Complex Living Brain Simulation Replicates Sensory Rat Behaviour (cell.com)

New submitter physick writes: The Blue Brain project at EPFL, Switzerland today published the results of more than 10 years work in reconstructing a cellular model of a piece of the somatosensory cortex of a juvenile rat. The paper in Cell describes the process of painstakingly assembling tens of thousands of digital neurons, establishing the location of their synapses, and simulating the resulting neocortical microcircuit on an IBM Blue Gene supercomputer. “This is a first draft reconstruction of a piece of neocortex and it’s beautiful,” said Henry Markram, director of the Blue Brain Project at the Swiss Federal Institute of Technology in Lausanne. “It’s like a fundamental building block of the brain.”

Comments

I for one

[JustAnotherOldGuy]

I for one welcome our Simulated Rat-Brain Overlords.

(1st pest?)

Re:I for one

[Tablizer]

Congress?

Re:I for one

[JustAnotherOldGuy]

Congress? Oh no, that would assume they have brains, lol.

Honestly, I think a Beowulf cluster of synthetic rat brains would be more productive than the whack jobs we have in Congress right now.

Re:I for one

[Applehu Akbar]

" the results of more than 10 years work in reconstructing a cellular model of a piece of the somatosensory cortex of a juvenile rat."

Although we can't rush such new tech into production, we might already try having it plead cases in municipal and county courts. Then we will see if it can represent a patent troll in East Texas.

Re:I for one

[turbidostato]

You say so, but look at the first declarations from this rat brain experiment:
-Gee, Brain, what do you want to do tonight?
-The same thing we do every night, Pinky - try to take over the world!

Re:I for one

[whimdot]

I smell a rat

Re:I for one

[JustAnotherOldGuy]

I smell a rat

Yes....but is it a real rat, or a simulated one?

Isn't it a bit late?

[Giant Electronic Bra]

So I guess the implosion of the Human Brain Project has FINALLY gotten Dr Markram to publish something on brain simulation? He might have deflected a lot of criticism and saved himself a lot of grief if he'd done this 2 or 3 years ago.

Re:Isn't it a bit late?

[SuricouRaven]

They are trying to reverse engineer one the most complicated structures in the known universe, which operates on principles unlike those of any construct of human engineering. Even greatly simplified simulations of the most miniscule parts require a supercomputer to run - and that's just for rats. Do you expect progress to be rapid?

Re:Isn't it a bit late?

[jcdr]

From the page 487:
Received: December 16, 2014
Revised: May 4, 2015
Accepted: September 11, 2015
Published: October 8, 2015
Add to that the time needed to redact a 30+ pages paper in coordination with all the authors.
That said, I don't know if it's a normal timing or not in that research field.

Re:Isn't it a bit late?

[Giant Electronic Bra]

I'm just saying, its been 10 years since Markram began this line of research, and NOTHING has been published in the formal literature up to now. One might question whether or not they've actually ACHIEVED anything, and if 10 years of effort isn't enough to publish a paper if the project is worth funding at a huge level going forward (Because the HBP is essentially just funding Markram's project as he's in complete control of all its funds, or was until recently). I think a year for a publication to be reviewed and finalized is maybe longish, but maybe not. Still, the HBP was started in 2013, so it was still begun on the basis of a project that hadn't published in 8 years. THAT is very unusual indeed!

Re:Isn't it a bit late?

[jcdr]

Now that the paper is out, did you think there archived sometime or not ?

Re:Isn't it a bit late?

[Giant Electronic Bra]

I'm not sure. I have read the popular reports. They sound interesting in some ways, and I don't think the effort is worthless. I may not be qualified to say exactly whether it is the best way forward or if other initiatives, which now seem to have gotten the upper hand, are better options. I will read the paper though if I can. I have an open mind about it, and 'simulate the human brain' certainly has a certain visceral appeal.

Re:and it already proved itself

[fyngyrz]

and all trolls

Perhaps

[nospam007]

But the obvious question is:

Can it drag a slice of pizza down the stairs?

But can it do Pizza Rat?

[michael.karl.coleman]

Real science is cute, too.

WHOOSHHH!!!!

