How the people who build robots answer the fear of robots

The Short Version

The people who build robots mostly answer the fear of robots by describing the machines. Russ Tedrake of MIT and the Toyota Research Institute says the fear is "a bit cultural," and that people who see an animal-like robot are "assuming that they have a level of intelligence that they don't yet have" [1]. Robert Playter, who runs Boston Dynamics, says people have been taught by a century of fiction to expect robots to take over, and that a physical machine is something "we can judge" for safety in a way we cannot yet judge software [2]. Michael Littman, a reinforcement learning researcher, is "not particularly moved" by the idea that we will accidentally build a superintelligence that destroys us, because building such systems keeps people in the loop [3]. Kate Darling offers a different picture to think with: robots as animals, partners with other skills, not replacements for people [4]. None of them says there is nothing to worry about. They point the worry elsewhere, at misuse, at software that spreads instantly, at data and at the disruption of work [1][2][3][4].

Table of contents

What They Said

Russ Tedrake: the fear is cultural, and the robots are not what they look like

Lex Fridman puts the fear to Tedrake in its television form: the Black Mirror episode in which a robot dog like Boston Dynamics' Spot goes rogue. Tedrake recalls showing a Spot video to MIT faculty and watching horror cross the room [1].

His first answer is about where the reaction comes from. "I think the fear is a bit cultural," he says. People of his age in the United States grew up watching Terminator. "If I had grown up at the same time in Japan, I probably would have been watching Astro Boy." There is a very different reaction to robots in different countries [1].

His second is about what the machines can do. "I don't think robots are going to come after me with a kitchen knife" right away, he says, and he used to joke about the Atlas humanoid that "all I had to do was run for five minutes" and its battery would be flat, though he notes that later versions ran for over an hour [1]. The gap between appearance and ability is the heart of it. People attribute to a robot that looks like an animal a level of self-awareness or consciousness it lacks, and "that might be part of the fear" [1].

The rest of his conversation shows how wide that gap is. At the DARPA Robotics Challenge, his MIT team's Atlas fell on its face getting out of a vehicle because someone forgot to tick one box on a checklist, leaving a controller running that moved the robot's foot at the worst moment. Nothing in the software knew what to do when the robot was sitting with its feet in the air [1]. Even with a robot running constantly, new failures appeared days before the competition [1]. Touching the world remains one of the field's hardest problems: most robots make contact only with their feet or fingertips, and he says "we mostly in robotics are afraid of contact on the rest of our body, which is crazy" [1]. Planning quickly through rich contact, as a wrestler does, is still beyond the field in his account [1]. The engineers' own fear runs the opposite way from the public's. They are trying to get the robot to stop avoiding the world.

Tedrake does not dismiss concern. "I think we have to be careful about the way robots influence our society," he says, and he calls that a responsibility roboticists need to embrace [1]. He accepts that technology can be used in many ways. Drones are one example, and he thinks "there are purely software attacks that somebody could use to do great damage" [1]. He does not want to build technology without thinking about it. But "I would consider myself a technological optimist": we will create new challenges, "we'll screw something up, maybe," and "we'll invent ourselves out of those challenges and life will go on" [1]. He adds that "there are many scary things in the world," and that people are right to ask the questions [1].

On the long-term, existential version of the fear, he borrows an answer from the roboticist Rodney Brooks: "It's not going to be the robots versus the people. We're all going to be robot people," given that we already carry smartphones and that some people have hearing aids, pacemakers or prosthetic limbs [1]. He sees the larger change in how online interaction is reshaping human relationships [1].

He has an observation about why robots stay frightening as a category. "Robotics always has to solve the next problem and doesn't get to celebrate its past successes," he says; a dishwasher stops being called a robot once it works [1]. What remains under the name is whatever is still strange.

Robert Playter: a hundred years of being taught to be afraid

Playter comes to the question with a prop. He shows Fridman a Wall Street Journal article about ChatGPT illustrated with a menacing humanoid robot announcing that it will replace you [2]. "It's unfortunate that they chose to use a robot as that embodiment," he says, because the two things are very different [2].

His explanation of the fear is historical. The word robot was coined by the Czech playwright Karel Čapek in 1921, in a play in which the robots take over at the end, and audiences have been entertained by that plot ever since [2]. "People love to be afraid," he says, and they have been taught for a hundred years that takeover is the logical progression. "I think it's fiction" [2].

