Future Computers Will Be Radically Different

Future Computers Will Be Radically Different

Visit https://brilliant.org/Veritasium/ to get started learning STEM for free, and the first 200 people will get 20% off their annual premium subscription. Digital computers have served us well for decades, but the rise of artificial intelligence demands a totally new kind of computer: analog.

Thanks to Mike Henry and everyone at Mythic for the analog computing tour! https://www.mythic-ai.com/
Thanks to Dr. Bernd Ulmann, who created The Analog Thing and taught us how to use it. https://the-analog-thing.org
Moore’s Law was filmed at the Computer History Museum in Mountain View, CA.
Welch Labs’ ALVINN video: https://www.youtube.com/watch?v=H0igiP6Hg1k

Crevier, D. (1993). AI: The Tumultuous History Of The Search For Artificial Intelligence. Basic Books. – https://ve42.co/Crevier1993
Valiant, L. (2013). Probably Approximately Correct. HarperCollins. – https://ve42.co/Valiant2013
Rosenblatt, F. (1958). The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain. Psychological Review, 65(6), 386-408. – https://ve42.co/Rosenblatt1958
NEW NAVY DEVICE LEARNS BY DOING; Psychologist Shows Embryo of Computer Designed to Read and Grow Wiser (1958). The New York Times, p. 25. – https://ve42.co/NYT1958
Mason, H., Stewart, D., and Gill, B. (1958). Rival. The New Yorker, p. 45. – https://ve42.co/Mason1958
Alvinn driving NavLab footage – https://ve42.co/NavLab
Pomerleau, D. (1989). ALVINN: An Autonomous Land Vehicle In a Neural Network. NeurIPS, (2)1, 305-313. – https://ve42.co/Pomerleau1989
ImageNet website – https://ve42.co/ImageNet
Russakovsky, O., Deng, J. et al. (2015). ImageNet Large Scale Visual Recognition Challenge. – https://ve42.co/ImageNetChallenge
AlexNet Paper: Krizhevsky, A., Sutskever, I., Hinton, G. (2012). ImageNet Classification with Deep Convolutional Neural Networks. NeurIPS, (25)1, 1097-1105. – https://ve42.co/AlexNet
Karpathy, A. (2014). Blog post: What I learned from competing against a ConvNet on ImageNet. – https://ve42.co/Karpathy2014
Fick, D. (2018). Blog post: Mythic @ Hot Chips 2018. – https://ve42.co/MythicBlog
Jin, Y. & Lee, B. (2019). 2.2 Basic operations of flash memory. Advances in Computers, 114, 1-69. – https://ve42.co/Jin2019
Demler, M. (2018). Mythic Multiplies in a Flash. The Microprocessor Report. – https://ve42.co/Demler2018
Aspinity (2021). Blog post: 5 Myths About AnalogML. – https://ve42.co/Aspinity
Wright, L. et al. (2022). Deep physical neural networks trained with backpropagation. Nature, 601, 49–555. – https://ve42.co/Wright2022
Waldrop, M. M. (2016). The chips are down for Moore’s law. Nature, 530, 144–147. – https://ve42.co/Waldrop2016

Special thanks to Patreon supporters: Kelly Snook, TTST, Ross McCawley, Balkrishna Heroor, 65square.com, Chris LaClair, Avi Yashchin, John H. Austin, Jr., OnlineBookClub.org, Dmitry Kuzmichev, Matthew Gonzalez, Eric Sexton, john kiehl, Anton Ragin, Benedikt Heinen, Diffbot, Micah Mangione, MJP, Gnare, Dave Kircher, Burt Humburg, Blake Byers, Dumky, Evgeny Skvortsov, Meekay, Bill Linder, Paul Peijzel, Josh Hibschman, Mac Malkawi, Michael Schneider, jim buckmaster, Juan Benet, Ruslan Khroma, Robert Blum, Richard Sundvall, Lee Redden, Vincent, Stephen Wilcox, Marinus Kuivenhoven, Clayton Greenwell, Michael Krugman, Cy ‘kkm’ K’Nelson, Sam Lutfi, Ron Neal

Written by Derek Muller, Stephen Welch, and Emily Zhang
Filmed by Derek Muller, Petr Lebedev, and Emily Zhang
Animation by Iván Tello, Mike Radjabov, and Stephen Welch
Edited by Derek Muller
Additional video/photos supplied by Getty Images and Pond5
Music from Epidemic Sound
Produced by Derek Muller, Petr Lebedev, and Emily Zhang


  1. Austin Riess on June 23, 2022 at 8:16 pm

    Another comment, right at the end the usages of analog to digital, both have desired traits we want so swapping back and forth between them is similar of how our brains function. Digital to categorize while analog observes the phenomena happening.

