10 ms·
It's nice that someone now has the neural wiring diagram for part of a mouse brain. But we have the full wiring for a thousand-cell nematode, and OpenWorm still
by Animats 7y ago
It's nice that someone now has the neural wiring diagram for part of a mouse brain. But we have the full wiring for a thousand-cell nematode, and OpenWorm still doesn't work very well.[1] OpenWorm is trying to get a neuron and cell level simulator for what's close to the simplest creature with a functioning nervous system - 302 neurons and 85 muscle cells. That needs to work before moving to more complexity.
[1] http://openworm.org/ http://openworm.org/
- jimfleming 7y ago> That needs to work before moving to more complexity. It really depends on what level of abstraction you care to simulate. OpenWorm is working at the physics and cellular level, far below the concept level as in most deep learning research looking to apply neuroscience discoveries, for example. It’s likely easier to get the concepts of a functional nematode model working or a functional model of memory, attention, or consciousness than a full cellular model of these. More specifically, a thousand cells sounds small in comparison to a thousand layer ResNet with millions of functional units but the mechanics of those cells are significantly more complex than a ReLU unit. Yet the simple ReLU units are functionally very useful and can do much more complex things that we still can’t simulate with spiking neurons. The concepts of receptive fields, cortical columns, local inhibition, winner-take-all, functional modules and how they communicate / are organized may all be relevant and applicable learnings from mapping an organism even if we can’t fully simulate every detail.
- Reelin 7y agoThe trouble is that (assuming sufficient computational power) if we can't simulate it then we don't really understand it. It's one thing to say "that's computationally intractable", but entirely another to say "for some reason our computationally tractable model doesn't work, and we don't know why". Present day ANNs may well be inspired by biological systems but (as you noted) they're not even remotely similar in practice. The reality is that for a biological system the wiring diagram is just the tip of the iceberg - there's lots of other significant chemical things going on under the hood. I don't mean to detract from the usefulness of present day ML, just to agree with and elaborate on the original point that was raised (ie that "we have a neural wiring diagram" doesn't actually mean that we have a complete schematic).
- staticassertion 7y agoBut we understand tons of things without simulating them.
- Reelin 7y agoIf you think you understand something, write a simulation which you expect to work based on that understanding, and it doesn't work - did you really understand it?
- staticassertion 7y agoOK, what if the simulation works, did you understand it before?
- Reelin 7y agoNot necessarily. A working simulation (for some testable subset of states) doesn't carry any hard and fast logical implications about your understanding. On the other hand, assuming no errors in implementation then a broken simulation which you had expected to work directly implies that your understanding is flawed.
- burnte 7y agoMaybe, maybe your simulation is just buggy. I can write a simulator of how my wife would react to the news I'm cheating on her, and fail miserably, but I'm quite positive I understand how she would actually react.
- Reelin 7y agoYes, you have to debug your code. I suspect that the people who implemented OpenWorm are capable of and have done that.
- gogoincar 7y agoCan you give some examples? I'm guessing there is a different in definition of understanding here. As I interpret GP, the claim is you can't describe something in sufficient detail to simulate it, then you don't actually understand it. You may have a higher-order model that generally holds, or holds given some constraints, but that's more of a "what" understanding rather than the higher-bar of "why".
- __s 7y agoA good comparison exists with emulators, where transistor level emulation is ill advised for most hardware
- Animats 7y agoIt really depends on what level of abstraction you care to simulate. The article starts out "At the Allen Institute for Brain Science in Seattle, a large-scale effort is underway to understand how the 86 billion neurons in the human brain are connected. The aim is to produce a map of all the connections: the connectome. Scientists at the Institute are now reconstructing one cubic millimeter of a mouse brain, the most complex ever reconstructed." So the article is about starting with the wiring diagram and working up. My point is that, even where we already have the wiring diagram for an biological neural system, simulating what it does is just barely starting to work.
- jes5199 7y agoI recently saw this video of living neurons: https://www.youtube.com/watch?v=2TIK9oXc5Wo https://www.youtube.com/watch?v=2TIK9oXc5Wo (I don't actually know the original source of this) and just looking at the way they dance around - they're in motion, they're changing their connections, they're changing shape - is so entirely unlike the software idea of a neural network that it makes me really doubt that we're even remotely on the right track with AI research
- jacobush 7y agoAmazing video! And these poor neurons are squished between glass, imagine them crawling around in a 3D space.
