A neuroscience experiment failed to build a connectome for a 6502 chip
wired.com
wired.com
It's as silly as expecting to get a good theory of artificial intelligence by studying artificial neural networks.
(I'm aware of the irony in the above statement, but stand by it earnestly. An excellent engineering artifact whose functioning we can't explain is not scientific understanding. An excellent engineering artifact whose functioning we refuse to explain is bad philosophy, too.)
Also, there exist brain areas and regions where we do in fact have a few good good models, and connectomics has the potential to help us resolve them -- see http://www.nature.com/nature/journal/v500/n7461/full/nature1...
Turns out, even where they should connections don't constrain circuits to a sufficient degree.
I guess the key lesson is -- don't rely on a single approach, because its limitations may well lead you astray. Applies to connectomics, physiology, modelling, etc.
Everyone who's paying attention understands that individual neurons have the potential for very complex, but precise, behavior.
But knowing the connections across the brain (at some resolution) is helpful for plenty of reasons. If I want to understand how areas in the brain communicate, it's immensely useful to know where they're connected, for instance. Let's say I have 200 sensors I can place in the brain wherever I want. Placing them at crucial nodes or connected areas could be tremendously useful, since we can't yet put sensors everywhere for most spatial and temporal resolutions we want.
It is as though someone declared, "Powered heavier-than-air flight will never be possible." Uh, of course they will? It's totally obvious?
Like how can any reasonable or even poorly-informed, uninvolved person who is in a hurry, possibly not see how completely obvious this is? Help me understand this. Never is a long time.
-
EDIT: case in point, for some inexplicable reason this comment, a direct question to you stochastician, in the above form (I didn't edit except to add [serious] and the last sentence) was just downvoted twice. (I take it by people who consider everything I just said to be BS.) Just to be clear none of the above is sarcastic, etc. To me - and I suppose you - it's pretty clear what will eventually be happen, and "never" is a strong claim. I completely fail to see how anyone can claim "never" when we're ahead in terms of hardware, today.
Contrast AI with, say, interplanetary space travel. We're a long way away from that too, but there's a more obvious path to getting there. We need better control of nuclear energy, more robust robotics designs, and more efficient environmental control. These are engineering challenges: for the most part, the science is already laid out.
With AI the science isn't there. We don't know which direction to move in, let alone how to move in that direction.
"If you had a near perfect hydrogen -> helium fusion engine, it'd take about 6 million tons of fuel (about the mass of the Pyramids of Egypt or 2,000 Saturn V rockets)".
Does this sound like a roadmap? "If you had a near perfect hydrogen-> helium fusion engine"... "mass of the Pyramids of Egypt"?
Contrast this with the hardware I'm talking about: by my estimation there are 1,000 clusters today, today, that can run the appropriate software if we had it. The idea that someone would call it "impossible" at a time when conventional human thinking challenges such as playing Go, are being mastered by machine using machine learning, seems crazy.
I absolutely do not agree with you that we have any roadmap for interplanetary, certainly extrasolar, travel.
Contrast this with the Human brain. We can count the neurons (100 billion or so). It uses as much energy as a 20 watt lightbulb[1] but doesn't have light-speed interconnects, unlike server rooms. The roadmap we are following, simulating neural nets on bigger and bigger hardware, is pretty much the exact thing you'd expect us to be doing. It's yielding amazing results in machine learning.
Why would anyone think that the biological, slow-firing chemical analog process these huge, super-fast digital processes replicate, will be out of reach literally "forever".
It is literally possible that not one human being will ever travel outside the solar system - not in the next 1,000 years; 5,000 years; or ever.
The idea that we will "never" simulate a hundred billion analog neurons running at 20 watts seems childishly naive. It seems like saying "no machine will ever fly - period" when machines are already doing more than enough work to do so. Come on.
Anyway I am really interested in stochastician's answer here - this is why I wrote them after they offered to answer any questions ("appy to answer questions.").