  2. Dot Alzero on June 23, 2022 at 8:18 pm

    How do you make it pass 2:10? My ears cannot :(((((

  3. Jonathan Anderson on June 23, 2022 at 8:18 pm

    damn vid talks about how much energy and time it takes to teach and run these AI with only 100 layers of neurons. Nature does 86 billion neurons like its nothing. Nature is scary bro.

  4. james kirk on June 23, 2022 at 8:18 pm

    I have been thinking about this for a while, and although size and power still seem to be a bottleneck, I see future AI to be tunable lasers, fiber optics, and optical detectors. Very much mimicking the brain itself. Wavelength being the neurotransmitters, signal strength, duration, etc, I can only imagine a neural network being an analog computer of the future.

  5. גבריאל on June 23, 2022 at 8:18 pm

    Analog computers will always be special purpose and will become obsolete with the advent of quantum computers.

  6. Orange on June 23, 2022 at 8:19 pm

    I think future computers will have a digital processing unit and an analogue processing unit working together. That way, all the advantages of analogue and all the advantages of digital can be used together.

  7. oscvr on June 23, 2022 at 8:20 pm

    I wonder what quantum computing is doing for AI 🤔

  8. Onio Saiyan on June 23, 2022 at 8:25 pm

    Sounds to me this is analog with a digital front end and digital governance. Kind of like a CNC machine. Only instead of G-code it’s machine learning.

  9. en Cross on June 23, 2022 at 8:27 pm

    So its more like a hybrid of the two instead of one or the other

  10. Austin Riess on June 23, 2022 at 8:27 pm

    First, genius idea to utilise the digital piece for analog usage and then alotting values to them to determine answers. Second, I like the fact our brain uses tons of differently sensitive wired neurons to determine values! Im in robotics so this stuff fascinates me🤯

  11. Mohammad Hosein Aliyoldashi on June 23, 2022 at 8:29 pm


  12. Magnasium on June 23, 2022 at 8:29 pm

    I was initially expecting quantum computing, but this is cool.

  13. Karkess on June 23, 2022 at 8:29 pm

    My undergraduate work was actually with a professor who did research in the brain as an analog computer and using neural networks and analog computing as an attempt to achieve super-turing computation. A researcher who’s name is worth looking into in all this from my research would be Hava Siegelmann. At the time I understood much less about the problem. My task was essentially to try and prove that analog computation could be modeled with a neural network on a digital computer. Not sure if my comment will be buried or not, but it’s an area worth looking into if you’re more deeply interested in this problem.

  14. jimmy Burnett on June 23, 2022 at 8:29 pm

    OK fix the homeless problem. How about a climate change solution? How about stopping Hyper missiles?

  15. Kyle Vin on June 23, 2022 at 8:32 pm

    Too bad there are Soo many ads in the vid… 7ads is ridiculous

  16. Armin Lutz on June 23, 2022 at 8:33 pm

    So youre saying:Analog computer: really great at computing and showing a chaotic system – also analog computer: no exact imputs and about a 1% error. Now that sounds helpfull.

  17. Airship Today 2022 on June 23, 2022 at 8:33 pm

    Quantum Computers: Am I a hallucination?

  18. MrVipitis on June 23, 2022 at 8:36 pm

    Well, inference I suppose. Not training…yet?

  19. Rushiatruefan on June 23, 2022 at 8:37 pm

    So how do this going to help me watch rule 34?

