Atomic Brain? - Computerphile
Вставка
- Опубліковано 15 січ 2025
- How about a Neural Net where the neurons are actual atoms? Professor Phil Moriarty shows a paper demonstrating the principle from researchers at Radboud University in The Netherlands.
Professor Moriarty's blog with more detail: bit.ly/C_Atomic...
/ computerphile
/ computer_phile
This video was filmed and edited by Sean Riley.
Computer Science at the University of Nottingham: bit.ly/nottsco...
Computerphile is a sister project to Brady Haran's Numberphile. More at www.bradyharan.com
![The moment we stopped understanding AI [AlexNet]](http://i.ytimg.com/vi/UZDiGooFs54/mqdefault.jpg)
Next video: "we got Doom to run on a single atom"
Can a single transistor run doom?
@@incription yes but at 1 frame per year
@@Yobleck No. A single transistor cant even store a bit.
@@deathtothebeardless2959 Yes it can, that's what NAND flash memory is based on. Floating gate MOSFETs. A single transistor can even store 4 bits in QLC memory. I'm pretty sure a single transistor can't run doom though.
@@DehimVerveen What the
How?
"This is a big wet squishy thing with lots of very strange, weird stuff happening" 7:31
As a neuroscientist, I approve this message.
As an avid e621 enjoyer, I too am a dragon curve enthusiast.
That's too bad for the both of you
@@thekingoffailure9967 I like you
as a corrupt-minded, i misinterpreted this in a very nasty way
Prosecutor: "Why did you kill this man?"
Me: "Well the system wants to reach its minimum energy state."
A professor through and through. That power of voice and engagement even in a small room with just a couple dudes and he's speaks like he has an auditorium of a thousand.
Apart from the whole turn-it-off-and-on-again thing.
he literally has an audience of millions though
MORE ON THIS TOPIC PLEASE!! This video needed to be WAAAAY longer
Math not meth
I genuinely thought we would never see a day where we could do this at the atomic scale. That is absolutely wild
they've basically invented computronium, right?
Modern computers already operate on atomic scale. The real importance of this is that simulating neurons requires disproportionate amount of gates, whereas with this kind of technology you only need equivalent of 1 gate to simulate a neuron.
@@michaelbuckers I'm aware. I've seen how components like SSD's and other components work, but this is a 1 to 1 ratio of atoms to bits, instead of using a coupled hundred or thousand atoms to represent transistors and whatnot. That part is what blows my mind
We've had AFM and STM (these microscopes) since the 1980's
And we've been experimenting with these ideas ever since.
Check out "a boy and his atom" to see a movie made entirely in atoms.
@@allhumansarejusthuman.5776 I've seen that video, and I know the tech has been around for a while, but we have just been playing with these ideas for so long, that I never expected something like this to actually be worked on right now
Science today: Atoms can remember (12:50 - 13:10).
Homeopaths tomorrow: We were right all along.
I don't know whether to praise or curse you for pointing this out. I just hope it's not a self-fulfilling prophecy
Oh no....
used
"No actual physics was -harmed- in the creation of this illustration"
lol, that's fantastic
9:35 for anyone trying to find it
Humans: "Okay, we don't need to use Cobalt any more after we invent better Li batteries..."
Cobalt: "01001110 01101111"
Cobalt is also useful in the semiconductor industry. But it this research goes anywhere then it will become even more prevalent in computers. Which is TERRIBLE news. Cobalt is mined in Congo using child slave labor.
@@neoqueto Maybe in the future when we mine in space, then.
Can Cobalt be in asteroids, though?
@@neoqueto But the number of atoms in 1 gram of kobalt gets you a lot of computing power. Like, more than all living human brains combined.
That's one of the most exciting aspects of the atomic brain, approaching maximum information density.
Plus, think of how much energy well be saving when most data processing and storage happens at that scale.
@@neoqueto someone should tell them not to do that
@@neoqueto maybe cobalt isn’t the problem, maybe it’s the child labor
Very fascinating, thanks Dr. Moriarty and Shaun!
Quite the cliffhanger with "…the system can learn about its previous experiences." More on this please!
Convenient my name is Shaun so the "Hi Shaun" really feels personal :D
Hi Shaun
Hi, Shaun.
Shaun, hi
If we had quantum like effects in spoken language he could say hi to all of us at the same time and we would each resolve the actual sound he made to mean what we personally expected.
Hi Shaun
Why do I feel a sudden urge to rewatch all those videos on AI safety? Robert Miles, heeeelp!