[Giant Electronic Bra]

I think you should read up on the Human Brain Project. Markram has been doing this rat brain thing for 10 years now, and he's launched a billion Euro super project based on it, yet this is the FIRST time he's ever published any results. The HBP is also crashing and burning right now, bigtime. In fact it looks like Markram is pretty much been kicked off it.

Re:WHOOSHHH!!!!

[NoImNotNineVolt]

I'm fascinated by this field of research but haven't been keeping up. News of HBP crashing and burning right now is indeed news to me. Aside from the objections from the computational neuroscience crowd (that I would summarize as "we don't fully understand the brain yet, so simulating it to gain a better understanding doesn't make any sense!") and Markram's unwillingness to entertain that claim, is there anything actually wrong with it?

Re:WHOOSHHH!!!!

[Giant Electronic Bra]

Well, the whole rest of the neuroscience community pretty much rose up in rebellion. It turns out the structure of the HBP was pretty much entirely whatever Markram wanted it to be, it was his little dictatorship and he was answerable to nobody in effect due to the way he structured things. The EC gave him virtually carte blanc, didn't properly oversee the project, etc etc etc. There's a long and complex litany of issues. Finally the outcries of the rest of the field became so deafening that they HAD to look into it. A panel was formed, etc. Markram is effectively dethrowned at this point, though I guess technically he's still the head of the project for the time being. The entire thrust of the endeavour has been renegotiated, its now designed to be a project developing technological support and software/hardware infrastructure to support neuroscience research. The 'build a human brain simulation' goal is perhaps not gone, but its not realistically expected to be accomplished or even attempted within the scope of the HBP as it exists today. Its still an open question what level of funding the whole effort will retain, it was a $1.3 billion project, but it could well be scaled back to 10 or 20% of that.

And one of the big issues is that Markram was just pushing his research agenda, yet he'd never formally published anything on the last 10 years of research with IBM! Nobody really has any idea except his pop science articles about what exactly he's actually accomplished. The whole notion of throwing EUR 1.0 billion into a research project that's never published a single result was frankly absurd.

Now, maybe Markram's goals are valid and we'd gain some great insight or at least a hugely valuable experimental tool, if we could execute his vision, but its simply not clear that it CAN be executed. This publication is the VERY FIRST one that can be evaluated on its merits to help come up with an answer to that question. The fact that Markram didn't publish SOMETHING at least 3-5 years ago is crazy. Either he's very shy of criticism or has a huge ego (and the later is thought to be the case anyway).

Its a hot mess in any case, and this publication is just a very interesting footnote to the greater story.

Re:WHOOSHHH!!!!

[NoImNotNineVolt]

So, basically, management issues, not technical ones?

Personally, I'm eager to see what happens with a brute-force approach to simulating a human brain, and I thought that was the sole goal of HBP. If it's being turned into a project developing technological support and software/hardware infrastructure to support neuroscience research, I'm definitely disappointed. While I'm sure that's a worthwhile endeavor also, I also suspect that there's lots of other people contributing towards that type of work also. I don't know of any other team seeking to simulate a whole brain, though.

It's unfortunate that one man's ego could derail an endeavor as fantastic as this. I hope the committee that Markram is replaced with doesn't abandon this opportunity for a moon shot, as I don't know if the will for creating another opportunity like this will exist any time soon. It's not every day that $1.3B of public funding gets dumped into brain simulation.

Re:WHOOSHHH!!!!

[Giant Electronic Bra]

Well, I reserve judgement on whether his ideas have merit or not. I might have some slight insight into models and certainly software and such. Still, its an interesting project in one sense, and yet in another sense I understand the objections perfectly. If you end up with a simulation that is as hard to understand as the actual brain, what have you gained? The value MUST all be in the process, not the result, so why fix now on one specific result which may not prove to be the best choice?

My guess is this project is over in its current form. Some funds will probably be withdrawn, and the scaled-back project will be structured to complement the US 'BRAIN' project, which is looking at neuroanatomical tools. The two research programs could be highly complementary, with Europe developing the software to manage and manipulate data about the brain, and the US developing the tools to acquire that data. The whole simulation project probably should be built on top of these tools anyway, and it may actually be helped more by a stepwise approach to that vs Markram's all-in strategy.

We shall see, but I suspect we'll never know exactly what Markram was going for.