  20. Mack .Doggs on June 23, 2022 at 8:37 pm

    Can analog computers replace the digital processor in cochlear implants? Wouldn’t this make them much more effective at translating pitch and timbre? Essentially having an analog processor would make it quite easy to accurately translate soundwave to the brain with infinite fidelity,

  21. Dr.SmoothBrain on June 23, 2022 at 8:38 pm

    Yeah it won’t be 1’s and 0’s it’ll be 6’s and 9’s

  22. Hiboyboy123 on June 23, 2022 at 8:39 pm

    If we can use digital and analog as well as our brains, we can test things on how to help fix ppls brains that would kill people who were tested on. We can also work on that, and make it think so much faster than we would, and start coming up with things, but by then, I think we’d have them far away from us

  23. Ammar Faiz on June 23, 2022 at 8:39 pm

    no way, you mean…. there is no way… NUMBER 2❗⁉️⁉️⁉️‼️

  24. Chris Hamberg on June 23, 2022 at 8:41 pm

    1:10 Yes there are 🤡

  25. Robert Weekes on June 23, 2022 at 8:43 pm

    Facebook / meta is gonna try to buy Mythic for VR full body tracking any day now 😉

  26. NeXt Wave on June 23, 2022 at 8:47 pm

    nice , video

  27. Majin Canon on June 23, 2022 at 8:49 pm

    One day we gonna get dual system home computers

  28. Nickle Spale on June 23, 2022 at 8:49 pm

    All his closing statements just confirm what I’ve been believing all along

  29. Divya Jha on June 23, 2022 at 8:52 pm

    Even u r puppet

  30. Thilo on June 23, 2022 at 8:52 pm

    Reality in nature is rather analog than digital, which is why I think that digital computing did twist our thinking into dualism – hence creating the many mad outbursts of conspiracy theories and sarcasm in social networks.
    If in the future this technology will be made analog again it might change our thinking and even feeling and therewith bring people closer to a non-dual self-realization.

  31. Domaine Botha on June 23, 2022 at 8:52 pm

    That marker on the cardboard though…

  32. Justin Case on June 23, 2022 at 8:53 pm

    I vehemently disagree.. While niche uses will indeed employ analog computers, the vast majority (>99%) of all computing needs are perfectly met via comparatively very cheap IC chips. Consider the fact that aside from playing video games on my PC, I am perfectly content to run three screens capable of displaying two simultaneous Full HD Video feeds with 5.1 audio on any combination of screens. The point being that aside from streaming videos, I use less than 1% of my 10 year old computer’s processing power on average, and even streaming Full HD, it’s still less than 30%.

    Besides, by means of programming, digital computers have been adapted to perform thousands, if not millions of different tasks.

    By comparison, nearly all analog computers are highly specific in the tasks they perform. If it involves a singular application, the analog solution, which must be designed and precisely machined can cost thousands to millions of dollars. A World War II firing control computer, for example, would today cost millions of dollars each, and had to be updated with new cams every time a new shell with different ballistics came out. IIRC, they developed a parametric approach involving a series of dial-in correction factors which accommodated new shells. However, digital firing control computers can be updated via software, a usually performed by a visit from the supporting contractor, but which probably could be accomplished in a pinch by military fire control technicians themselves.

    Then again, a general ballistics program with maneuverable shells renders the issue moot, except for the fact that modern electronics are both light and sturdy enough to handle the the 100s of Gs involved in the firing process while using EMP and EW hardened electro-optical viewing and target tracking systems to home in on the target. Example: The Javelin missile system*, light enough to be carried long distances by one person, with a reusable launcher that doubles as an IR scope, and capable of taking out tanks at a distance.

    Let’s see you do that with an analog computer!

    *I’m sure there exist many non-military applications, including the electronic ignition, emissions control, and gas mileage optimization circuits in my truck; the electronic temperature control and energy optimization circuits in my refrigerator-freezer, the electronic control circuits in my 10,000 BTU window A/C which enable it to draw just half the power of my 14,000 BTU dual-hose portable A/C; my voice-recognition TV remote…

    A voice-recognition TV remote is an outstanding example where digital excels and analog will not ever be able to match!

    That said, some applications for analog computers exist where they’re actually cheaper to make and will work under conditions digital computers find hostile.

    One such example is as a mechanical RPM governor. Small engines use air vanes connected to the carburetor. As the RPM increases, the air vanes are blown by the engine’s cooling fan, thereby reaching an upper RPM limit. Most people have seen governors on larger engines, usually as two spinning balls hanging at an angle. As they spin, they rise, applying a counter-force to the throttle, thereby limiting the RPM.

