Neuroscience Can't Explain How a Microprocessor Works (economist.com) 169
mspohr writes: The Economist has an interesting story about two neuroscientists/engineers -- Eric Jonas of the University of California, Berkeley, and Konrad Kording of Northwestern University, in Chicago -- who decided to test the methods of neuroscience using a 6502 processor. Their results are published in the PLOS Computational Biology journal. Neuroscientists explore how the brain works by looking at damaged brains and monitoring inputs and outputs to try to infer intermediate processing. They did the same with the 6502 processor which was used in early Atari, Apple and Commodore computers. What they discovered was that these methods were sorely lacking in that they often pointed in the wrong direction and missed important processing steps.
Massive failure from all involved (Score:4, Interesting)
Anyone with even an elementary education in cognitive science will tell you that attempting to model thought processes is always done according to the dominant technology of the time in question. First it was machinery, then it was circuits, then it was computers.
This does not mean the model is accurate or even useful.
Re:Massive failure from all involved (Score:5, Insightful)
This isn't so much about modeling thought processes as it is about illustrating how even in a simplified model one of our debugging approaches fails.
The logic that they're arguing appears to be:
"If we can't even properly reverse engineer an extremely simple deterministic computer chip using fault modeling, it's extremely unlikely that we can infer the mechanisms of an extremely complex non-deterministic processor like the brain."
Re:Massive failure from all involved (Score:5, Insightful)
"If we can't even properly reverse engineer an extremely simple deterministic computer chip using fault modeling, it's extremely unlikely that we can infer the mechanisms of an extremely complex non-deterministic processor like the brain."
But that logic only makes sense if microprocessors and brains were similar enough that comparable methods could be used to attempt to understand them. But that isn't true. That is like saying you can't understand how to plow a field with a horse if you don't understand how a tractor engine works. Although horses and tractors have some similarities, understanding how one works doesn't really help you with the other.
Re: (Score:2)
I don't want to enter the high-level debate here - I'm not qualified (and that's not sarcasm) - and I do know that this example doesn't really mean anything, or add to the debate, but:
Watch this:
http://www.visual6502.org/JSSi... [visual6502.org]
then watch this:
https://www.youtube.com/watch?... [youtube.com]
and think about them for a minute. It never fails to make me stop and wonder.
Re:Massive failure from all involved (Score:5, Insightful)
The point of the argument is to challenge the implicit assumption that current neuroscience methods work as well as people think they do. If you just assume your research methods work, you are resting on blind faith in your methods. One step in showing the need to challenge those foundational assumptions is to use this example to //illustrate// how then can fail. Using microprocessors allows is the luxury of total knowledge as to what we are investigating, at the expense of being quite different to the brain. The quoted bit needs fixing:
"If we can't even properly reverse engineer an extremely simple deterministic computer chip using fault modeling, it's extremely unlikely that the same fault modelling will work reliably with something extremely complex like the brain."
It does not show whether or not 'fault modelling' works or not for the brain, but gives good justification for the claim that we cannot take the efficacy of 'fault modelling' for granted when studying the brain.
Re: (Score:2)
Indeed, and I vaguely gather there's the notion that "science" actually starts with "thinking about thinking"
ie.
asking not just, how do we know?
but asking, how do we know whether, how we know, really allows us to know?
or to be less wordy,
why do we trust this method?
Re: (Score:2)
Using computer chips as test subjects does not validate or invalidate research methods used in neuroscience. What validates or invalidates the methods and the results of the methods is how well it predicts human behavior. That's all that counts. Do these results provide insights or not. Studying a different subject matter does not get you any closer to understanding how the human brain functio
Re: (Score:3)
No the point of the article was "questions whether more information is the same thing as more understanding."
