My goal with this blog is to find the coolest and funnest ideas in robotics, and to have fun with them. It’s a big job, I don’t know if I’m up to it, but it’ll be fun to try, and I hope it’ll be fun to read. Even better, I hope readers will join in the fun and offer their own contributions in the comments.
If you’re still with me, you’re probably wondering, “What are these fun ideas?” Here’s a good one:
- Animals are robots. That includes humans. We are robots.
If a robot is an intelligent machine, then we are robots, and evolution is a roboticist. Everything we learn in robotics, we can ask whether it might apply to ourselves. Here’s a variation on that idea:
- Each of us has an inner robot.
When you walk, run, clean, cook, or whatever, it seems that almost all of the required brain activity is subconscious. (Sometimes roboticists refer to that subconscious stuff as “physical intelligence.” See the Ted talk by Sangbae Kim.) If you, the conscious self, had to figure out what every muscle should be doing, every fraction of a second, you would be in big trouble. Instead, you have an inner robot that takes care of the details, and you are mostly just along for the ride. Your inner robot is like a pet! You can train it, and guide it, but it doesn’t speak English. Be sure to take good care of your pet robot!
Here are two more fun ideas:
- Anthropomorphism: Robots should look like humans.
- Functionalism: Robot designs should optimize performance.
If we really want to have fun, we could have a fight: Anthropomorphism versus functionalism! Should robot designers be good engineers and let form follow function? Or should they copy evolution’s design?
Some tangential questions come up: Is the human form optimal for our tasks and our environment? Is evolution done tweaking the human form? Maybe you’d rather have your eyes on stalks. That way you could have one eye looking forward and one backward, when needed, and you could get your eyes into all those places your head won’t fit. And you wouldn’t need a mirror! And while we’re at it, should we have lights alongside our eyes?
Some of the funnest ideas are pretty far out there. Let’s call them the wild ideas. For example:
- The obsolescent economy.
When robots can do everything humans can do, including making more robots, then all costs go to zero, and the economy as we know it will cease to be. This idea was recently pushed by Elon Musk, but I first saw it in a research proposal when I was in grad school at the MIT AI Lab. I don’t know who wrote it, but Marvin Minsky is a likely source. It said that one day, when robots are building robots, robots will be so cheap that each head of lettuce growing in a field will have a robot squatting next to it, watching it grow, and flicking off the occasional grasshopper.
- Uploading our minds.
When AI is solved, we can upload our minds to the cloud, and then we can share with each other as deeply as we want, even merging minds with each other and with AIs.
Besides the fun ideas and the wild ideas, there are the dangerous ideas. Dangerous, in the sense that they are fun, but they can absorb all your thinking without meaningful progress. Black holes. Bottomless pits. Tarpits. You might start pondering one of these ideas, and never be heard from again. Keep your distance! I have examples, but I’m not sure I should share.
My plan is to stick to the fun ideas, and mostly stay away from the wild or dangerous ideas. But, who knows? I hope to follow the fun wherever it leads. I hope you will enjoy it enough to contribute.
I was once writing a sci-fi novel close to uploading your mind, where poor people’s brains are used as “GPUs” for rendering hi-res scenes for rich people… However, I only made it to Chapter 1.1 😉
Regarding Anthropomorphism vs Functionalism: let’s think about home robots, would it be a full adult size or kid-size humanoid? Or maybe a collection of Roomba (tiny) like special purpose robots/devices?
Hi Matt!
Congrats on your Blog!
Proteins are robots. They can sense, react and self-organize to carry out precise tasks automatically.
Hey, Chris! I’m sure I would agree with you, if I knew more about proteins. Your comment has goaded me to address this ignorance. I ordered a copy of Bruce Donald’s book “Algorithms in Structural Molecular Biology.”
I had the same thought but with DNA! DNA are the real robots while almost everything living thing (plants, animals, fungi) are vessels for these robots to live and replicate themselves in different environments. I believe Richard Dawkins called this the selfish gene theory.