[1] 20 watts
Supercomputing instances built w/ EC2 are in the 0.25Petaflop range. Assuming that's only a fraction of their machines, maybe the major cloud providers play on the same scale - and at this point, you have to deal with communication across DCs, which might make realtime simulation utterly infeasible. That's 2 yes, and 3-5 maybes, not "over 1000 organizations".
Then you're looking at the fact that recent research indicates dendrites carry processing power as well, which would move any brain simulation well into the exascale range. And Moore's law is shakey by now.
But given all that is overcome - we seem to be doing OK at the scaling up thing - you would need a complete connectome of the brain. So far, we've succeeded doing that for roundworms. (And 6502's, as the article shows :)
The human brain is a far cry off from that.
But let's be optimists, and stipulate that is at some point done as well. We're still in the dark, because we have no idea if a fully connected brain will work without being in a specific starting state. We're nowhere even close to map the state of a brain. And at this point, we're definitely in the realm of questions where we don't even know yet if we will find an answer at all.
Want more curveballs? There's a non-zero probability that the brain can only be fully explained using QM explanations. You've just significantly shifted the scale of the simulation.
I don't know NEVER is an appropriate word, but it's not totally obvious, either. Using the word "obvious" in this context (just like the word "never") means you've reduced the problem to pop-science, where everything is either black or white.
Where are you getting this? Did you account for the fact that the average cortical neuron fires 0.16 times per second? [1] Put in terms of gigaherz, that is -- haha just kidding about gigaherz, that is 0.16 herz per neuron. Times 100 billion neurons. "Combined with the observation that 90% of neurons rarely fire" so you actually might not need to touch all 100 billion of them in a cycle.
100 billion * 0.16 gets you literally 16 gigahertz. This is four cores running at 4 gigaherz. Memory can be heavily compressed if necessary, without slipping out of the realtime budget.
People don't seem to get that these are ANALOG SIGNALS. These neurons aren't etched in silicon. Simulating that is like simulating a mechanical watch or something - with 100 billion pieces. Your guess is ridiculously too-high.
Do you know how long it takes for a signal to cross from one half of the brain to the other? 100 m/s is widely reported progation speed limit, and the average brain length is 167 mm[2] . That means you get 167 mm / 10000 mm = 1.67 milliseconds. That's long enough to cross a server room approximately 25,000 times.
(I divided 20meters by C, getting 66 ns - the time at speed of light to cross a 20 meter room, and this value goes into 1.67 ms 25,032 times.)
All of my numbers here are super-conservative, above-all the fact that I said only 1,000 server rooms were able to do this today. It's considerably higher.
We're ahead of the computational budget required, not behind it. I simply disagree with your calculation.
The rest of this paragraph is sarcastic: When mentioning that there's a non-zero chance that neurons use quantum mechanical effects (the way our etched silicon does, since gates are so small now that this has to be accounted for), you forget to also mention that there's a non-zero chance that souls are connected to brains using an as-yet unidentified organ and therefore nothing in the brain can be simulated given any amount of computational power. I mean, the chances of that aren't zero, right?
[1] "Based on the energy budget of the brain, it appears that the average cortical neuron fires around 0.16 times per second" - http://aiimpacts.org/rate-of-neuron-firing/
[2] https://www.disabled-world.com/artman/publish/brain-facts.sh...
Second, the complexity does not lie in the neurons, there one gets away with something like 100 billion neurons * 100 Flops/ GHz (Prozessor) ~ 100 Prozessors per dimension if we completely neglect the gradient.
But what we need is the dynamics of the connections. And based on that, the estimate of 100 PFlops simulation gives roughly 10^5 clock cycles per connection, which is not very much if you compare it to the latency of a standard database in the microsecond range. (Network latencies are in a similar order of magnitude.)
About your argument, that it would be enough to only simulate firing neurons, well we have to simulate a neuron to detect if it is firing.
For completeness, Kurzweil mentions in The Singularity is near much lower estimates for functional simulation of a brain in the TFLOPS range, however that is with optimized algorithms and we probably need a singularity first to get these optimized algorithms.
This page I linked before , https://www.disabled-world.com/artman/publish/brain-facts.sh... , says "The brain has a processing capacity of 0.1 quadrillion instructions per second" (though that phrasing is bizarre), saying that the cluster Roadrunner at 1.06 petaflops did 10x more computation.