No, that was not the point of the article at all. The point of the article was that there is an implicit assumption in the field that we just lack sufficient data. That the methodologies used to analyze that data are fine, but because we don't have enough data, we fail to successfully understand cognition. The authors argue that, no, there is not enough but data, but also that the methodologies are flawed; that the methodologies themselves need to be validated. But because we don't have a ground truth with
Re: (Score:2)
If they were looking at calcium channel opening, then I'd agree with you. They appear to be looking at things from a much more abstract level. And their results aren't proof, but certainly raise reasonable questions.
Re: (Score:2)
Re: (Score:2)
But that logic only makes sense if microprocessors and brains were similar enough that comparable methods could be used to attempt to understand them. But that isn't true.
Actually, they are arguing that it is true. From the article,
"Obviously the brain is not a processor, and a tremendous amount of effort and time have been spent characterizing these differences over the past century [22, 23, 59]. Neural systems are analog and and biophysically complex, they operate at temporal scales vastly slower than this classical processor but with far greater parallelism than is available in state of the art processors. Typical neurons also have several orders of magnitude more inputs
Re: (Score:2)
More to the point: "Gaël Varoquaux, a machine-learning specialist at the Institute for Research in Computer Science and Automation, in France, says that the 6502 in particular is about as different from a brain as it could be."
You knowledge and understanding of brain research is very primitive.
Re: (Score:2)
No, they argue that the 6502 (or just a microprocessor) is an acceptable model for validating the approaches used in neuroscience to analyze complex data sets, which is exactly what I said in my comment above. In other words, if they can successfully determine the ground truth of the microprocessor using those approaches and with limited a priori knowledge, then the methodologies have potential. Otherwise, they need to be refined until they are able to do this. Validating against an imperfect model is bette
These methods were hoped to be useful in general (Score:2)
But that logic only makes sense if microprocessors and brains were similar enough that comparable methods could be used to attempt to understand them. But that isn't true.
Yes, there's no guarantee that the methods that work on the brain also have to work on microprocessors. However, there's also no guarantee that they won't work in both cases. There are many methods/observations that are so generally useful that they apply to a huge range of problems. This is important because there's no guarantee that methods that work on one part of the brain also work on another part of the brain. Maybe the part of the brain responsible for facial recognition and the part of the brain tha
Re: (Score:2)
Re:Massive failure from all involved (Score:4, Insightful)
Or, for that matter, why alt-right trolls are such stupid bigots?
Neuroscience can't, but eugenics can. Eugenics can explain anything. There are some thing neuroscience can't explain.
That's why neuroscience is science but eugenics is not.
Re: (Score:1)
Eugenics is a science. Science can explain anything, so can logical reasoning, so can emotional reasoning. It does not make one correct or incorrect.
Re: (Score:2)
I agree.
If your model fails to predict an event, your model is faulty. Full stop.
So whatever methods they used, aren't enough to capture the 6502. A random number generator given an infinitely long time, would build a 6502 eventually. So the point here is not "It cannot be done." It's simply that "Given the methods we tried--which may be ALL the ones available to us in 2017--we couldn't do it." But we couldn't do it with our tools != nobody could ever do it with newer tools.
Re: (Score:2)
What they need is super fine granularity on an fMRI, which they will have to wait for...
OR
Use a normal fMRI on a HUUUUGE brain!
If only we could find one of those...
Re: (Score:3)
[...] an extremely complex non-deterministic processor [...]
[citation needed]
Since it's my job to put the nondeterministic stuff into your CPUs, I don't need no stinking citation.
The top three source of non determinism.
A) RNGs
B) Asynchronous interfaces
C) PLLs
If your computer is a phone or otherwise has a wireless interface, the second largest source of non determinism is the antenna.
Re: (Score:2)
I'm the systems guy who loves to design that non-determinism in eg to avoid accidental deadly embraces, and to give bad actors a harder time, so thank you!
And yes, for embedded devices, I'd put sensor least-significant bits and jitter between different clock sources high on my list of genuine entropy sources, and I'll count radio (eg RSSI measurements) in the first category.