Here is a dangerous one: Robots should (will) be conscious.
You’re probably right. Maybe I should have killed your comment!
Welcome to the world of blogging. This looks like it will be fun.
Congratulations!
“When you walk, run, clean, cook, or whatever, it seems that almost all of the required brain activity is subconscious.”
This reminds me of Daniel Dennett’s competence without comprehension: “Competence without comprehension is the way of life of the vast majority of living things on the planet and should be the default presumption until we can demonstrate that some individual organisms really do, in one sense or another, understand what they are doing.” From Bacteria to Bach and Back.
Thank you for starting this blog Matt!!
Thanks Kostas. I thought the work would be in writing the blog posts. But now I see I will be chasing down and pondering new leads. I’m a big fan of Dennett, but hadn’t read that one. Now i have to! That quote — perfect.
That sounds like how deep neural network behaves. May be explainability (XAI) is not always necessary and useful.
No matter what the down sides are, I’m pretty sure robots will be fun
I like it. Imagine Skynet is real, the Terminators are wiping us out, and the last two humans turn to each other and say “Wow, isn’t this fun!”
And, the blog seems to be more work than I expected. I didn’t realize I was going to want to respond to every comment!
Amazing! Thank you for making this Matt! Here is a fun one:
Robotic Dream Synchronization: If we each have an “inner robot,” what if we could tune our inner robots to synchronize through shared dreamscapes? Imagine a future where we connect our subconscious “robotic” minds through neural interfaces to dream together — if you know kung fu, your inner robot teaches my inner robot kung fu — bam! The Matrix 😀
Oh, interesting. My thinking hadn’t gone in that direction. Hmmmm. If the science / technology advances far enough we can upload our minds, then all kinds of things are possible. It might go far enough to reveal to us all of those dark corners in our minds, not just the inner robot. Yikes. Seems like we are getting into cyberpunk novel territory.
Matt congratulations — I enjoyed the blog and look forward to the next one! I love how you say our pet robot does most of the heavy lifting, and thank goodness it doesn’t consult us every time it wants to lift a foot or flex a finger. I’m not up for micromanaging my muscle fibers 24/7!
Matt says: “Animals are robots. That includes humans. We are robots.”
Are all living things robots? What sets all living things (including robots) apart from non-living things? Is “living” the same as “intelligent”?
Answers: Yes. The interactions between components. Yes, I suspect it is.
Is it not surprising that there is a part of science concerned with material things, where – historically speaking – it has been relatively easy to come up with laws, principles, and other statements that generalize very well? Like f = m . a, it seems like it holds for a wide range of conditions, across time and space.
And then there seems to be a second branch of science about animate things, i.e., matter arranged in such a way that it behaves distinctly from “just” matter. In fact, it behaves so differently that we call it alive. But in this second branch, science has not been as successful in coming up with laws and principles that hold under a wide range of conditions.
What are we missing in the second branch of science? Should we approach the second branch differently than the first branch, i.e., differently from how we approached traditional disciplines that physics? Is that maybe even something that is holding us back? That we are attempting to approach living things with a physics mindset, kind of bottom-up from a hard-core empirical perspective? Should be open to significantly expanding our scientific methods and tools to tailor them to the phenomena in the second branch? Or is life just hopelessly messy and there will never be such a neat and tidy list of principles as there seems to be in the realm of physics?
Hey, Oliver, thanks for jumping in. By a weird coincidence, I just picked up a copy of Norbert Wiener’s “Cybernetics,” and it begins by noting the difference between Newtonian mechanics, where time can be reversed and still obey all the physics laws, versus e.g. thermodynamics, where that is not so. I think Wiener might have said that is the same distinction you are talking about. Yes? No?
Dear Prof. Mason and Prof. Brock, Thanks for raising a very interesting discussion. I am thinking a lot about physics of complex systems which I believe is relevant here.