Hi I'm an anti vaxxer anti Vicks ViX VEX methamphetamine in nasal decongestants as well as ketamine 💉 and *cough dxm olney lesion brain damage and CYP2D6 neglectful health care industry opiate pushers and murders ... Mask up mosque up 🤪🤐😷 ignorant🤐 alla🤐 myanmar gold thieves.. pot is a crime Optimus prime last knight DEBIAN linux anti pill gate way drug megatron metatron qubes os lover.. math not meth rip dmx 😔 *cough
No idea, are you trying to hide a massacre or something? Or some fine as* puntang 🤪
1984 exodus 3:14 serious SIRIUS david divide ✡️/✡️ star of david sirius epoch 1894.13 🧐
💯% A+ E=Mc² einstein.... That's MEc²a MEcca Mecha Michael not cheMicheal chemical 🧐 math not meth
J e s u s C H R I S T .. I thought I said MATH NOT METH 🤬
Professor Moriarty looked almost wistful when mentioning how we know Mike Pound.
Sir, fear not, the comment section’s capacity for love is enough for both of you
I think there isn't a zero energy state in Prof Moriarty... Loving this guy to bits!!!
Moriarty is a rad surname. Even if I have to fight the urge to make a Sherlock joke at any turn. Many of us have much less neat names :p.
Every time I see him I want I to say " But Bluebottle, he's fallen in the water."
The first job I worked they gave me a key to the woman's locker room
Literally the best computer science channel in the existence.
There's a 99.9599999% chance that I really enjoyed this video.
Cobalt atoms: One one one one one one zero! One one one one one one zero!
Tunnelling microscope: You did it! You broke computing down to its bare essentials!
This approach gives new meaning to the term "electronics"
It's always used electrons, lol.
@@andybaldman Not *individual* electrons, though.
@@gloverelaxis Yes, individual electrons. Just a lot of them!
@@andybaldman
Uh.....
@@gloverelaxis well you do have single electron transistors...
This is like suddenly being able to travel across the Galaxy in an instant and its just another good UA-cam video
I just loves Phil Moriarty presentations, they make me wish I'd continue to university back in the days (getting close to 50 it's a tad late for such career change).
His engagement, energy and the clear ways in which he explains these subjects are astonishing to this machinist.
Statistical mechanics blew my mind as an undergrad and even today it's nearly unbelievable it's almost magic
Everybody gangsta till the Cobalt atom starts asking what's the meaning of life
Damn I love his passion about this topic. That's how professors should be!
That's how all professors are when it is a topic they care about, you just had profs teaching the wrong classes
@@sheeplessknight8732 What are you, a freshman? That's blatantly not true.
I don't know if this supports or disputes your theory, but I had a diabetic professor teaching evolution and sociobiology, he jumped up on his desk and shouted "I really believe this is true!", then he regained composure and asked if anyone had a sugary candy. This was almost 30 years ago and I cherish that memory.
What a decade to be alive!
"We've got an atom on a surface in ultra high vacuum", so this is how excited a physicist gets when his dream of a spherical cow in vacuum finally becomes a reality!
And you program it in Cobalt?
I just saw the thumbnail and automatically read the title with that sweet accent :)
Moriarty :This is a big wet squishy thing with lots of very strange, weird stuff happening.
Cobalt Atom: Hold my quantum brain.
Great video. Definitely keeping an eye out for future developments.
Phil should start a channel about stuff he's interested in, but for people who have a STEM background so he can communicate at a higher level about the cool bits and not say stuff like "the devil is in the details." I want the details!
Whenever i see Prof. Moriarty i get excited like a 9yo... i mean come on the dude works with the freaking atoms, its so interesting! I only wish James Clewett returned with a new video or two, havent seen him in ages.
Curious if they are using the frequency of the state change in any meaningful way
Professor Moriarty is the most interesting guest on your channel
Love Prof. Phil
This seems like an actual possible way forward for Artificial General Intelligence. I wonder if we'll run into a practical limit, or not.
What a mind blowing video! 1 atom transistor.
If I'm remembering this right, what he means by *"changing the electronic configuration of an atom"* is _adding or removing electrons orbiting the atom._
As in "1s 2s 2p.." transitioning to either "1s 2s" or "1s 2s 2p 3s", etc? Please correct me if I'm wrong.
I'd love to hear more about this stuff, brain computations, consciousness, etc.
I don't see how this is like a Neural Net. It seems just like regular computer components at the atomic level. Which is still cool, but what's Neural about it?
The fact that it adjusts the weights of the connection between atoms. He mentioned that towards the end
More! I'm in love with this topic, absolutely thrilling. But don't less us distract if we are.
what's up with the tinfoil shelf? are those all the tools that build up static? (I couldn't see a wire grounding it though)
💩
"The atomic brain" sounds like a 1950s horror/sci-fi flick.