Re:WHOOSHHH!!!!

[NoImNotNineVolt]

If you end up with a simulation that is as hard to understand as the actual brain, what have you gained?

Determining the presence/absence of interesting emergent properties alone would be invaluable.

The value MUST all be in the process, not the result, so why fix now on one specific result which may not prove to be the best choice?

See, this is where I disagree entirely. There are those who seek to gain some fundamental understanding of the brain, and then there are those who believe that a brute-force approach to simulating the brain might lead to the creation of virtual brains with emergent properties similar to those observed in biological brains. I fall into the latter camp. I just want to see what happens.

Granted, I can see why people might take issue with a $1.3B project that amounts to "let's just do something ridiculous and see what happens!"

Re:WHOOSHHH!!!!

[Giant Electronic Bra]

Yes, but what you fail to understand is these "emergent properties" would be what? Intelligence? Suppose you simulated a human brain and it talked to you intelligently. What have you learned about intelligence that you can't learn from talking to me? The learning process involves the ability to analyze and control your simulation, not just to make it. In essence if you can 'instrument' this brain simulation in a way that lets you 'dissect' its process and decompose it into more abstract representations then you've learned something. Thus the 'journey' to the brain simulation, not the simulation itself, is the value. Its what you learn when you tinker with some parameters and different things happen. Its when you can build models of PARTS of the brain and analyze them in isolation and with controlled inputs, etc.

Simulating an entire brain by 'brute force'? I don't see any inherent value in that at all. I think the problem is Markram seems fixated on this 'whole brain' part of the question and he's just either not expressing or genuinely not understanding the perspective I'm putting forward. The neuroscience community generally seems to believe that we have to approach this all bottom up, that we need to parameterize the processes of mental function and model them as abstract systems in order to really understand them. This seems to be very different from Markram's vision and its not clear that his approach is really valid.

Re:WHOOSHHH!!!!

[physick]

Let's examine a few points:

1) " the whole rest of the neuroscience community pretty much rose up in rebellion."

156 people signed an open letter that was started by a tiny number of neuroscientists who disagreed with the HBP premises; fair enough. Disagreeing is good. But this is hardly " the whole rest of neuroscience". There are, I believe, more than 80 universities and research institutes in the HBP, and several hundred people work on it. So, several hundred people disagreed with several hundred other people.

After the letter was published, a lot more people signed it (800 or so) but not all of them are scientists as you can see here (http://www.neurofuture.eu). If you read some of these comments they are illuminating as to motives. I quote:

"As a clinical neuroscientist I am convinced that the human brain project does not succeed. The ressources should be used for more promising research in clinical neuroscience"

2) "HBP was pretty much entirely whatever Markram wanted it to be" "And one of the big issues is that Markram was just pushing his research agenda, ."

If by "pushing his agenda" you mean writing a grant application that says " I am going to do X" and then getting the money and then doing "X", what's wrong with this? It's how science gets funded. The HBP won the contest (actually, half the context, the graphene flagship won the other half), so they get to spend the money in the way that they said they would.

3) " if we could execute his vision, but its simply not clear that it CAN be executed."

That's why it's called research. Some smart people think it can be done, other smart people don't. It takes a while to publish a paper based on 10 years work with dozens of collaborators. Now that the work is in Cell, let's see what people can make of it.

4) "he'd never formally published anything on the last 10 years of research with IBM! "
"This publication is the VERY FIRST one that can be evaluated on its merits"
" this publication is just a very interesting footnote to the greater story."

So, first you castigate Markram for not publishing anything, and then when he does, it's just a footnote.

In a project of this size, there are bound to be teething problems and probably mistakes. They can be corrected. What surprises me is the level of animosity, ad hominem arguments and plain spite against a project that won an open contest to pursue a well-publicised research goal. Why people who don't share that goal, and are not in the project, should dictate the terms of the project seems to defeat the point of giving the grant.

Full disclosure: I work on the HBP.

Re:WHOOSHHH!!!!