The key features which define living systems often include non-equilibrium, self-replication, metabolism, and compartmentalization. As Rasmussen et al (2010) says,
“The heritable information (genes) is of course also crucial to the bottom-up approach. But without energy, clearly no life is possible, so a metabolism capable of fuelling the life process is just as necessary. A container also seems unavoidable: the energetics and information need to support each other’s production, which can happen most conveniently in some sort of corral, such as a membrane.”
Robots appear to be intelligent as a human-invented algorithm drives its sensor-actuator networks. In my opinion, a human-made robot only represents a “technosignature” (signature of the existence of life) but it is not a living system. The difference between a living and non-living system is very clear when one considers a robot such as Boston Dynamics robotic dog and a real dog. The robotic dog mimics a few life-like characteristics that emerge as an outcome of a human-designed algorithm. The universal definition of life is indeed a complex problem, and we often describe “Life as we know it”. However, all living systems harnesses complexity to persist and perform complex tasks that require information processing. We recently showed that molecular complexity could be used as a signature of life and could distinguish living systems from non-living systems (ACS Cent. Sci. 2024, 10, 5, 1054–1064).
I believe if all living things are considered as robots, it could signify that the universe is computational at the bottom as suggested by Wolfram. Do you consider the universe to be computational?
Physics is analytic (I believe it is referred as the first approach) and complex systems including biology/living things are often considered as algorithmic (referred as second approach) in nature (Thurner et al. Introduction to the Theory of Complex Systems, 2018). The key difference comes as physical systems are governed by a fixed set of laws. However, laws governing biological systems are dynamic in nature. A simple example is replication-mutation dynamics, such as the Quasispecies model developed by Manfred Eigen. Additionally, there is no concept of novelty in current models of physics. The key reason is that there is no clear distinction between a system and its surroundings, as often discussed in statistical mechanics based models. The contingent variation within a living system as an outcome of evolutionary processes becomes part of the environment that influences its future outcomes. I believe the two approaches, analytic and algorithmic, have strong connection as suggested by Charles Bennett (Thermodynamics of Computation). However, a lot is required to be figured out to have a consistent theory that could combine physics and biology together. Recently, we developed Assembly Theory as an attempt to combine physics and biology by considering the physically plausible combinatorial spaces of future possibilities in an evolving system (https://rdcu.be/dYLL5 ). This allows us to define a generalised definition of the selection process (pre-Darwinian), which is essential for any system having a propensity towards life (and intelligence). I believe even at a very early stage, Assembly Theory has the potential to bring two approaches together.
Looking forward to hear from you.
Dear Abhishek,
Thanks for bringing a “life” perspective to this discussion.
To most of your questions I have no good answers. When it comes to life, philosophy, and almost everything, come to think of it, I am a dilettante. So here are a dilettante’s answers. Maybe it will provoke some better-informed readers to answer.
I think you are suggesting that only living things can be intelligent. Being a generic AI guy, my view is that intelligence is computation. The brain is a computer, and the mind is a computational process. How do we define intelligence, in such a way to differentiate it from lesser forms of computation? I don’t know, so the easiest path is to assume there is no meaningful distinction. John McCarthy had a beautiful example of this view: a thermostat can have either of two ideas: it’s too hot in here, or it’s too cold in here.
Do I think the universe is computational? No, I think of the universe as a dynamic system following the laws of physics. Even a computer is a hunk of stuff following the laws of physics. Signals, symbols, communication, and computation are all models that we share and apply as appropriate. A computer is not inherently computational, but the computational model of computers works incredibly well, thanks to the efforts of zillions of engineers.
( So these “words” are actually just patterns of dots on your screen, produced by the laws of physics. But I thank you for choosing to model them as words, and for modeling me as an intelligent agent 🙂
Re different types of dynamic systems: I cannot even venture an answer to this one. Assembly Theory looks interesting. It seems to address a question that has bothered me for years, about how to compare living things with engineered things.
Hi, Matt. It’s been many years since we met at AAAI etc.
“When robots can do everything humans can do, including making more robots, then all costs go to zero, and the economy as we know it will cease to be.”
