"ChatGPT Moment" for Robotics Is Coming. The Real Problem Isn't Intelligence | Stanford, Catie Cuan
In a Nutshell
Katie Cuan argues that robotics' core unsolved problem isn't building more intelligent machines but designing legible, emotionally engaging human-robot interactions. She critiques the field's narrow focus on humanoids optimized for "dirty, dull, dangerous" tasks, advocating instead for diverse robot forms that can inspire, delight, and provide companionship. Her VLI model and projects like Music Mode demonstrate that measuring success by human emotional response, rather than just task completion, is essential for robots that billions of people will encounter in everyday spaces.
These notes were generated by AI and may contain inaccuracies.
It is a tremendous time to be working in robotics. Every week or multiple times a week there are new announcements about amazing things happening. We are in this absolutely unbelievable moment right now. The thing missing in a lot of this discussion is what the human interaction piece looks like.
The question in terms of building humanoids is whether this is really what we want. The speaker asks whether this is what a 77-year-old English as a second language dad wants, even if the robot can do some tasks. The field has locked itself into a slightly myopic vision of what the future with robots can look like, and the speaker would love for that aperture to be broadened.
The speaker's name is Katie Cuan. She is the founder and CEO of Art Lab, which stands for AI Robot Technology. She is a robot choreographer, roboticist, researcher, entrepreneur, and artist. Her mission is to make nascent technologies like AI and robots more humanistic, whimsical, and widely accessible through business, research, and art.
Her definition of humanistic is human-centered. She focuses on driving interactions and experiences with machines that help humans feel more empowered and safe while also helping them live healthier, better lives.
The location is the Stanford Robotics Center on the Stanford campus in Palo Alto, California. The idea was to bring together a variety of disciplines and engineering backgrounds to work on flagship projects in robotics including domestic robotics, industrial robotics, medical, underwater, and exploration. For the speaker, the focus areas are education, art, culture, movement, and dance.
There will be billions of robots in our lifetimes. It only took 13 years for there to be a billion iPhones on Earth. With so many people working in robotics now, billions upon billions of robots will show up in everyday spaces: offices, hospitals, hotels, homes, and caretaking facilities.
When robots are taken out of niche, narrow vertical applications and given more intelligence while placed in situ with people, there is a huge opportunity and challenge: how to make humans and robots legible to one another. This is what the speaker's work is interested in. Human interaction is a grand challenge for robots and has been for the last several decades.
Humans are really unpredictable. They tend to do and say things and move in ways that might not always be expected. The problems of interest are how to make robots more legible to people, but also more animating and more inspiring to delight and bring awe into lives. The speaker has been curious throughout her career about what kinds of interactions can be facilitated that both inspire and surprise people while also opening the door into what technology can be. Technology is not exclusively a tool but also something one could socialize with or learn from.
A model called VLI, a vision language interaction model, is being built. This is the first kind of model that takes the impetus of what it should be doing from humans in the environment. LLMs are acting on human information and then map that to a series of robot actions, but measure the success or failure of those robot actions on the basis of a human response. The system can tell whether the human is increasing in positive affect or decreasing, and whether that was what they wanted or not. This is the first start for robots that can really naturally and intuitively socialize and interact with people. The goal is for those interactions to be safe, legible, and clear.
The speaker grew up in the Bay Area. From a young age she always loved numbers and loved to dance. She loved math and science too. Her dad is Cuban, so the whole family grew up social dancing. The statement is made that if you don't know how to dance, you're probably not Cuban. She grew up with these two very strong interests.
Toward the end of college, she had interned at a big technology company in the Bay Area. She knew she really wanted to try what it was like to be a professional dancer, and there is only a very small window in early adulthood when one can go for it. She was a little scared because she did not know anyone who had been a professional dancer in her immediate sphere of friends and family.
After graduating from college, she took a job doing private equity consulting in New York. She realized quickly that was not for her, mainly because she had this childhood dream. She figured she was only going to be 21 once, and if she did not go for this now, she would regret it for the rest of her life. She left that job and became a professional dancer in New York City. She had her own dance company for a long time. She danced for the Metropolitan Opera Ballet at Lincoln Center in New York, which was a crazy experience. She will never forget what it is like to bow on that stage.