[Giant Electronic Bra]

I don't think your point 2 actually is fair. HBP is not 'a research project', at least that isn't how the EC CONCEIVED of it, they conceived of a large-scale multi-disciplinary project with many different research groups working on related issues. So it wasn't intended to be simply a research project that was proposed by Markram, it was intended to be an umbrella program that directed the overall high-level goals. But Markram and a very small group of his people had an outsized influence on the direction, and were able to steer much of the resources of the project to their own research groups. I'm not sure to what extent IN PRACTICE they did this, but they clearly didn't listen to a lot of other input, else why the 'rebellion'?

And that brings us to point 1. The objections were so serious, so widespread, and involved such highly successful and influential academics that they couldn't be ignored. When a panel was convened 56 findings were made, and the EC people have largely agreed with those findings. Now, I'm sure there's plenty of politics involved here, but this project had a VERY unusual governing structure, and it was clearly having some real issues. It wasn't just some griping from what I've seen. Maybe "only 156 people" objected to start with, but I see very few people out there on Markram's side, and there are a LOT of questions that have been asked and the answers aren't good.

3 doesn't make sense. If something is known to be impossible or its value can't be established then maybe it isn't a suitable target for such large-scale research spending. Shouldn't more effort go into answering the basic question of if this kind of simulation is possible and what the benefits might be? Like maybe starting with simulations of much simpler nervous systems?

and I don't understand 4 at all. His publication isn't 'a footnote', its just way late in the scheme of things, because its a horse coming well after a cart. His publication record on the project should have been out there demonstrating what was actually being accomplished, BEFORE EUR 1 billion was committed. Now, its not Markram's fault perhaps that the EC was stupid enough to launch the project, but he might have achieved a very different result than being tossed off the project board and perhaps getting the whole thing defunded if he'd published 3 or 4 years ago. SURELY there was something interesting enough in the first 6 years of his research to warrant the publication of a paper? If not then one REALLY needs to ask if this research was leading somewhere or not. I think it is at least unfortunate that publication was a low priority for Dr Markram and he probably made a bad mistake there, which was really all my original comment was saying.

So I'm full circle to that original comment, "isn't this kind of too little, too late, now that he's publishing after being politically defeated."

Re:WHOOSHHH!!!!

[physick]

Thanks for your comments. I think maybe a lot of the management problems arose because the FET flagships were a new funding mechanism, and the EC may not have had clear ideas how they should be managed. Management problems have certainly loomed large in the HBP, but I think they have still been blown out of proportion. I wonder if the FET mechanism is only a beta version, and next time the EC will devise the release version. After all, the EC has many goals that are not only scientific, and 0.5 billion is not in european terms a huge amount (think Greece bailout funds). From the EC point of view, if side products of the HBP research led to a 1% increase in european employment and proved that brain simulations were not possible, the EC would be delighted.

"The objections were so serious, so widespread, and involved such highly successful and influential academics"

It is not clear to me that this is true. Some highly influential people have blown up a story that has flown in the press, but that does not mean the objections are serious. The press, including the scientific press, love a good battle, and being negative about something that costs a lot of money (waste!) is more saleable than describing the truth about a complex story. And even if some influential academics don't like the project, that is not evidence that they are right and the project is wrong. The HBP also has a lot of influential scientists on board, whose arguments do not get so much coverage.

  " If something is known to be impossible or its value can't be established then maybe it isn't a suitable target for such large-scale research ". Clearly, if it is "known" to be impossible, there is no point doing it. But in the HBP case, a certain subset of neuroscientists think a cellular level simulation of the mammalian brain is either impossible or not worth doing. This is not the same as it being impossible, Maybe the HBP would demonstrate its possibility or impossibility. Its a judgement call whether that knowledge is worth 0.5 Billion euros ( not 1 billion as all the press report, as graphene gets the other half, and the partners in the project have to stump up the matching funds).

By comparison, when the Apollo missions started, rockets failed and people died, and it was not known that it was possible to survive a journey to the moon and back. But people building trains did not get to interfere in the development of rockets.

Finally, about publications: there are more papers that have been published now than just the Cell one. Maybe, earlier publication would have been better, but I doubt that it would have changed the minds of those scientists who object to the principle of a brain simulation. And this is, for me, the big point. Some arguments against the HBP are of the form "its badly managed, too concentrated in a few hands, not obviously worth the money" - these are reasonable arguments, and the mediation committee have addressed them and the HBP is taking them on board. But the other arguments, especially those expressed in the open letter, and Nature commentaries, are just invective and not subject to rational argument. Go and read some of the comments on the letter. These comment authors are not interested in the finer details of the argument, they just hate the idea of a brain simulation, or they hate Markram, or they hate the money going to someone else. And this, for me, is the depressing aspect for a community that is supposed to be made up of rational scientists.