Uh…huh? Mice reproduce themselves, but there are certainly resource costs involved. Mechanical systems involve more complexity, logistically, to create than biological ones, so how would costs decrease to zero? Anything that requires energy entails cost.
But meanwhile, robots and AI just allow humankind to extend our reach and investigate more; there’s always more fun (and work) to be had. I wrote a blog on it a decade ago:
Fun ideas. I personally think specialized, function-based robotics will dominate, but I have to admit that being able to rotate your head 360 like Boston Dynamics’ latest could come in pretty handy.
Anyway, nice to see you’re still keeping people thinking!
Yes, long time no see! I didn’t know about your blog. I have some catching up to do.
Congrats~
And look forward to the follow-up blogs！
Dear Professor Mason,
Your blog reminded me of some thoughts I had in the past. A few years ago, I watched the Netflix documentary Babies, which introduced various studies about infants. It mentioned reflexes that babies only have during the first few months after birth, like the grasp reflex. I started to wonder if these reflexes might be similar to a computer system’s built-in programs. If we consider humans as robots, perhaps these reflexes are like factory-installed algorithms that help the “robot” quickly acquire essential survival skills. The timing of these reflexes disappearing seems to coincide with when this initial learning is complete, as if the program finishes its purpose and shuts down permanently. I wonder if understanding these reflexes might even provide ideas for robotics research.
Fun! Insects (especially social ones) are 100% just wet-ware robots. And if you are interested in the idea of uploading minds I highly recommend the short story “Lena” https://qntm.org/lena
On functionalism vs. anthropomorphism: Our robot, Digit, is focused on function. The function is interacting with humans, in human spaces, doing workflows designed for humans. So it ends up looking a bit like a human. It’s very much the “biomimicry” vs. “bio-inspired” discussion, and I think understanding why a thing is the way it is in biology and then engineering to that purpose from a blank page is the right way to go! But also hard to know why a thing is the way it is in biology. I love it when we discover something about biology through our engineering exploration on robots.
Jonathan,
Well said. I’m tempted to print it out and hang it on the wall.
For readers who don’t know, Jonathan is the co-founder and Chief Robot Officer of Agility Robotics. And Agility is the creator of Digit, which I would guess is the most successful humanoid robot to date, except, is it a humanoid? As Jonathan says in his post, it looks a bit like a human.
Jonathan, I’d love to learn more about the engineering process and the evolution of Digit. But, I know you have to be careful not to give your competition an edge. If I could ask one question, I would ask about wheels. And I found the answer here. Nice video, with some great shots of Digit in action, and a discussion very much to the point, on the issue of wheels versus legs. So let me tweak the question. What about skate shoes?
In any case, thanks for your wise words.
Thanks Matt! I Just noticed this post, so here’s a reply: I think the applicability of skate shoes is narrow. In theory, walking can be close to as efficient as rolling, and I expect engineered systems to approach that eventually. Feet are useful because you can change the center of pressure, and help with balance. More importantly, feet (in animals, not yet seen in robots) “catch” the mass and reflected inertia of the leg on touchdown. It would be difficult to get a wheeled foot to do that; I think the complexity and engineering of a wheeled foot would make it much less good at legged locomotion. I don’t see a first-principles reason that some form of skate feet can’t transform and have the best of both worlds, but I think in practice, you either get good legged locomotion, or you get wheeled locomotion that can articulate well.
Thanks Jonathan!
Matt — great seeing your blog.
“When you walk, run, clean, cook, or whatever, it seems that almost all of the required brain activity is subconscious.”
I looked deeply into this when at NASA — it is not just subconscious, most of the details of physical execution are occurring in the spine, not the brain. You can see this from the “decerberated animal” experiments that happened almost a century ago, where the brain stem was severed, but the animal kept alive and studied for the range of complex physical actions it could execute, such as changing gaits as a treadmill went through different speeds. The implication is that there is a hierarchical dynamical system where the brain sets high level goals, and the spine executes complex reactive behaviors — and the spine is a decentralized modular system, so that itself has very interesting implications.