The second part of his answer is about why bodies are easier to trust than code. The point is made in his exchange with Fridman that even a very capable AI in a robot body is less threatening because "It's visible. It's verifiable. It's right there in physical reality," and humans know how to deal with physical reality [2]. What we lack experience with is "software that can be propagated worldwide in an instant," and for that new protocols and tools are needed [2]. That work, Playter says, should not hold robots back, because "I think we can judge when a robot's being safe" [2]. He is uneasy about large language models spreading disinformation and not about learning in robots, and Fridman supplies the reason: "The physical reality is a good verifier" [2].

Like Tedrake, he says people misread the machines. They identify with Boston Dynamics' robots because the only things they have seen move that way were alive, which he calls both a benefit and a curse, since "they read into the robot's intelligence and emotion that isn't there" [2]. The company's history supports him. Natural walking on Atlas took about 15 years to arrive [2]. At the DARPA challenge, he says, the robots mostly failed and fell down, and he recalls one that reached for a door handle, missed its grip, and threw itself over because it expected the door to push back [2]. Around 40 developers share four modern Atlas robots [2]. The business is inspection of breweries, chip plants and power stations with Spot, and unloading trucks with Stretch [2].

On jobs he gives the industry's standard answer with specifics. Every technological shift has brought "automation anxiety" [2]. One day, he says, people will look back at workers unloading boxes by hand from trailers that can reach 120 degrees and wonder why it was ever done that way [2]. The same people can operate the machines, and at his customers the workers who unloaded trucks are now running the robots [2]. "We used to farm with hand tools and now we farm with machines. And nobody has really regretted that transformation" [2]. He adds that shrinking populations in the developed world make such jobs hard to fill, and that "People want to work" [2].

He takes one fear seriously enough to act on it. After videos appeared of people strapping remotely operated guns to commercial robots, Boston Dynamics wrote a letter, co-signed by several other robotics companies, pledging not to put weapons on their robots [2]. Trust, he says, is essential for the industry to grow, and he has hired a policy director to work with lawmakers and regulators [2].

On machine consciousness he is doubtful. He describes large language models as "a very elaborate word completion" and does not think that has much to do with consciousness [2]. Fridman disagrees that the question can be put aside so easily, arguing such systems may simulate consciousness convincingly [2].

Michael Littman: we will be in the loop

Littman addresses the largest fear, that AI will end humanity. He says he tries to see every question from all sides, but on this one he is fairly settled: "I am not particularly moved by the idea that if we're not careful, we will accidentally create a super intelligence that will destroy human life" [3].

He knows the argument he is rejecting. As he summarizes Nick Bostrom's book, computers help design smarter versions of themselves until they bootstrap past us and pursue goals out of step with ours [3]. He once wrote an op-ed answering Elon Musk's warnings. His reading of Musk is that his success comes from believing in the power of ideas, and that such a person naturally pushes an idea to its extreme: "So I think his strength is also his undoing because that doesn't mean it's true" [3].

His main objection is about process. Building systems that can really act in the world will teach us a great deal about how they work, so we will be shaping them as we go [3]. He borrows the creationist image of random parts assembling a 747. Airliners do get assembled, he says, but by a process we understand and control [3].

Several of his other views explain why he doubts a sudden runaway. Breakthroughs depend on rare human skill: for years his students could not reproduce the backgammon program TD-Gammon's success with neural networks, and he came to regard its author as "a neural net whisperer" [3]. Progress is narrow: of the claim that AlphaZero's ceiling has not been found, he says "it's not getting better at other things it's getting better at go" [3]. Every finite game has an optimal strategy, so improvement runs into diminishing returns [3]. He suspects the growth in computing is "a sigmoid curve on the front half which looks a lot like an exponential" [3]. And language models trained on our writing miss something, in his view: "I think you really need to be pushed back at" by people to learn nuance [3].

His children's driving lessons gave him a concrete case. What stunned him was "the degree to which driving is a social interaction activity," which requires thinking about what other drivers are thinking, something he had not heard discussed in talks about self-driving cars [3]. Machines, he notes, also get credit they have not earned. We project intelligence and compassion onto devices like his Roombas, and "One of the things we're learning from AI is where we are smart and where we are not smart" [3].

Fridman pushes back. He cites Sam Harris and Bostrom's point that nothing in physics forbids superintelligence eventually, and notes that even an emergence over 20 years is fast in historical terms [3]. His own bigger worry is that we would not notice it happening, since simple social media algorithms may already be steering collective human behaviour [3]. Littman agrees that algorithms optimizing for advertisers already manipulate people, and says where his alarm would begin: if they tried to stop us from shutting them off [3]. He adds that he trusts Wikipedia's values more than those of Facebook or YouTube [3].