Re:WHOOSHHH!!!!

[Giant Electronic Bra]

Thanks for your comments. I think maybe a lot of the management problems arose because the FET flagships were a new funding mechanism, and the EC may not have had clear ideas how they should be managed. Management problems have certainly loomed large in the HBP, but I think they have still been blown out of proportion. I wonder if the FET mechanism is only a beta version, and next time the EC will devise the release version. After all, the EC has many goals that are not only scientific, and 0.5 billion is not in european terms a huge amount (think Greece bailout funds). From the EC point of view, if side products of the HBP research led to a 1% increase in european employment and proved that brain simulations were not possible, the EC would be delighted.

Yeah, its nice to have actually intelligent conversations ;) Thx. I think you may be right, and it is certainly true that as a govt institution EUR 1 billion must not seem like a very big stake. The other flagship project it seems has been quite successful so far though, so its not clear that the issue is entirely the overall mechanism, it seems more likely that project's individual characteristics are a really important aspect of the equation. Certainly I think people will agree that in the future the lessons learned from these projects should be used to improve the overall process. I would assume that the public should expect this, as it only meets something like CMM level 2 grade management, and we should insist on at least that level of competence from large governmental institutions (I'd argue that we should demand CMM level 5 performance from ALL public institutions and any failure to reach it should be rectified actively).

"The objections were so serious, so widespread, and involved such highly successful and influential academics"

It is not clear to me that this is true. Some highly influential people have blown up a story that has flown in the press, but that does not mean the objections are serious. The press, including the scientific press, love a good battle, and being negative about something that costs a lot of money (waste!) is more saleable than describing the truth about a complex story. And even if some influential academics don't like the project, that is not evidence that they are right and the project is wrong. The HBP also has a lot of influential scientists on board, whose arguments do not get so much coverage.

Sure, and not being a neuroscientist of any sort I can only make a relatively arms-length and perhaps superficial analysis of the whole crisis. Still, the EC seems to have drawn largely the same conclusions as these objectors and they've restructured the project (or are in the process anyway), so its likely there was at least SOME substance to it. I think my conclusion is that AT LEAST Dr Markram failed to manage the project in a way that allayed these fears and objections. I suspect that earlier publication of substantial results from his own project would have helped, but I could be wrong, maybe there was no avoiding this crisis, it was just purely a product of politics and ego.

" If something is known to be impossible or its value can't be established then maybe it isn't a suitable target for such large-scale research ". Clearly, if it is "known" to be impossible, there is no point doing it. But in the HBP case, a certain subset of neuroscientists think a cellular level simulation of the mammalian brain is either impossible or not worth doing. This is not the same as it being impossible, Maybe the HBP would demonstrate its possibility or impossibility. Its a judgement call whether that knowledge is worth 0.5 Billion euros ( not 1 billion as all the press report, as graphene gets the other half, and the partners in the project have to stump up the matching funds).

OK, I didn't delve that deeply into the details of the funding. Its a pretty chunk of change from the science perspective in an

Re:WHOOSHHH!!!!

[physick]

"AT LEAST Dr Markram failed to manage the project in a way that allayed these fears and objections. I suspect that earlier publication of substantial results from his own project would have helped, but I could be wrong, maybe there was no avoiding this crisis, it was just purely a product of politics and ego."

I agree, but there were external pressures that meant that it was not easy to direct the project in a way that would prevent what happened. Some decisions were forced. And now that it has happened, people's views have got fixed and resetting the project's position in the minds of the neuroscience community is probably impossible. It has to evolve into a different kind of project. But at least now there are some results out there that people can test. There will be more software tools released over the next few years which will allow people to do their own computer "experiments". If they are useful to the wider community then maybe this will all die down and interesting work be done in the HBP. Here's hoping...

Re:WHOOSHHH!!!!

[NoImNotNineVolt]

Yes, but what you fail to understand is these "emergent properties" would be what? Intelligence? Suppose you simulated a human brain and it talked to you intelligently. What have you learned about intelligence that you can't learn from talking to me?