We need more videos like this
I miss him too. Him, Pound and Brailsford are the three G.O.A.T..S. ... hehe
"Hey check this out! I've constructed an atomic register!"
"You can get funding from ignorant people if you call it 'neuron'.'"
"Hey check this out! I've constructed an atomic neuron!"
Pretty much, neurons have way more going on in them then a simple on/off.
"It can learn from it's experiences"
Looks at the dents in my car... yep, where's my funding cash! XD
@@beskamir5977 ...which is exactly the point I make in the video and in the associated blog post (see link in the video information.) It's the weight of the synaptic connections that is key and that is tuneable in Khajetoorians et al.'s model.
Philip (speaking in video)
@@thequantumworld6960 Okay fair, but how accurate is the tuning of that and even then I don't think you could reduce a neuron to either on/off plus some connections.
Are the electrons perfectly efficient? Always able to oscillate?
Yup!
I really didn't understand what the professor meant by changing the electron configuration of an atom. From what I have read in my high school chemistry, each element has a specific electron configuration. For example, carbon is 1s2 2s2 2p2. Then I read that it is the ground state and the excited states have different configurations. It was very exciting! 😁
"Configuration" in terms of the number of electrons are fixed, but the shape of the "orbit" or probability of positions can change. One example is the energy state which can change when affected by energy from an external source, and then collapses to release that energy; like in a neon sign or a laser.
How is this idea different from the traditional Qbit? The possibility of many states- 0 or 1 or anything in between- is already in use, right?
8:03 -> Could we change the electron bonds of an atom while such a computer is running? Because in that way the more bonds, the more likely it would be to transmit energy (more easily) and from a chemistry perspective, you would be effectively "fixing" the connections based on how much the paths are used?
hey that guy from sixty symbols. great explanations.
Even at the time that video was released you were preparing for the Omicron variant :))
(it can be seen on the table side at 2:07)
Is it possible to have a link to the paper(I haven’t finished the video so sorry if he mentions where to find it later)?
See the link to the blog post given in the video information for much more information and detail.
Thank you
So that is how Intelligent Calcium (iCa) works. I'm glad they worked with Cobalt, or we could have a Helvetica Scenario.
I'm glad they didn't replay Helvetica Scenario scene. Once was enough.
Wow, these stuff are crazy brilliant!!!
Question is there a smaller atom that could function similarly? Imagine if you could do this with hydrogen and helium atoms, would be extremely compact, more efficient perhaps? Also could we drill even further down to quarks level?
maybe you could even go down to strings, if those are what make up quarks. Those strings (if they exist) would also have to vibrate inside of 6 dimensional calabi-yau spaces… so at that point the intelligence would be transcendent?
Wow, this is really fascinating and cool! 😀👍🏻
Looks like the boys are back in town.
Please more of this
Pls Available captions too 👍🏻
I feel like some time in the future, some one is gonna figure out how to make neural-nets be perfectly trained nearly instantly on quantum computers.
I think Sir Roger Penrose suggests that the brain works at the quantum level where there are 3 states instead of just 1 and 0s. He is doing research on microtubules which are potentially responsible for the quantum processing between neurons. This is one hypothesis why we will never be able to "write" a program to simulate consciousness with 1 and 0's.
"neural net" analogy Smorgasbord.
Where is this filmed?
We are just a dream of some Boltzmann brain anyway?
Just when it started to get really interesting, part one ended!!
Have they made the smallest possible transistor for a classical computer?
Or, is this the building blocks for a quantum computer?
Its not habit tech, which is probably why he emphasizes the "neuron" comparison, not raw computation.
Damn it, spell correction changed "qubit" to habit on me!
This is very interesting! I wouldn't dare call this a neuron tho.
The lack of detail makes it sound sensationalist.
@@NortheastGamer It's a bit funny. I think calling them neurons was pretty sensationalist in the first place... but you could also be right, so I'll give a bit of context.
Boltzmann Machines have this problems:
Networks only form simple undirected graphs with symmetric weights. That means no multiple connections between elements, no connections to themselves, and information computed has no directionality. Also elements only have two states (0 or 1).
This atomic BM has one additional BIG restriction:
Connections to a single atom are limited by the atoms one can place in its vicinity.
In contrast:
The amount of neurotransmitter released at a synapse can be modulated (so it's not really a binary thing), multiple synapses to a single other neuron are possible, there's directionality in these synapses, the geometry of neurons (and synaptic distribution) also plays a role in computation (non-linear relationships between input synapses), and more importantly in this case, a neuron can synapse not only to itself but also to thousands other neurons in distant (and very specific) locations at a time.
