Rodney Brooks

Ah nice I've tinkered a bit with robotics few years ago, e.g. https://twitter-archive.benetou.fr/utopiah/status/1760260544450138140/ and it's pretty empowering... but also showcases how amazing the human body is. Our dexterity, ability to sense, etc. Reminds me of https://rodneybrooks.com/why-todays-humanoids-wont-learn-dexterity/ which is basically a stark warning against the hype.

Rodney Brooks has a great essay on why he's skeptical that the current humanoid hype will deliver and the central claim is that human dexterity is extremely advanced any today's humanoids lack even the sensors and data needed to start building the models needed to match human performance. https://rodneybrooks.com/why-todays-humanoids-wont-learn-dexterity/ I saw him post this article on his Bluesky saying that they're the first ones he's seen that are close to cracking this issue (he's an investor/adviser).

Perhaps change the headline to say famed roboticist == Rodney Brooks... because that matters. A lot. And I do agree with him. Today's humanoid robots are poorer than any task compared to other, specific / single-purpose robots. And they are more expensive. And they don't generalize. All humanoid companies are burning their $$$$ for sake of cute stories. None of the companies today are revenue generating, let alone profitable. Perhaps in a decade or longer, when we have better, more robust vision-multi-modal-action models we will have humanoid-ish robots that are actually valuable. And these too are unlikely to be human-like. Brooks's essay is worth reading in full: https://rodneybrooks.com/why-todays-humanoids-wont-learn-dexterity/

Rodney Brooks' predictions have not aged well. For example, in his 2018 blog post (http://rodneybrooks.com/forai-steps-toward-super-intelligence-i-how-we… he rates various approaches from 1-3 (with 3 being the best). Neural nets scored: Composition: 1 Grounding: 3 Spatial: 1 Sentience: 1 Ambiguity: 2 At that time, the potential of neural nets was already very clear.

Geofenced limited access highways also seems like a fairly promising area. It's amazing how many people bought into the hype that there would be widespread self-driving taxi services by the end of last decade and that kids growing up today would never need to learn to drive. Even a relative skeptic like Rodney Brooks is looking to be overly optimistic at this point. https://rodneybrooks.com/my-dated-predictions/

Great read. To follow up also read a comment on that piece titled "A better lesson" by Rodney Brooks: https://rodneybrooks.com/a-better-lesson/

His dated predictions about self driving cars, robotics & AI, and space travel are well worth reading: http://rodneybrooks.com/my-dated-predictions/

Interestingly, the iRobot founder and brilliant robot scientist Rodney Brooks puts flying cars and robot taxis in similar buckets as to wide availability: http://rodneybrooks.com/my-dated-predictions/

> best products Agreed, somewhat, but you are conflating "product" and "architecture". I specifically meant architecture. Intel was usually at least one process generation ahead of competitors, and often more than that. With Moore's law still going at full tilt, that meant not only more transistors, but also faster transistors, completely overwhelming the architectural deficits. > occasional superior alternatives but "often" is kinda silly Hmm...just off the top of my head: Motorola 68K, MicroVAX, LSI-11, SPARC, AMD 29K, Power, RISC, MIPS, Transputer, Alpha, HP-PA, ARM for desktop, Motorola 88K, Fairchild Clipper, etc. I'd be hard-pressed to name an architecture that wasn't superior to i386. But architecture did not matter. With Moore's law no longer also getting us faster transistors, it might matter again: https://rodneybrooks.com/the-end-of-moores-law/

Brooks' take is worth considering: https://rodneybrooks.com/the-seven-deadly-sins-of-predicting-the-futur… Musk knows nothing substantial about AI except what he's read in SF. Not a great resume for a Cassandra.

Feel free to ask for more details if you have specific questions! I worked in robotics for many years, I have some decent familiarity with this space. Here’s some more detailed thoughts “for free”: Humanoid robotics are largely a publicity stunt. Our actuators, sensors, and algorithms are better adapted to other form factors. The nice thing about humanoids is that you (in theory) don’t have to change the interface, since they can use the same interface humans can use. In practice that doesn’t hold well, because we don’t have great force/pressure sensors to cover large areas like human skin. Likewise, it’s difficult to apply the fine forces that are sometimes needed (grabbing an egg, moving a joystick, etc). And there’s risk of the robot doing something unpredictable, so you always have to set a good safety bound around it anyways. In the end it’s often better to adapt the process to modern robotics, rather than the other way around. There are many good practitioners that write about these and other limitations, I think Rodney Brooks has some good discussion of it, eg. https://rodneybrooks.com/why-todays-humanoids-wont-learn-dexterity/

Home assistance and nursing assistant jobs require competence in a large number of highly varied tasks and many require human-style intelligence which is among the hardest to automate from our experience with AI research. (See Moravec’s Paradox.) It will be at least a couple decades, likely longer, until we can really automate those jobs cost effectively, which gives us time to reconfigure socio-economic systems to suit the burgeoning new industrial revolution. Check out the following predictions by a noted roboticist and former director of MIT’s CSAIL https://rodneybrooks.com/my-dated-predictions/ Example from the page: Dexterous robot hands generally available. NET (Not Earlier Than) 2030 It will take at least a decade and often longer for a hardware product to become affordable to the masses. So we are talking about 2040 or later when most of home caring jobs will be at risk. Even then, the need for human interaction will still be there so it is unclear if most elderly would not want human services if it is accessible to them.