Rgds
Damon
Re: (Score:2)
Re: (Score:2)
>In computers, "random number generators" are often only pseudo-random, and are in fact deterministic.
Unless they are the ones commonly found in every modern CPU, which include an entropy source.
E.G:
Intel : http://www.rambus.com/wp-conte... [rambus.com]
VIA : http://www.rambus.com/wp-conte... [rambus.com]
Many Arm Socs: https://community.arm.com/mana... [arm.com]
Re: (Score:2)
That is, observing your own thoughts isn't perfect, but it can give you a ton of data if you're willing to look.
Re: (Score:2)
except this isn't about any given thought, or emotion. this isn't about muscle feedback loops and controls.
This is about how each neuron fires and why does it fire in that order.
We can make a logic gate, but the brain doesn't use logic gates yet it still gets the correct answer. (sometimes) how it does that is the biggest mystery of neuro science. In the brain memory and processors are one and the same.
What could a computer do if you gave it 32 gigabytes of level 3 cache? what if you gave it 1 terabyte
Re: (Score:2)
Re: (Score:2)
Computer main memory, even SSD drives, are faster than human memory.
The elementary parts that make up a brain can be emulated by digital logic. The connections and their changes can be emulated by digital logic. The appropriate question is not can it be done, but how can it be done and is it practical to do it?
Fundamentally flawed logic (Score:2)
There a fundamental flaw.
Brain are extremely parallel and highly distributed processing units.
Some region are more specialised in some tasks, but as a whole, no part of the brain absolutely needs another part for the brain to keep working.
From that perspective, CPU are a small single function device. They either work, or not. It's hard to have a *half functionning" CPU (unless you very specifically manage to burn a peculier par of the silicon that isn't core to the functionning. I don't see how that would b
Re: (Score:3)
Indeed (Score:2)
Yup, have never had experience coding for 6502. (Only from 8088/8086 up)
Just noticed now that it lacks multiplication/division instruction (and thus probably lacks microcode to do them as a series of addition/substraction and shifts).
Thank for correcting me.
Re: (Score:2)
I agree. I mean, an 6502 is a pretty simple piece of electronics and a description of the complete functionality and instruction set can be done on 20-30 pages or so. In addition, it is completely deterministic and has a very small internal state (around 8 bytes). If you cannot model that, then forget about modeling more than a single neuron or a very small cluster of neurons.
Re: (Score:2)
This isn't so much about modeling thought processes as it is about illustrating how even in a simplified model one of our debugging approaches fails.
The logic that they're arguing appears to be:
"If we can't even properly reverse engineer an extremely simple deterministic computer chip using fault modeling, it's extremely unlikely that we can infer the mechanisms of an extremely complex non-deterministic processor like the brain."
I do wonder at what level the reverse engineering is done. Also I wonder if their method was pure enough to initially consider the 6502 to be analog rather than digital. That would be a nice trip down the garden path right from the get go.
Now I would say that many fields of study at the higher levels, such as economics, medicine, etc. etc., are incomplete. There's a lot still to be learned. And taking a sidestep of looking at an artificial "brain" from a neuroscience perspective is a good way to navel gaze,
Re: (Score:1)
Ironically it is possible to reverse engineer the brain. Once you begin to understand its core algorithms at heart the brain is not even a very complex machine. The secret to the mind is abstraction and generic logic and the Turing machine - the key to all those is computer and CPU engineering. That's the irony.
Re:Massive failure from all involved (Score:4, Interesting)
Exactly, I have a degree in cognitive science and this is what we are taught. So much of the language of computers has crept into psychology it's unbelievable. And most of it is wrong and misleading. Hundred years ago the personality was being modelled in hydraulic terms (the new cool tech of the age) and even physical models were made. All wrong of course.
