milk in the bottom
of the cereal bowl. Second tweet is, I came on here to complain that I had a really bad day
and saw that a bunch of you are hurting too. Love to everyone. Thank you for listening.
I hope to see you next time.