Around that time, her dad also got sick and spent a while in the hospital. Seeing him be sick in the hospital was really scary because here was somebody who she cared about and loved so much, and she wanted him to be healthy and safe. The fact that he was surrounded by all these big machines that were supporting him, meant to help him literally survive, and they were beeping and booping with proportions and colors and an overall presence that was oppressive to him was notable. It was also opaque. He did not understand what they were doing and there were very few opportunities where someone could explain it to him. Seeing him be that vulnerable led to the realization that a lot of other people are in that same position all the time.
One of the projects worked on at Google was called music mode. When first arriving, there were 200 robots driving around Google buildings doing useful work: wiping tables, sorting trash, resetting conference rooms. They were doing valuable things. As an artist in residence, the speaker went around and asked software engineers what they thought about the robot. They responded that it was fine, it was a robot. It was kind of cute or kind of cool initially and now it very slowly does these different tasks. This was viewed as a missed opportunity. Here was this amazing new flagship vanguard technology showing up in lives and people felt a little bit ambivalent toward it. For some people that would be a success, but the view was that a lot more could be done.
The first idea was to get the robots to play music. The thought was that they could sort some trash and wipe some tables and then go over and play the cello. Work continued for a few weeks. Then a friend and colleague named Tom came up and said that the robot generates a lot of data. Instead of the robot playing music, what if the robot was the music? What if this data became the music? That was a clever breakthrough. Work then started with a composer named Peter van Straten on a piece of software called music mode. Every time a different part of the robot moved, it would be mapped to a different sample of music. For example, opening and closing the gripper goes cling cling cling. Turning the torso of the robot would make a beautiful bass noise like boom.
Peter was an amazing artist to work with because he understood that all these joints of the robot can be moved in any combination. It was a real combinatorics problem. Work continued for a couple months and then the software called music mode was shipped on the robot. One could just say to the robot, "Hey robot, turn on music mode." Then as it was wiping the table, a symphony would be heard. Emails started coming from people at Google who were not previously known, saying this is the most incredible thing ever seen. People were sitting at their desks crying because they never thought a robot could be so beautiful or that a machine could make them feel that way. Music mode was shipped on every single one of the robots. Nine robots could be sorting trash at the same time and one could go around and say "Hey robot, turn on music mode" and a beautiful symphony would arise.
This was a proof point that these experiences with technologies are needed. People want to feel something. People want to feel engaged. We have been so influenced by fiction and media. There are dozens of humanoid companies making robots that look like humans. Often the argument is that they are made for human spaces, so they need to look like people. Rodney Brooks, a titan in the field, says the closer something looks to a person, the higher the expectations people have for it. One cannot help it. Mirror neurons go off. Expectations start that because it is bilaterally symmetric and roughly in the same proportions as a human, it should be able to do what a human does. The expectations set are so high.
If asked whether it is the 77-year-old dad or one of the nieces and nephews, "Hey, is that the robot that you want?", there is not enough expansive thinking into what robots can be. The broader robotics field has locked itself into a slightly myopic vision of what the future with robots can look like, and the aperture should be broadened.
There is a phrase called dirty, dull, and dangerous. For a long time robots were useful for tasks that were too dirty, dull, or dangerous for people to do. Then there was appetite for robots to do other things that were not exclusively dirty, dull, and dangerous jobs. The field has still gotten locked into the paradigm that robots are for utility. They need to alleviate certain parts of physical labor in lives when robots can do a lot more than that.
The Paro robot was a very soft seal robot popular in the mid-2010s that would just provide companionship for people who were ailing or lonely. There have been robots used for helping children with autism learn social skills. There are a lot of opportunities for robots to show up in lives and be beneficial for humans that are not exclusively humanoids that do dishes and laundry.
A lot of different robots have been worked with throughout the career: Roombas in a dance project, small humanoid robots, drones, big industrial robot arms, mobile manipulators from Google, and quadrupeds from Boston Dynamics.