Re: (Score:2)
FTA: "Gaël Varoquaux, a machine-learning specialist at the Institute for Research in Computer Science and Automation, in France, says that the 6502 in particular is about as different from a brain as it could be. Such primitive chips process information sequentially. Brains (and modern microprocessors) juggle many computations at once. And he points out that, for all its limitations, neuroscience has made real progress. The ins-and-outs of
Modern (pseudo)-"Science" (Score:5, Insightful)
In order to understand the DNA of an Orange, we "scientists" dissected an alarm clock. This _proved_ that our methods of studing oranges, and fruit in general, have been wrong for centuries.
Re:Modern (pseudo)-"Science" (Score:5, Insightful)
I see a lesson in humility here by looking at how poor human scientists do at modelling-by-studying-defects in a general sense.
It suggests that models of the brain derived by seeing what effects damaged sections have on patient behavior may be worse than originally expected.
Re: (Score:2)
I see a lesson in humility here by looking at how poor human scientists do at modelling-by-studying-defects in a general sense.
It suggests that models of the brain derived by seeing what effects damaged sections have on patient behavior may be worse than originally expected.
But just like any science, that's not the only thing they do. They compare different methods and models and come to a consensus. If you see that people who have damage to region X of the brain can't do Y and you see that region X of the brain is active in healthy people when they do Y then you have two points of data that point to the same conclusion. They do the same thing with carbon dating, quantum physics, gravity waves, etc... As long as the different measurements all agree then you assume that you
Re: (Score:1)
That shows that X has some relationship to Y. But the researchers were caught over-interpreting this with descriptions such as "X controls Y" in the chip experiment.
Re: (Score:2)
In the case of the orange, this method isn't wrong [wikipedia.org]...
Re: (Score:1)
Now I'm annoyed the editors rejected my submission "Carpentry Can't Explain How a Poem Works".
Intelligent design (Score:5, Funny)
I know I'll catch hell for my religious beliefs, but...
I think that the 6502 was not the result of evolution, but rather it had a Creator and was the product of Intelligent Design. There are just so many subtle clues that suggest features that were deliberately put in there. Could natural selection really explain how it had two different indirect access modes, one that selects a direct index from an offset, and the other adds the offset to the index?
These researchers may be trying to apply the wrong methods to a device that is almost certainly the product of a higher power.
Re: (Score:1)
Don't be silly. The 6502 evolved from earlier 4-bit microprocessors. This is clear because "evolved" now refers to anything whatsoever where B follows A.
If still unconvinced, adjust your confirmation bias upward until you're blissfully avoiding risk of academic crimethink.
Re: (Score:3, Funny)
Re: (Score:1)
Because, obviously, there were only a few necessary mutations between 4-bit and 8-bit processors, and, naturally, each transitional random rearrangement of the transistors met the constraint of full functionality of the processors at each progressive step.
We only acknowledge one meaning of "IC" here.
Re: (Score:2)
Re: (Score:2)
Re:Intelligent design (Score:5, Funny)
And the great and powerful Woz spake thusly: Let the Intel become the brain of my new creation! But lo! The Book of Jobs decreed the creation be cost effective and priced by the Number of the Beast. And so out of the land of Silicon came Forth the 6502 to eat from the Apple tree. Eight shall be the number of bits, no more and no less.
Re:Intelligent design (Score:5, Insightful)
All jokes asside, I think the point here was that both devices (6502 or fatty-thinkmeats) were modeled as a black box. I'd be willing to be that a significant fraction of the neuroscientist population would argue for a 6502 being the simpler system, so the blackbox approach should (one would hope) be able to model that device more easily. If they find that their blackbox approach to understanding a 6502 leads to incorrect results, then it raises questions as to the effectiveness of the approach on the thinkmeats.
Re: (Score:1)
Re: (Score:1)
These researchers may be trying to apply the wrong methods to a device that is almost certainly the product of a higher power.
That may well be the case, but if so, it's also quite clear that the higher power used evolution and natural selection as his development tool.