Nothing.

The learning process involves the ability to analyze and control your simulation, not just to make it.

This statement is misleading on account of its overly narrow scope. From the abacus all the way to the integrated circuit, we've been simulating (at a very low level of fidelity) some subset of the human brain for ages. The goal has never been to learn how the human brain does math, for example, but instead to augment natural human mental abilities. Gaining an understanding of the human brain is not the only valid goal to chase after, as it is evident that the pursuit of better computational technology has, in practice, proved to be very beneficial to the human race.

Simulating an entire brain by 'brute force'? I don't see any inherent value in that at all.

So you see no value in potentially creating virtual rational agents? Ones that could enable universal automata, androids, etc.? No value at all?

I think the problem is Markram seems fixated on this 'whole brain' part of the question and he's just either not expressing or genuinely not understanding the perspective I'm putting forward.

I think the fact that Markram is fixated on this 'whole brain' part of the question isn't a problem. It's evident that his goals are orthogonal to yours, but it would be no less wrong to say that your focus on a 'partial brain' is some sort of problem. Different goals, different approaches. I don't see either goal as inherently more objectively valid.

The neuroscience community generally seems to believe that we have to approach this all bottom up, that we need to parameterize the processes of mental function and model them as abstract systems in order to really understand them. This seems to be very different from Markram's vision and its not clear that his approach is really valid.

The neuroscience community doesn't have one single unifying goal, and the fact that there are multiple competing visions underscores that fact. It's entirely likely that any given vision will not fulfill everyone's disparate goals, but no reasonable person would expect that. Do you expect a bottom-up approach is likely to enable the creation of virtual humans any time soon? If not, then why isn't your approach also "invalid"? After all, it's about as likely to bring us functioning virtual brains as Markram's approach is to bring us any meaningful understanding of how the brain works.

Re:WHOOSHHH!!!!

[Giant Electronic Bra]

To be clear, what I think is that you have to develop conceptual models, not just simulations. If all you did was literally make a neuron simulation and wire a huge number of them together you would learn what? Nothing much really, because you'd then have the same questions about how that simulation works that you have about how a real brain works. What we need are the equivalent of Kirchhoff's Laws and Ohm's Law, etc for neurology. That is a set of principles from which you can engineer functionality that meets requirements. When you have something like that, then you have understanding. (not to imply that these principles would be of the same character as the simple laws of basic electrical circuits, or so deterministic, etc, just that generalized rules of this sort ARE what constitute understanding, and it can only be measured in terms of mastery of the subject, the ability to employ those generalizations to accomplish novel things as needed).

The point is that Markram's work doesn't seem to put any focus on that. Actually we don't KNOW how much it bears on that because, until now, he hadn't published anything!

My personal opinion is that an overall framework goal of reproducing the behavior of very large sets of neurons (IE on the scale of mammalian brains) might not be a bad direction to take, but we're probably not going to succeed without developing these generalizations and using them to construct less detailed models which still perform in 'brain-like' ways that we can use as I've stipulated above. And it may well be that the low-level simulation approach simply isn't the most efficient way to get there. Maybe there are higher level neurological functional units that we can abstract, but we don't have a good understanding of what those might be yet. Maybe small-scale neuron-level simulation is the way to gain THAT understanding too, but I'm certainly not the person to even have an opinion on that.

So, my guess would be that work like Markram's is almost surely going to be valuable, at least up to a point and in service to the right goals.

Re:WHOOSHHH!!!!

[NoImNotNineVolt]

To be clear, what I think is that you have to develop conceptual models, not just simulations. If all you did was literally make a neuron simulation and wire a huge number of them together you would learn what? Nothing much really, because you'd then have the same questions about how that simulation works that you have about how a real brain works. What we need are the equivalent of Kirchhoff's Laws and Ohm's Law, etc for neurology.

If your goal is to gain a fundamental understanding of how the brain works, I agree. If your goals are to simply duplicate the functional properties of a human brain, I disagree.