His favourite robot film shows what he wants instead. In Robot & Frank an ageing jewel thief pulls his helper robot into his own odd life, and Littman loves the idea that people should be able to make technology their own [3].

Kate Darling: think animals, not people

Darling, who studies human-robot interaction at the MIT Media Lab, thinks much of the fear starts from a bad comparison. Type "robot" into a search engine and you get humanoids, and the constant measuring of robots against people and of AI against human intelligence trips us up [4].

Her alternative is the subject of her book. We domesticated animals because what they do is different from what we do and useful. The ox changed what people could do economically; ferrets still run wires through narrow pipes; navies still use dolphins for mine detection because their echolocation beats machines [4]. Robots, she says, will reshape economies and even cities in the same way, and society will adjust socially as it did when animals became pets [4]. "It's not a perfect one," she says of the analogy, but it helps us get away from the idea that robots will replace people [4].

She applies it to work. "I do think that robots and humans have very different skill sets," she says, so while a robot may take over a task, "it's not going to take over most jobs" [4]. Her example is mining, where autonomous trucks let workers sit in air-conditioned control rooms instead of on dangerous sites. That is a much better job, and fewer people are employed [4]. From a bird's-eye view more jobs are created, but she warns against ignoring "the rubble on the ground" during the transition [4].

She does not treat the public as foolish. People will believe machines are sentient, and so do experts. In her lab someone can be clinical about a robot they built and then coo over a different one "having fun." "We are all this guy," she says of the Google engineer who claimed a chatbot was sentient [4]. The goal is not to discourage the tendency, which she thinks is impossible, but to acknowledge it, even embrace it, and design for it [4].

She also shows where fear is earned. The grocery store robot Marty drew complaints less about jobs than about being a creepy surveillance machine that got in the way [4]. When robot dogs were deployed with police, she says, the reaction was "oh my God, the cops are using a robot dog," and the maker had a public relations nightmare [4]. Fridman adds an awkward observation about incentives: for companies whose money comes from industrial customers, "it's almost good PR for people to be scared of these things," because fear spreads videos [4]. Darling's own strongest worries are about privacy, data and manipulation by systems that people bond with [4].

On the humanoid form she parts company with much of the industry. She calls it a distraction for social robotics, notes that people love the trash-can-shaped R2-D2 more than the humanoid C-3PO, and would rather see environments redesigned to suit many kinds of bodies [4]. And she ends on the opposite of fear. "I think love is so important and I also think it's not zero sum," she says. "Artificial agents will just if done right will just expand your capacity for love" [4].

Where They Agree, and Where They Don't

The machines are less capable than they look. Tedrake, Playter, Littman and Darling all say people project intelligence, emotion or awareness onto robots that do not have them [1][2][3][4]. The builders' stories of robots falling over, missing door handles and taking 15 years to walk naturally are the evidence [1][2].

The fear was learned. Tedrake blames the films of his American childhood and Playter a century of fiction beginning with Čapek [1][2]. Darling adds the habit of comparing robots with humans [4].

The serious risks are somewhere else. Tedrake names misuse, drones and software attacks [1]. Playter names software and language models that spread instantly, and weapons mounted on robots [2]. Littman and Fridman name social media algorithms [3]. Darling names surveillance, data and manipulation [4]. In every case the feared thing is people using technology on other people, or software, and not a machine deciding to turn on its makers.

They differ on how much weight the fear deserves. Playter says plainly "I think it's fiction" [2]. Tedrake says people are right to ask and that roboticists must think about the implications of their work [1]. Darling treats specific fears, of being watched and of police use, as reasonable responses to real deployments [4]. The difference tracks their positions. Playter leads a company whose growth depends on public trust. Tedrake is a researcher. Darling studies how people actually react.

They differ on jobs. Playter stresses upskilling and says nobody regrets the mechanization of farming [2]. Darling agrees about the long run and insists on counting the workers displaced along the way [4].

They differ on anthropomorphism. Playter calls it a benefit and a curse [2]. Tedrake sees it as a source of fear [1]. Darling wants it accepted and designed for [4].

They differ on the humanoid body. Tedrake says the best argument for humanoids is that the world is built for the human form, though that may fade [1]. Playter builds one [2]. Darling thinks the form feeds the wrong comparison and wastes attention [4].

They differ on the long term. Littman's doubt rests on people staying in the loop [3]. Fridman's reply is that nothing rules the outcome out and that the change may be too gradual to notice [3]. Tedrake's own phrasing keeps the question open: the robots lack intelligence "yet" [1]. Playter expects the next generation of Atlas to do things a person cannot [2].