- ramraj07 7y agoI'm planning to work on a project like this at some point soon (worm biology is so cheap you can do it in your garage for the price of a Mercedes). The main roadblock I want to work on is to get more connection strength measurements - we already have the full wiring diagram for the worm connectome, but it's not obvious that the connections are all equal strength, they definitely are not. Many labs are trying to image the neurons firing realtime at the whole organism level, but they're stymied by the head which has a hundred or so neurons in very close proximity (and they typically fire at rates faster than 3D imaging modalities can keep up). I'm definitely excited to start working on this in a couple of years! My hopeful guess is that observing the full nervous system while it's firing full throttle is the only way to start understanding the algorithms that are running there. And from there hopefully we can start finding patterns! Needless to say I agree with you. The people who say they have a wiring diagram of the mouse brain need to reign in their enthusiasm. We are not anywhere close to start understanding even a fly or zebrafish brain leave alone a mouse or human one. Sydney Brenner himself agreed (though he's arguably biased towards nematodes).
- mapcars 7y ago>worm biology is so cheap you can do it in your garage for the price of a Mercedes I don't think you understand the depths of what a worm is. Is there any worm created by humans without using previous worm material?
- galileo5 7y agoNothing to fear even though that tiny little calculator can calculate whats 123456789 x 34556 faster than any of us dodos
- galileo5 7y agoIt is not 'If' but a matter of 'When'
- stagas 7y agoThe numbers seem to suggest that it's a simple mechanic - hey just 302 neurons - but a single neuron cell is immensely complex, containing millions of molecules, trillions of atoms that all interact with each other in unknown, unpredictable or even unobservable ways. Even if we had all that data in a model, the biggest problem is that our computations, unlike nature's, are done serially, meaning we only get to work at one bit at a time whereas nature is computing all of the interactions in parallel. If you've ever done a physics collision system you'd know how the performance starts to degrade rapidly with just very few elements (being an O(n^2) problem) and you need to make workarounds. So we need different hardware to start with, like analog computers, if we'd ever have a chance to simulate a single living cell (for starters), then move on to larger organisms.
- stagas 7y agoExactly, it's a kind of hubris that people seem to claim knowledge about the brain just by observing vague electrical signals from afar. The delusion comes from the fact that we seem to know a lot about the heart, the kidney, the bones, so why not the brain? Well, the brain's cells are communicating to each other in a network with trillions of connections, that's a major difference that seems to be ignored, and the other is that the each cell contains a nucleic acid code beyond our scale of understanding which runs all the time while those electrical signals pass through deciding its next messages and actions. If you ignore these two major bottlenecks, then you could be tricked into claiming knowledge, but it will be limited and, most of the time, wrong.
- nil-sec 7y agoThe worms are very special. Their nervous systems are weird because they have been heavily optimized to be small. A huge push to get the full connectome of Drosophila is coming to an end right now, we are not there yet fully, but close. This research has already elucidated many things about how their brains work. People are discussing how to do the mouse now. In conjunction with functional experiments this research already explained many pathways such as olfactory associative learning, mating behavior and many others. None of these understandings came from simulations, but from multiple experiments and study of selected subcircuits. There were also successes in simulation of the visual pathway of drosophila, based on the connectomic data. For example this study [1] was able to reproduce direction sensitivity of a specific set of neurons. It isn’t necessarily true that we need to fully understand the worms before we should and can move on to more complex nervous systems and successfully understand them. It might just be that neuron abstractions don’t work well in worms, because of the mentioned evolutionary pressure to optimize for size. [1] https://arxiv.org/abs/1806.04793 https://arxiv.org/abs/1806.04793
- magicalhippo 7y agoA computer analogy: it's usually far more straight forward to reverse engineer a regular C program, than one of those tiny hand-optimized assembler demos. For a concrete example, consider the "modbyte tuning"[1] in the 32byte game of life. [1]: http://www.sizecoding.org/wiki/Game_of_Life_32b#Modbyte_tuning.2C_jumping_into_modbytes.2C_code_path_alignment_:_32_bytes http://www.sizecoding.org/wiki/Game_of_Life_32b#Modbyte_tuni...
- GuB-42 7y agoLife is different. There is no logic to the way it solves problems. A programmer writing a game of life in C will probably do it from scratch, in a straightforward manner. Read the corresponding machine code and you are going to see familiar structures: loops, coordinates, etc... Now here is how life does it: you have a program that outputs prime numbers, you then have to change it into a sorting algorithm, then in a fizzbuzz, and then in a game of life. You are not allowed to refactor or do anything from scratch. If the program become too big, you are allowed to remove random instructions.