I talked about this in-depth in a lecture I gave some years ago for the NASA Ames Chief Scientist and explored the implications for how motion control occurs in animals and how we translate that to biologically inspired robots. The section on the spinal control of higher level motions starts at the 4:49 minute mark and includes videos of the decerberated cat experiments. I suspect you will enjoy the whole presentation if you have not seen it yet.
https://www.magicalrobot.org/BeingHuman/2017/01/my-2016-nasa-ames-summer-series-presentation-superball-a-biologically-inspired-robot-for-planetary-exploration
Vytas
Hey Vytas, long time no see.
Thanks for the pointer to the lecture. It is excellent.
Good point — a lot of the “physical intelligence” is not in the brain, it’s in the spinal cord. If I were less lazy I would say “central nervous system” instead of brain. I had heard of the cat experiments, and I was eager to see the video, but … sheesh, what a horrible thing to do to an animal.
One crazy idea: with perfect humanoids and brain-computer interfaced, put human brains in the robots to achieve true intelligence.
Hi Matt
Wonderful to meet up again in blogspace. Former CMU robotics student, way back in the 1990s.
Scary robots are sometimes funny robots? Bongard’s Xenobots [https://cdorgs.github.io/] come to mind.
I have a course on Robots and Animals for STEM and non-STEM students and will definitively work your blog into the materials.
Thanks for all the inspiration.
-marc
Marc! Great to hear from you. Your project at CMU was one of the coolest and funnest robotics projects ever. Is there any publication or video?
I was unaware of Bongard’s Xenobots. Awesome; thanks for the pointer.
HI Matt
Here some documentation on that robot-chicken interaction
Links to two papers at the bottom of that page.
The whole Earth is self-sustaining robot, perhaps…
Some new work on GeoAI maybe timely, if decidedly off topic.
Hi Matt. The blog is a good idea.
But saying that we’re all robots is just taking the useful semantics away from a perfectly reasonable word, “robot”, which was imported by Karel Capek from the Czech word for “worker” to mean an artificial human that could be built in factories (from artificial protoplasm) and treated as a slave. (Read the original, and see how it came out!) If you want to say that we are “meat machines” (or “protein machines”) just say so.
Meanwhile, let’s look at the economy. It doesn’t exist just to make goods, services, and money. It also exists to provide people with meaningful work. Through robots and other automation, we can create more goods, services, and money, but how do we provide meaningful work for everyone who needs it? It’s not going to happen because of some “invisible hand”. But we can figure out how to create that meaningful work, and make life better for all of us, if we have the smarts and the will to do it.
I think there’s some really fun ideas along these lines, and there’s a lot of clever folks reading your blog. So let’s solve another important problem or two.
All the best,
Ben
Hey, Ben, thanks for contributing!
But I like saying that humans are robots. I like saying that humans are meat machines, too! By the way, I was re-reading Cybernetics, by Norbert Weiner (trying to anyway, it’s not easy) and he says that animals are automata. Actually he says that “Descartes considers the lower animals as automata.” So, that’s pretty good precedent, even if it isn’t exactly the same word.
About finding work for people. Yes! Even if the invisible hand does eventually address the problem, a lot of people might suffer in the process. I wonder if any readers have ideas about how to do it?
Thanks, Matt! As you know, I’m not a big fan of military force as a scalable approach to conflict resolution. However, the US military is one of the largest and most effective educational institutions in the world, transforming the lives of countless young people by teaching them both practical skills and the discipline needed for cooperative action. I think if we look to the future, and we want our society to provide meaningful work for lots of people, some of those can come from teaching younger people skills and discipline. As they learn, they can be building housing and improving our environment. There will also be work for robots to do, but work is a valuable resource, and lots of it should be reserved for humans.
H Matt! Great blog! Your old pal John Craig.
Wir sind die Roboter ‘D
Evolution is a roboticist! Great sentence.