Much like we didn't need to gain a fundamental understanding of materials science or chemistry (or even a table of elements) to develop bronzeworking, ironworking, even steelworking (which are practically valuable in their own right, even without any deeper understanding of the how or why) it's entirely plausible that we don't necessarily need to gain any fundamental understanding of the human brain in order to extract practical value from constructing simulated brains. In the aforementioned example of metallurgy, the cart did precede the horse, in that we were making valuable things long before we ever understood how or why these valuable things worked, and I don't think any reasonable people are saying that we should've abstained from developing these technologies until we first gained a fundamental understanding of the nature of matter.

My personal opinion is that an overall framework goal of reproducing the behavior of very large sets of neurons (IE on the scale of mammalian brains) might not be a bad direction to take, but we're probably not going to succeed without developing these generalizations and using them to construct less detailed models which still perform in 'brain-like' ways that we can use as I've stipulated above. And it may well be that the low-level simulation approach simply isn't the most efficient way to get there. Maybe there are higher level neurological functional units that we can abstract, but we don't have a good understanding of what those might be yet. Maybe small-scale neuron-level simulation is the way to gain THAT understanding too, but I'm certainly not the person to even have an opinion on that.

These are all reasonable, valid points. Going back to my previous analogy, it also would also have seemed reasonable for people to wait until our understanding of metals was more fully developed before trying arbitrary combinations of materials just to see what the result would be. Think about how wasteful that must have been, about the level of resources squandered on these early metallurgical forays that led nowhere. I imagine the alchemists of that era were incensed about this waste, thinking that spending resources on gaining a fundamental understanding of matter would be much more wise than mindlessly trying things out just to see what the end products would be. After all, even if the end products turned out to be valuable, we still wouldn't have really learned anything about how they work or why they work they way they do! However, it was many centuries or millennia before a thorough understanding of chemistry was developed, and indeed, much of the research that provided this understanding was itself dependent on the existence of metal tools. Without our unscientifically-developed metallurgical technology, it's not clear that we'd have developed this level of understanding in the same time frame, or even by today.

In much the same way, I propose that there is potential value in Markram's 'hail mary' approach, not because it might help us gain understanding into how the brain works (in the same way that developing steel didn't directly help us understand the myriad relationships between iron and carbon), but merely because it might yield a new technology which could help us in novel and unforeseen ways. Perhaps the creation of virtual human brains (again, without unders

Re:WHOOSHHH!!!!

[Giant Electronic Bra]

I would argue that it took 1000's of years, in fact if you go back to the invention of fire it took almost 4 hundred thousand years to go from there to bronzework by simple process of blind experimentation, and almost 10 thousand more years to go from there to iron, and another 2 thousand to get to a sophisticated steelmaking process. That's a very tough row to hoe, very slow progress. We would be in the same position if we're just fiddling with some brain model. Brains are also a LOT more complex than steel, so I'm not even sure how much progress we would make.

I'm not sanguine about the value of the results. Brains are horribly complex things, and without understanding how they relate to the minds they embody we have no way of 'creating a research team' of such artificial minds. I'd say we'd have a HUGE challenge just creating a simulation so perfect it worked at all. Any small problem at the brain level could create a catastrophically disordered mind, but you'd have no way of knowing how to fix it without any conceptual framework. Beyond that, most of the most high-value potential uses of AI are likely to be highly specialized niche applications which won't benefit that much from emulations of humans (or rats, etc for that matter). Emulations of some of the low-level neural circuitry I can definitely see as having value, that's already happening, but it seems to be a bit differently focused effort than Markram's.

Given that Markram was set on controlling a HUGE part of the research in this area, it seems like maybe it wasn't prudent to go ahead with it that way before at least getting some publications out on the existing research program. That was mostly my point.

I mean, I don't doubt that nobody can really predict the exact value of future research, or which is the fastest route to interesting technology. Markram's approach could hit a gold mine, or it could be a wasteland. Its primarily a question of how much effort to put into each direction.

Re:WHOOSHHH!!!!

[NoImNotNineVolt]

That's a very tough row to hoe, very slow progress.

And yet, with the hindsight that we have today, can you say that we likely would've more quickly progressed to the point of steelmaking by, instead of blindly experimenting, waiting until sufficient understanding of chemistry was achieved? I don't. I think that the absence of metal tools would've delayed scientific progress, pushing back the date at which we really understood how steel works (which itself was preceded by the first development of steel either way). It's unlikely that an absence of metal tools would've actually increased the rate of scientific discovery, in my opinion.