Roadrunner was built in 2008.
1) Neurons are independent spatially and temporally.
2) Dendrites are not computationally active, again false.
3) State is not important.
4) Neurons are used to simulate a binary function and each only one.
For me the most authoritative answer is what the Human Brain Project in europe is able to achieve: http://www.nature.com/news/fragment-of-rat-brain-simulated-i...
31k cells with 37M synapses. That's a start, but pretty far from a mouse brain (100M cells) or a human brain (80B? ).
Also if Moore's law really is dead, this all gets much more depressing.
You mention membrane dynamics, past spiking history, etc - but neural nets accomplish a lot without these things.
A simplification surely works. You are not thinking about simulating a brain using a large cluster with a lot of back and forth.
Why not? Don't you get how incredibly slow 100 m/s is? This is 7 orders of magnitude difference in speed. Silicon is seven orders of magnitude faster at signal propagation than the brain, which isn't electrical.
To you this doesn't unlock vast possibilities for simulating the analog in real time? It just doesn't seem that permanently impossible, given the results you are already claiming are occurring.
Also although I directly addressed you, I guess you don't have any answer as to why someone would choose a target date like "never" or "300 years" as opposed to some reasonable extrapolation of the results being achieved today.
It looks like many analyses were performed by binning transistor switching as you might spikes from neurons, or by linearly combining transistor activity across transistors or time.
This is slightly puzzling to me, because transistors would appear to not work on the basis of average activity at their level of computational composition. I would be surprised if you could really understand circuit activity at a computational level with most of the measurements you describe.
On the other hand, I suspect the methods used work slightly better in a real brain. Of course, the computational purpose of many neurons is difficult to parse or even construct useful hypotheses about. But at the very least, neurons in the outer periphery seem to behave in a way that is at least consistent with firing rate, e.g. impulses at the neuromuscular junction to cause muscle contraction, and retinal ganglion cells to fire at higher rates when the input matches their certain luminance features.
For the next paper, though, we're taking a more careful look at a lot of these encoding questions, as well as trying to use processors with a greater degree of functional specialization so we can more accurately capture what's going on.
For the processor I can see the properties of individual transistors being (near) constant, but do you have a reason to think so for neurons in the brain as well?
It's basically saying that even IF we had a full connectome of the brain that we could run in a simulator (and presumably produce a simulated personality) that we'd be no closer to actually understanding how the damn thing works.
The question I would like to see answered is what happens if we do this to someone with a grown brain and wake them up in a simulation of a space ship that traveled to a far away planet. Again presumably, the person would need nutrition and care and education for some time, but then would they be able to carry out significant advances on behalf of humanity. Would they even want to? What if they feel betrayed by their fate and creator? Then again, who doesn't? Someone write a book about this.
You need the morphology and structures that happen in nature to get the structures and functionality needed to properly simulate a brain. We simply don't know about enough about learning, development, brains, and consciousness to just plug into a simulated adult brain and make it all work. We can only follow the path nature has laid out.
I'm sure you can skip some parts like fetal stages, but we develop language at a very specific range of ages using specific techniques. We can't skip the language acquisition stage and then expect our simulated brain to magically know language. We can't skip the angry "me" stage of a toddler and expect the mind to have self-assertiveness. We can't skip the sexual and social awakening stage of teenhood and expect the mind to understand complex social concepts and sexuality, etc, etc.
The only scenario I can imagine out of this problem is a per neuron copy of a working brain on the cellular level. I imagine this isn't, or won't ever be, in the cards for practical reasons. Or if you could someone divine the algorithms that create consciousness, assuming this is even possible on a transistor based machine, then that would also be a work-around. Right now we can't do either, so following nature's approach is the probably the sanest way forward.
I wouldnt be surprised if the first AI ends up being more a biological 'computer' than anything having to do with x86 instruction sets and transistors, but that's just a guess on my part.