Why It Works

Familiarity with failure. People who have watched a humanoid fall out of a car, or a fleet of robots walk 2,000 kilometers a week just to find rare faults, have a calibrated sense of the distance between a video and a dependable machine [1][2].

Physical things can be inspected. A robot occupies one place, can be seen and tested, and comes out of a long tradition of certifying machines as safe. Software copies itself everywhere at once [2].

Culture explains the reaction. The same robot horrifies one audience and charms another, depending on whether it was raised on Terminator or Astro Boy [1].

Movement triggers projection. Darling says much of the magic of both animals and robots is our own projection, amplified by autonomous movement [4]. Playter notes his robots move the way only living things had moved before [2]. That is why a machine with the awareness of an appliance can feel like a predator.

Building teaches. Littman's 747 argument is that nothing this complex is assembled by accident; it is put together through a process people learn from at each step [3].

Progress is narrow and bounded. Systems get better at the task they are trained on, finite games have ceilings, and what looks exponential may be the first half of an S-curve [3].

Intelligence in the world is social. Driving turned out to require reading other minds, and Littman thinks machines will need to be pushed back at by people to learn nuance [3].

History offers a precedent. Humans have lived and worked with capable non-human agents for thousands of years, and Darling's analogy draws on that record [4].

These are the views of practitioners arguing from experience. Darling and Littman cite research in their fields, and the others cite their own laboratories and companies. None presents a study showing the fears to be unfounded, and Littman admits a past misjudgment: he predicted self-driving cars would be introduced gradually in closed communities, and they were instead largely dropped onto public roads [3].

Where It Applies, and Where It Doesn't

The reassurance covers physical robots as they are now. It rests on current limits in balance, contact, batteries and perception [1][2]. Both builders speak of abilities the machines do not have "yet" [1][2].

It does not cover software. The builders themselves move the concern there, and Fridman's argument about social media stands largely unanswered [2][3].

It does not cover what people do with robots. Guns have already been strapped to commercial machines, and the industry's pledge against weapons is voluntary [2]. Tedrake mentions drones [1].

It does not cover surveillance. Spot's main business is inspection with cameras and sensors, and Darling's examples show the public objecting to exactly that [2][4].

It does not remove the cost of change at work. Even on the optimistic account some people lose jobs they cannot easily replace [4].

The speakers have a stake. Playter needs to sell thousands of robots, and says so [2]. That does not make him wrong, but Darling's and Fridman's remark that fear can be good publicity for robot makers is a reminder that incentives run in more than one direction [4].

The animal analogy has limits. Darling says it is imperfect and that language-using systems make it trickier [4].

What To Do With It

1. Ask what the machine can actually do. Before reacting to a video, find out how long it runs, what happens when it is pushed, and how often it fails. Tedrake's and Playter's stories suggest the answers are usually humbling [1][2].

2. Notice where your picture of robots came from. Tedrake's contrast between Terminator and Astro Boy is a useful check on any instinctive reaction [1].

3. Separate the body from the software. A robot can be seen and tested. Put more of your concern into systems that spread instantly and act out of sight, as Playter advises [2].

4. Watch the people, not the machine. Ask who is deploying a robot and for what: inspection, policing, selling, or weapons [2][4].

5. Expect yourself to anthropomorphize. Darling says everyone does. Enjoy it, and do not mistake it for evidence about the machine's mind [4].

6. Use the animal comparison. Ask what a robot can do that people cannot, the way an ox or a ferret could, instead of asking when it will match a person [4].

7. Judge automation at work by both views. Look for the better job it creates and for who is left out, and press for the retraining Playter describes [2][4].

8. Hold makers to their commitments. The weapons pledge and Playter's hiring of a policy director are things the public can check and regulators can build on [2].

9. Keep one clear line for alarm. Littman's is a system that resists being switched off [3].

10. If you build these machines, accept the responsibility Tedrake names. Think about how they affect society, be honest about what they cannot do, and remember Darling's examples of what happens when the public's reaction is treated as an afterthought [1][4].

Sources

1. Russ Tedrake on Underactuated Robotics, Contact, Touch and the DARPA Robotics Challenge · Lex Fridman Podcast
2. Robert Playter on Boston Dynamics: Atlas, Spot, Stretch and the Business of Legged Robots · Lex Fridman Podcast
3. Michael Littman on Reinforcement Learning, AI Risk Skepticism, and the Future of AI · Lex Fridman Podcast
4. Kate Darling on Social Robots, the Animal Analogy, Privacy, Manipulation and Lessons from MIT's Epstein Scandal · Lex Fridman Podcast