The very first experience dancing with a robot in Amy's lab in 2017 was transcendent. It was the first time getting to imbue oneself inside of something and then also interact with it. Choreography was created for the machine, then it performed the choreography, and then dancing with it happened. The feeling was "Whoa." There was a feeling of being on a very long continuum of humans and tools. Everything from early stone tools and fire to the wheel and the printing press and the modern computer and these robots. The thought was that this is part of definitionally what it means to be a human: to be in relationship with tools.
Aristotle used to say we should not be called Homo sapiens, wise man or intelligent man. There are a lot of species in the world that are intelligent. We should be called Homo technae. The thing that makes us unique is our relationship to our tools and our ability to make tools.
What has been learned since then includes so much frustration, so much debugging, and so many limitations. The relationship has changed to become much more acutely aware of how amazing human beings are relative to all of the robots worked with. Robots do not have the expressive potential that people have. Humans need to give themselves a lot of credit. We can do a wide diversity of things.
If thinking about what was required to come into the building today, a door had to be opened. It was a door never seen before with a handle that could not have been predicted. There are infinitely many different kinds of door handles with different friction and torque required to get the door open. Stairs never seen before had to be walked down in a space with proportions. A large object had to be moved around, whether a backpack or something else being carried. Socializing and showing up and making conversation was needed, being acutely aware of timing. The dexterity and acuity and specificity that humans have took not only a lifetime to evolve, but also the many generations before that encoded all that stuff in DNA. We are really remarkable animals.
A beautiful article was read about writing as a craft and how it is a hard-earned craft in many cases. Anything worth doing, anything that is a craft in the richest sense of the word, takes effort. That feeling applies to many different things, not only writing. Anything really worth doing or worth earning is hard. Otherwise everybody would do it. Otherwise it would not be considered worthy.
A reminder comes from a professor at Stanford named Mark Skylar Scott, who said his goal in research is to 3D print a human heart, and it is going to take 25 years to do it. But it is a challenge worthy of life's attention because the impact at the end is so life-altering, literally, that it will have been worthy of life's attention. Things that are still challenging are needed.
Part of what was learned in dance training is that to really be at the top, to be at the highest levels, hard work is needed. It is not easy. Otherwise everyone else would do it. When showing up to an audition and there are 400 people there and they are going to hire four people, one better be good. Otherwise everybody else in that room is good too, and one should go home and get better.
A thing certainly not seen as delegating in the future is the pursuit of high-impact, challenging problems with solutions built with people who inspire.
A class is taught at Stanford called CS 334, Robots and Arts. It is the first class cross-listed between computer science and theater arts and performance studies. The invitation was that AI and robots could be used to make a creative statement with a political, philosophical, or social underpinning. One could not just have a robot do something. An assertion had to be made with that project about believing the world should be like this, or something that inspires about the world, or something not liked that should be changed.
On the very first day of class, thinking about all these amazing students at Stanford, the nexus of technology, culture, art, at a moment that is so special when everything seems possible. Researchers here at Stanford are 3D printing organs. An autonomous car can be taken all around San Francisco, or Austin, or Phoenix. Large language models are being used to potentially make sequences of DNA, so new life forms could be built. We are in this absolutely unbelievable moment right now.
Ostensibly the class is about making a project, but really what the class is about is when in a place of infinite opportunity to build so quickly, easily, and seamlessly. The interesting question is not what should be built, but why am I building it? Do I want it to say or to do? When in this space of infinite choice, understanding why decisions are being made and where values are is needed. We all have a finite amount of resources that can be dedicated to anything: energy, brilliance, network, relationships, technical skills, and the precious thing which is time. Time is the only resource that cannot be scaled, cannot be gotten back, that is equal for everyone, does not matter who you are or where you are. Time is tremendously valuable. The dedication of that time is one of the biggest choices made in life. Knowing why choices are made is needed because it is a reflection of the spending of this really precious resource, which is time. Knowing why one is building and making what one is doing, what the statement is that was wanted to be made.
The why for students and for everyone else, when authentic and real, will carry through a lot. If the why is known, it can create less resilience in challenging circumstances where overcoming and doing better, being better, working harder, learning faster, going farther than thought is needed. The why carries through a lot of that.
Keep EO in your library
Save the videos and channels worth coming back to, and find them again in one place.