This is only what comes to mind rn tho, there are surely many other things that could mentioned. These atomic computation devices seem super interesting and beautiful already (not to mention how useful they could become in the future), there wasn't really a need to call them something they are not.
tl;dr The computational capacity of neurons is given by their complexity and it's only exponentially amplified by the complexity of the networks they can form, these devices are too simplistic to be called neurons imo.
Agreed neurons are far more complex then what a lot of physists in computing seem to believe.
CS folks don't try to model the physical brain down to the atom or even anything spatial. NN are just a model, roughly based on neurological structure, the activation potential of a "neuron" in code can be driven by incoming edges from "neurons" anywhere in memory so again distance doesn't really come into play... The closest thing to "distance" that we model is the weight or amount one "neuron" affects another.
Also neural networks or NN are some of the most successful AI concepts around and has had recent revival due to its effectiveness when given a ton of data and a few thousand cuda cores on modern gpus (researchers now working on reduction of training sets so the networks can train faster and learn more)
Makes sense. Much of machine learning is fuzzy gradient descent, so why not take advantage of analog processes which naturally perform similar "operations"? Back in the day we used analog electronic chips to quickly calculate things like sine and cosine. Digital methods surpassed the analog approach, so analog computation fell out of favor. Maybe it's time for a comeback.
Finally some subtitles
Please do a video about yubikey🙏
That was way too short
Please tell me where I can find papers regarding the exploitation of atomic forces with an STM. This is a fascinating topic I previously brushed off as "oh that's a fun thought experiment."
Hi there.
If you go to the blog linked in the video information and search for "scanning tunnelling microscopy" and "atomic force microscopy", you'll get links to a number of posts discussing various aspects of atomic manipulation.
Philip (speaking in video)
If it can only work in a high vacuum can it ever be practical?
Nope
Even if the atomic ML stuff doesn't work out, we still got ourselves a one-atom transistor!
Can you run COBOL on this cobalt?
Another video with Professor Moriarty, the famous scientific criminal. Excellent.
6:02 I lost it. Thank you for that laugh.
Do we know the basic principle about how the neurons inside the brain is doing calculations?
I know it's not woking the same way like a CPU but do we at least have some clues about the fundemental elements that make our brain do what it is doing?
We have no idea why it works but we can see mechanically what our brain does when it's working.
We know what neurons do down to a chemical level. We know they are like muscles, in which they get stronger/faster the more you use them. This is essentially what they are doing. The part of neuroscience we are still doing a lot of research on is understanding how the brain works as a whole. We understand it at the macroscopic scale, we understand it at the micro scale, but somewhere inbetween is where things are more theory
Did they not have a spirit level when they put those shelves up?
Is there a minimum energy state for velocity?
"At rest"
Next topic; we learned how to focus a camera.
Wouldn't bringing in an additional atom make the 2 atoms interfere with each other creating more than just binary energy states and may result in local minima if we are using the ball down a curve visualisation.
6:10 “Actually there are very very few physics questions that cannot be answered with ’Well the system wants to reach its minimum energy state.’”
The Boy is Back in Town!
Oh thank god, it's Prof. Moriarty. I needed a good dose of him 😃
How it would be connected if you for example manage to make 100 of those atoms-neurons, and can you estimate or do you know exactly what is the power consumption of those 100 if you exclude support systems around it... like vacuum pumps and cooling and all other stuff around, what would be power consumption of your 100 atom-neuron-brain-CPU call it how ever you want ?
7:33 did he just describe ppl?
Just the tip.
Is 'learn', 'experience' and 'remember' not overly anthropomorphized here? Or is there sound reasoning to build up to that position?
This seems like a huge equivocation waiting to happen.
Positronic brain? Sonny where are you?
I was thinking of R. Daneel Olivaw and his Positronic brain when I saw this video! I wonder what Isaac Asimov would make of this (and the Newton -> Palm Pilot -> Galaxy Tab / iPad) if he were still around?
Hello, prof. Moriarty, is this "the end of quantum computing"? And, is this the reason why Intel wasn't even trying to compete with the rest? If yes, I have more questions, thank you.
Can you guys make a video on how will computers be able to think with their artificial neural networks
Moriarty for president
How does one scan say millions of atoms simultaneously, say if one were to scale this approach?
I don't think you need to. With basic artificial neural networks you only measure the output neurons to get the network's "answer". Everything else is hidden unless you're doing diagnostics.
could this be a candidate for computronium?
He looks A LOT like John Travolta
Intresting.
How many synapses can a single atom physically have though, especially on a 2D surface?
My guess would be that if you needed more than, say, four, that you could start bifurcating the channels and have infinite endpoints. But very small weights.
I guess the greater the distance between atoms, the more connections you can fit in?
12:16 - This is what you came for.