Less flippantly, I've never heard a credible argument why a working simulation of an actual sampled human brain, with appropriate sensory inputs, would not function just as that human did. (Note that the magic is in the phrase 'working simulation' - this may require more than just a connectome depending on how much crazy molecular-level processing actually takes place inside real neurons, but assuming this is possible...)
There was a news article a couple of days ago about some company in Japan "simulating a human brain" which disappointingly was actually just them simulating a randomly connected neural net with a similar number of neurons. That's not a 'simulated human brain', any more than a skip bin full of transistors is a 'simulated CPU'.
I think the Wired title is a little overblown. I work in a connectomics lab (software, I'm not a scientist yet so take what I say with a grain of salt), and we've been having enjoyable debates about the paper. It's hardly a "war".
What's kind of interesting about the field of connectomics is that the engineering challenges have been massive enough that it's been hard for many labs to do science with their datasets until recently. Several papers have come out (e.g. a fly visual circuit, two dual mouse visual circuits, a few others), but they generally are findings on relatively small volumes. Thanks to a big grant that created a multi-lab consortium, we are getting closer to doing science with a big volume. We're one year into a five year process which will result in a petabyte size open dataset for the world to use.
Because so few datasets have been adequately reconstructed, and most labs have been obsessed with the reconstruction process, I think in the future, as we do more science, we'll discover better ways to glean insight from the neuroanatomy and physiology traces. It's a new field, so I don't think it's unreasonable to expect that there are new methods to learn. I'm happy to talk about what goes into the reconstruction process. :)
Other labs at different phases of the pipeline are working with various kinds of physiological indicators including taking movies of GCaMP (https://en.wikipedia.org/wiki/GCaMP) (a calcium activity indicator) expressing neurons under various behavioral conditions. A glut of calcium ions provide evidence for recent firing of a neuron as calcium is involved in triggering the release of neurotransmitters.
If you're interested in methods, here's a paper describing a dataset we focused pretty heavily on (that's not related to the consortium dataset, but paved the way with developing techniques): Briggman, et al. "Wiring specificity in the direction-selectivity circuit of the retina" (http://www.nature.com/nature/journal/v471/n7337/full/nature0...)
We later did some science with it ;)
JS Kim, et al. "Space–time wiring specificity supports direction selectivity in the retina" http://www.nature.com/nature/journal/v509/n7500/full/nature1...
The pipeline has evolved considerably since those papers came out. We're living in exciting times. :)
To use a set of techniques to investigate the operation of a device whose function we fully understand, and still fail to draw meaningful conclusions, raises serious questions about the usefulness of those same techniques in a much more complicated and less-well-understood domain.
Most of what you mention is sort of negative information - what happens when the brain breaks. How it works is a fundamentally different question, and I suspect we'll only get real understanding by building computational models that are able to do similar computations to those the brain does.
Which is not a surprise. Imagine someone hands you an encapsulated CPU and you don't know what a clock is, what external memory is, or what an instruction set is.
How long will it take you to understand what it does and how it works?
Now let's say you know what a gate is, and you take the top off and reverse engineer the circuit. Does that help?
IMO, not as much as we'd all like. For a CPU the important information isn't in the connectome, it's in the design manual.
It's a fair bet the brain is rather more complicated. And there's no manual.
Having a crude map of some top-level systems is a good start, but I'm surprised that there seems to have been a lot of labelling and identification, but comparatively little interest in investigating how information is passed between the subsystems and how they function together.
I think we probably know a lot less than laymen assume. Its a big conversation to define "what we know" but "a lot" assumes at theoretical and experimental groundwork that most likely doesn't exist. I find the research I've been exposed to is a mile wide but an inch deep.
If we had more depth, we'd be hobby coding AI because we'd understand the basic concepts the same way a even jr developer can do enough research to hobby code a complex and featured 3D game engine, something a coder from the 60 or 70s couldn't do. We simply don't have these kinds of understandings.
This is the point. There are multiple levels of interpretation layered on top of the physical process. There is no reason to believe that they are tidy or hierarchical either, abstractions and indirections can leak left and right (or top to bottom). We lack the language and tools to describe these structures laid bare, and words like "abstraction" and "indirection" are probably a bad fit. Even "interpretation" might prove to be conceptually misleading when cracking this nut, so to speak.