    Even in these applications, however, digital has taken over. Back to my truck… The digitally-controlled RPM limit when it’s in park or neutral is about 3,200 RPM. While in gear, it’s about 4,500 RPM. Then there’s a MPH limit of about 93 MPH, so while driving on Germany’s Autobahn, I was a slowpoke! Doesn’t bother me here in the United States, though, as I do follow speed limits, with no speeding tickets in more than three decades. 🙂

  33. neuroscienxe on June 23, 2022 at 8:54 pm

    This reminds me of how our brains work. They aren’t binary systems entirely. They’re not electrical, they’re electrochemical. Essentially, that’s analogous to a hybrid of digital and analog systems. Not in the sense he says in the video, but at a neural/glial level.

  34. Mike Yancey on June 23, 2022 at 8:57 pm

    I was a finalist in the state science fair competition back in the 4th grade. So around 1977.
    My project was a board about 2 foot by three foot, full of 2 and three position switches and colored lights. It was a logic board that could solve various types of equations. Pretty cool in a time when almost no one had ever touched an actual computer.
    In the end I learned absolutely nothing about computers. But I learned to solder really, really well from it.
    Moral of the story, if you can’t learn to code, at least learn to solder. 🙂

  35. Balthasar King on June 23, 2022 at 8:57 pm

    lol nope, the future is quantum computers

  36. darude sandstorm on June 23, 2022 at 8:58 pm

    Shout out to the person doing the animations in this video! Must have been very timeconsuming with all the nodes and lines 🙂

  37. javed00 on June 23, 2022 at 8:58 pm

    Loved the video, well explained and sufficient depth to trigger those neurons! Definitely stirred curiosity in me to learn more.

  38. Deep Mukherjee on June 23, 2022 at 8:58 pm

    Elon Musk has left that chat

  39. Micha'el Orlev on June 23, 2022 at 8:59 pm

    our brains are not just analog and digital, they are electrochemical. The neurons are not activated through electrical signals (only), but through chemical neurotransmitters which are geometric shapes designed to interact with their individual receptors. This allows for the brain to be a lot more complex than digital binary, or even electrical analog, since encoding of information in geometry, is a different language than electrical signals or digital information of any kind. In order for us to truly harness the incomparable intelligence of the human brain, I don’t just mean calculus, I mean the creativity, and illogical, irrational, and intuitive nature of emotional feedback loops, we should be thinking of ways to biomimic the chemical nature of our electrochemical brains, not just digitally with neural networks, but geometrically with physical objects.
    Those are my thoughts anyway..

  40. Divya Jha on June 23, 2022 at 9:01 pm

    Those who told u is lying

  41. Antwan on June 23, 2022 at 9:03 pm

    This is amazing

  42. Mohammad Faisal on June 23, 2022 at 9:06 pm

    Assalamu Alaikum. You make such amazing content. I understood how neural network works fundamentally without even thinking about it….

    Love you man. Long live.

  43. Divya Jha on June 23, 2022 at 9:09 pm

    Even you don’t know anything

  44. Mike Stone on June 23, 2022 at 9:12 pm

    AI progress is scary for sci fi Machine VS Human war scenario.

  45. Zagstrug on June 23, 2022 at 9:12 pm

    I really love the way this video explains every step that was needed to get us here. Each mentioned topic makes sense in the context of of understanding the big picture and the transitions between them were done so well!

  46. Arun K on June 23, 2022 at 9:13 pm

    It would be a beautiful world… Where we don’t have to make assumptions while solving problems..

  47. Coach Vlad on June 23, 2022 at 9:14 pm

    Why not just hybrid computers, hell even add bio organic tissue if it enhances performance

  48. Fganr on June 23, 2022 at 9:14 pm

    Thank god for computer science.

  49. eSKAone on June 23, 2022 at 9:14 pm

    Wow! 💟

  50. Κώστας Κοζαδίνος on June 23, 2022 at 9:15 pm

    Human success rate isn’t 5,1%. If you have a dog and a book of all known species of dog in front of you, you can guess acuretly what dog you are looking at. In adition a computer have 3,1 chances to make a mistake but a human close to 0%.

Leave a Comment