If human brains had just been magic'd into existence by divine fiat, there would be no reason for them to look like a specialized version of the brains of earlier hominids (which in turn look like specialized versions of the brains of earlier mammals, and so on for as far back as you care to look).
Re:Intelligent design (Score:5, Funny)
Obligatory Princess Bride quote:
"Truly, you have a dizzying intellect."
Is there any other reason? (Score:1)
Re: (Score:2)
Re: (Score:2)
i'm sure there are people who believe that, but i don't particularly want to meet them.
or hear them praying.
Re: (Score:1)
Flame war on, it was not Intel Designed it was designed by MOS Technology!!!
Re: (Score:1)
And yet there were clearly indications of evolution at work. Subsequent generations, including the 65C02 and 65C816, clearly had not only new instructions that simply didn't exist in the early generation processors, but also expanded addressing and ever faster speeds.
Re: (Score:2)
the 6502 was not the result of evolution, but rather it had a Creator and was the product of Intelligent Design
And proof there is in the name: was created 6502 years ago
Re: (Score:2)
You forgot:
Checkmate, atheists!
Re: (Score:2)
Actually natural selection *can* explain how the 6502 had two different indirect access modes.
The PDP-11 (one of the great ancestor computers) had two different indirect access modes (6n and 7n). The computer eco system flourished and spawned many different types of computer chips, one of those which was the 6800 which shared the instruction set traits from that line. However, later, the computer eco system got more price competitive from descendants from other computer chip lines. This put evolutionary
Comment removed (Score:5, Interesting)
Re: (Score:2)
>The take-away point I get from this is that we may need another revolutionary technology or two (fully three-dimensional integrated circuits? IC's based on carbon instead of silicon?) before we can model the sentient mind as similar to an artificially created device
Memristors already exist and are going to revolutionize the computing world by combining processing with storage (and eliminating the difference between RAM and long term storage). If somebody knows if that will take 5 or 50 years to get out
Not completely true (Score:2)
Re: (Score:2)
You're close to a contradiction between your two statements. If you don't know how the brain works, how do you know that connecting a billion neurons won't make a brain?
Re: (Score:1)
It's been almost 50 years since memristors were proposed (1971) and over 5 years since they were produced in a lab (they were around before I finished undergrad, 7 years ago).
We have an enormous amount of production infrastructure built around producing transistors and not much else. I would not bet on memristors becoming competitive any time soon.
Re:The funny thing about infrastructure (Score:1)
Re: (Score:2)
Re: (Score:2)
Phrase the funding request to a gov/mil and enjoy decades of fundi
Re: (Score:2)
Re: (Score:2)
If you really want to blow your mind on something, watch this talk [youtube.com] on all possible sentient spaces, as in the set of possible intelligences/consciousnesses.
Re: (Score:2)
Throwing a shitload of computing power at a problem doesn't work unless you have far more than a vague idea of what you are trying to simulate.
If we had a magic SF computer available asking "what do you want me to do Dave?" we still have to h
No Surprises There... (Score:5, Informative)
Neurons aren't digital processors. A set of connected neurons isn't either. Neuroscience already knows that it's really difficult to learn about the structure and function of the brain from the available tools. What was more interesting was that they were able to pick up anything. They found that the chip had a master clock, for example.
There are people already challenging the use of viewing the brains as a computer (signal ins and outs) in terms of really understanding how brains organize and function. So, given all this, it's not surprising that the methods didn't fare well. The neuroscientists already knew they had a very tough task, it's those in CS and AI that are assuming that understanding the brain is the same as understanding a collection of digital circuits.
Re: (Score:2, Insightful)
Digital circuits, yes, but analog circuits exist (Score:1)
Re: (Score:2)
Re: (Score:1)
Nice strawman. Nobody calls machine learning "AI" anymore except for irresponsible marketing hucksters. The failure to manage expectations last time people got their hopes up is one of the most recited anecdotes about the history of the field.