That being said disabling neurons sounds a little crude and may not help much in gaining a finer understanding.
Then again I know nothing about neuroscience :)
Yes it's crude, but artificially creating 'lesions' and investigating brain disfunction is a time-honored tradition in neuroscience. It's usually done with rats/cats/monkeys, but humans with brain damage due to strokes or surgeries also serve as useful 'natural experiments.' HM is a classic example [0].
When working with something as complicated with the brain, taking out pieces and seeing what changes is a reasonable way to investigate its operation...depending on how you feel about utilizing animals in this way.
So, also not a neuroscientist. So i think the general point is, their current approach is to turn off neurons and see how things change. with careful evaluation, you should be able to pin down the role of that specific neuron. With a chip, it seems like you'd want to investigate the behavior when output when output is the same as the input - just a pass through. maybe you can deduce the effect of that specific gate.
My (poor) understanding is the standard tools weren't enough to identify stuff like ripple adders (or whatever's on the chip) so maybe they're using arithmetic when they need calculus.
Same problem with the binary functions like most kinds of digital electronics - they're typically combined in highly nonlinear ways.
However, knocking out a whole IC and seeing what it did is more reasonable and about the level we are at when understanding functions of subsystems in the brain.
Of course, nobody really knows how much indirection is used by the brain, but it's probably closer to a direct hardware implementation than a CPU. One respect in which it differs from both, of course, is plasticity; neither hardware nor software created by humans is given to constant self-modification.
Of course it is. A VM running on a CPU is basically just constantly self modifying data with the help of modular hardware. I'd hypothesize, the brain has inert structures at it's core, too (which is why we are all alike), and the rest is the data of a state machine (an IC being part of the rest, from a phenomenological point of view). Of course the complexity might be beyond a state machine or linear bounded automata, i.e. deterministic (think quantum).
The tools of a neuroscientist would be laughable in comparison, yet neuroscientists have cracked the auditory code (and you can have an artificial cochlea now), the lower visual system is close to be cracked, we have a pretty good understanding of hippocampus and how space is represented in the brain, and many other accomplishment. These are all results from the laughable tools we have for investigating brain.
Having a full connectome of the brain, and running the brain in a simulator will be a huge step of understanding it. We don't know how to use the data now simply because we don't have the data yet, and therefore no effort is putting on interpreting the data. However the usefulness can be glimpsed from neural structures where the connections are clear: the peripheral system, spinal cord, midbrain, and lower sensory and motor regions in the brain. We understand them far better than regions we don't even know where it connects: claustrum comes into mind.
A simulator of the brain, I imagine, will be similar to the human genome project: nobody will understand the whole thing quickly, but it will hugely prop neuroscience forward, sometimes in ways we cannot imagine now.
The inputs and outputs for a brain/neural network, e.g. pictures as inputs as recognized object as outputs, are clear. They are not clear for a chip-restricted view where your input binary stream is a post-transformed signal from an Atari joystick, and the output is a pretransformed signal to a display where the pong paddle moves. These complex transformed inputs and outputs at the chip-restricted perspective tremendously reduce the likelihood of making sense of anything.
If we know the adaptor functions for the signal to the first layer of neurons in a simple convolutional neural network, for example, and the output layer is indentifiable, it might be easier to track which neurons track which features, etc.
But there might be no clear intermediate output layers to a brain, so you might have a look at inputs and outputs wrt various subnetworks to see which is most likely to be the true functional subset corresponding to some definitely-occuring function (another obstacle to define). Or assume a function is occuring and see which neurons classify it under the most circumstances. But with so many connections that sounds like a big rough game to play too.
I can see how this project if successful would be another step to creation of the sentient robot overlords but I'm curious as to what people see as the more immediate medical applications would be. Some of the biggest breakthroughs in brain health in the last 20 years have been about getting rid of clots (current best is arguably manual labor) and immune modulation and new biochemical drugs whose effects would require excruciatingly sophisticated and accurate biomolecular connectomic models to model and may be a long time away.