How are you engaging with the fundamental problems explained in this blog post https://rodneybrooks.com/why-todays-humanoids-wont-learn-dexterity/ - the relatively crude tactile and proprioceptive sensing apparatuses of robots when compared to humans - the limited availability of multisensory, perception-action coupled training data Genuinely curious!

Probably both vision and dexterity, and the first mistake we make as roboticists/engineers might be to distinguish the two like they're separate problems to solve or that a solution exists where the two live a separate life. https://rodneybrooks.com/why-todays-humanoids-wont-learn-dexterity/

Rodney Brooks? https://rodneybrooks.com/predictions-scorecard-2025-january-01/

Just look up “fine dexterous manipulation with pressure feedback” to see the SOTA for dexterous manipulation There’s at least a half dozen products, two recently from Unitree and Allegro announced. Rodney Brooks wrote about the challenges - but frankly it was a submarine piece for his work https://rodneybrooks.com/why-todays-humanoids-wont-learn-dexterity/

So, they released a new gripper with touch sensors. Then there's this article: https://rodneybrooks.com/why-todays-humanoids-wont-learn-dexterity/ Talking about human hands having tens of thousands of receptors of several different types, and the difficulty of tasks like picking up a match, and the trouble with the project of learning dexterity by brute force.

That’s irrelevant though. If the system requires human intervention, then it’s not fully autonomous by definition. See https://rodneybrooks.com/predictions-scorecard-2025-january-01/ for example: The companies do not advertise this feature out loud too much, but they do acknowledge it, and the reports are that it happens somewhere between every one to two miles traveled. That’s… not very autonomous.

I think your comment has some threads in common with Rodney Brooks' response: https://rodneybrooks.com/a-better-lesson/

So going by https://rodneybrooks.com/rodney-brooks-three-laws-of-robotics/ That would be violating the second design principle: "When robots and people coexist in the same spaces, the robots must not take away from people’s agency, particularly when the robots are failing, as inevitably they will at times." With a physical robot, if it fails and freezes, it turns into a hazard. With Clippy, it intrusively stops humans from being able to do what they are doing.

Rodney brooks would disagree with you: https://rodneybrooks.com/predictions-scorecard-2025-january-01/ > Now self driving cars means that there is no one in the drivers seat, but there may well be, and in all cases so far deployed, humans monitoring those cars from a remote location, and occasionally sending control inputs to the cars. The companies do not advertise this feature out loud too much, but they do acknowledge it, and the reports are that it happens somewhere between every one to two miles traveled

The author ends with the thought that a robot that has a stop button is a failure but his most famous product for consumers, Roomba, has such a button. Does he talk elsewhere about this? He makes a paragraph about robots that interfere with ordinary people as bad and references autonomous vehicles in San Francisco. He has a whole series starting here https://rodneybrooks.com/autonomous-vehicles-2023-part-i/ or with no comments! https://news.ycombinator.com/item?id=37971352

Related: robotics pioneer Rodney Brooks just wrote up his recent personal experience where he was riding in a Cruise cab, and it stopped suddenly in the middle of an intersection, nearly causing a bad accident: http://rodneybrooks.com/autonomous-vehicles-2023-part-ii/. Consistent with discussion here of Cruise cars being dangerously timid.

Read articles like this and avoid the hype: https://rodneybrooks.com/what-will-transformers-transform/

I think a lot of people thought it would be exactly as hard as it has turned out to be but were dismissed as being luddites or lacking imagination. There may have been a lot of people like that, but there were very few knowledgeable people who were making such predictions in public. The only one I know of is Rodney Brooks, and even he went public for the first time only in 2017 [0]. [0] https://rodneybrooks.com/edge-cases-for-self-driving-cars/

Who do you have in mind? In my corner of AI it's pretty uncommon for researchers to even predict "timelines". Predictions have a bad track record in the field and most researchers know it, so don't like to go on record making them. The only prominent AI researcher I know who has made a bunch of predictions with dates is Rodney Brooks [1], and he puts even dog-level general intelligence as "not earlier than 2048". I imagine folks like LeCun or Hinton are more optimistic, but as far as I'm aware they haven't wanted to make specific predictions with dates like that (and LeCun doesn't like the term "AGI", because he doesn't think "general intelligence" exists even for humans). [1] https://rodneybrooks.com/my-dated-predictions/

The marketing hype has overtaken reality. The current AI technology is not ready for real world deployment and it will not be ready for the foreseeable future: https://rodneybrooks.com/my-dated-predictions/.

Full autonomy is not possible with the current technology [1]. Unless Apple is sitting on some breakthrough in machine learning this is purely marketing material. 1: https://rodneybrooks.com/predictions-scorecard-2021-january-01/

Let us note for a moment that Prof Brooks is a bona-fidi genius inventor and made contributions to AI that changed the nature and direction of the field. He also has a good blog. https://rodneybrooks.com/blog/ I once did a talk at MIT and he made a comment about people wearing cowboy hats but not having any cattle, which made me sad. Still, it's not surprising that we have work to live up to our idols standards.

https://rodneybrooks.com/a-better-lesson/ is an interesting retort to the Sutton post.