Just like our understanding of biology has been informed by better imaging techniques, optogenetics, Microelectrode Arrays, and better imaging techniques will surely continue to inform our understanding until simulations can replicate the results.
At the rate machine le
And that makes it a strawman, how? (Score:2)
Once we eliminate all the posturing around the concept of intelligence by changing the term to machine learning, all the arguments collapse.
Re: (Score:2)
the brain (and intelligence) is just a collection of neurons, and neurons can be modeled in circuits. Guess what? It can't.
Why not? If you could prove that, or even come up with a reasonable explanation of why, that would be the most important discovery in Computer Science in the last 50 years.
Re: (Score:3)
AI means artificial intelligence, artificial is the key here. The goal of AI is not to emulate a human brain down to the cellular level.
The point of AI is to perform functions that normally require human intelligence. For example a chess AI performs a function that normally requires human intelligence, and it does it artificially, so it is an artificial intelligence. Because it only does one thing, it is called a weak AI. When an AI is able to reproduce every function of human intelligence, it is called a s
C'MON (Score:5, Funny)
It's not brain surgery.
Re: (Score:2)
With very few exceptions... (Score:2)
...a 6502 is not a brain.
The issue is, the 6502 is several orders of magnitude less complex than a brain. It could be likened to a massively parallel computer that is running thousands of programs all at once. So it is completely reasonable, on the scale of the brain, to suggest that damage to an area in a dozen people that affects their hearing to draw the conclusion that that part of the brain is responsible for hearing. Damaging a couple transistors in a 6502, a single processor, is akin to damaging a
Rocket Surgery (Score:1)
It's like asking my boss to explain the technical details of what I do. Whenever he asks me to explain the details, I know it's going to be a really short conversation/meeting. About 3 sentences in, he waves his hands in the air and says "I don't need to know the details!"
It's like saying (Score:1)
"Neuroscience Can't Explain How a Microprocessor Works" is like saying "Herpetology Can't Explain How a Bicycle Works."
Academic Clickbait (Score:1)
I think the biggest flaw in this paper is that perturbing an analog system is nothing like perturbing a digital system. To be clear, if the brain is anything comparable to a computer, it is a computer built from millions of parallel analog processors. Perturbing an analog system can be informative in ways that perturbing an digital system would not be -- analog systems can reveal half answers and shades of grey even when severely disrupted. A microprocessor will throw a fit if a single bit gets flipped un
They could have asked me (Score:2)
Such a good idea (Score:2)
Trying to run before you can walk (Score:1)
Re: (Score:3)
What you're proposing is basically a GA: Genetic algorithm.
Even when you give a system a biological analogy as its base, the results are unpredictable, un-interpretable, and don't confirm to any logical architecture.
There is a famous example of a chip designed to detect two different fixed frequencies of an input signal, and output which is active (if any). Designing the chip by hand results in a working, logical model of a certain size.
If you allow GA to run random "evolution" over the circuit contents, p
Film at eleven (Score:2)
This is why I don't get a nephrologist to fix my car.
Was this news too fresh half a year ago? (Score:2)
Fucking Microprocessors (Score:2)
And biology... (Score:2)
can't explain how a steam engine works. So?
Oh /. where art thou? (Score:2)
Oh dear, do we really that stuff here these days?
Good God Stop (Score:2)
map != territory (Score:2)
From TFA: (Score:2)
The patterns were a mishmash of unrelated structures that were as misleading as they were illuminating.
This pretty much describes the state of every branch of science after a major influx of new data. Just look at the maps of the world produced after Europe became aware of North America. Early maps sometimes show California as an island [wikimedia.org]; and it's not because the cartographer is stupid; he just put the data at his disposal together into what was at the time a plausible conjecture. And in fact the problem might not even have been that he was ignorant. He may have misinterpreted some of the (at that stage)
First Thing First... (Score:2)
They should've funded a brain-scanning gadget for Apple IIs.
https://hardware.slashdot.org/... [slashdot.org]