We would be in the same position if we're just fiddling with some brain model. Brains are also a LOT more complex than steel, so I'm not even sure how much progress we would make.

We're also a lot smarter than people throwing arbitrary rocks into campfires. It's a challenging comparison to make accurately.

I'm not sanguine about the value of the results. Brains are horribly complex things, and without understanding how they relate to the minds they embody we have no way of 'creating a research team' of such artificial minds.

The context here was your previous hypothetical situation: "Yes, but what you fail to understand is these "emergent properties" would be what? Intelligence? Suppose you simulated a human brain and it talked to you intelligently." -- presumably, if a simulated human brain talked to me intelligently, it would be capable of talking to other simulated human brains intelligently. Admittedly, we're in some rather out-there territory at this point, with hypotheticals predicated upon hypotheticals. I'm not suggesting that this is the necessary outcome here, merely a possible one. A possible one that would unquestionably justify this sort of endeavor if realized.

I'd say we'd have a HUGE challenge just creating a simulation so perfect it worked at all.

Agreed wholeheartedly. But this goal can be reached iteratively, and we've already been making tremendous progress at smaller scales. From early, limited simulations of individual neurons which matured to be 'perfect' in many ways, to current larger-scale simulations of neural columns whose behavior also correspond rather well to that of biological ones, we're making progress. Markram's idea may (or may not) be being pursued too soon, in that we may need to work on perfecting simulation of "larger than neural column, smaller than whole human brain" neural structures, but that's not clear to me.

Any small problem at the brain level could create a catastrophically disordered mind, but you'd have no way of knowing how to fix it without any conceptual framework.

A valid point. Much like early steelworkers' ignorance of the actual chemical processes involved in their work prevented them from having any way of knowing how to fix a broken process, this would similarly be a hindrance to researchers in this field. If it works, it works, and if it doesn't, you have no informed way of doing anything about it. Quite a departure from science, I admit.

Beyond that, most of the most high-value potential uses of AI are likely to be highly specialized niche applications which won't benefit that much from emulations of humans (or rats, etc for that matter).

Ah, but that's AI, and AI is a very broad field rife with niche applications :P I'd argue (as someone who develops machine learning systems for a living) that a simulated brain would be categorically different than what AI researchers are interested in. It doesn't lead to a model of cognition, it doesn't lead to an understanding of intelligence, it doesn't really advance scientific knowledge at all. It's more of an engineering project than a science project.

Emulations of some of the low-level neural circuitry I can

Re:WHOOSHHH!!!!

[Giant Electronic Bra]

I think you are presenting a false analogy at multiple levels here. We might say today that maybe the world progressed faster to steelmaking one way vs another, but AT BEST that's 20/20 hindsight, you can't actually prove it, the analogy shows no particular connection with AI, etc. There was no 'plan' to 'get steelmaking' this is the real world not Civ IV tech trees, and people 1000's of years ago didn't even have a concept of progress as a general thing. Nor is anyone proposing that we halt everything, Markram's project is only one possible way of advance, not the only way, so you've got an excluded middle AND a red herring in there as well! But really, its mostly just a limited analogy. I don't think we can know how good or bad it is right now.

Are we really that much cleverer than cavemen throwing rocks in fires? Maybe, but not cleverer than bronze age metalurgists. In terms of what we know about how brains function who's to say we're much ahead of them at all?

Overall I think we're not in huge disagreement. I think brain simulation WILL continue. I think its likely to be a bit less grandly ambitious than Markram's for a while. I think I'd simulate a C. Elegans CNS first, then maybe a Planaria worm, and then perhaps some higher invertebrates, simple chordates, etc. Its going to be a good solid 30 years before the raw compute power exists to simulate a human brain anyway, maybe longer, so there's not that much of a rush if you ask me.

Re:WHOOSHHH!!!!

[NoImNotNineVolt]

I grant that the analogy is far from perfect and depends on many probably-flawed assumptions. While I wasn't suggesting that the only alternative was to halt everything, I think about this subject in the context of "which approach is most likely to yield a virtual human brain soonest". Looking at [what I call] the brute-force approach of understanding how neurons work, understanding how neurons are interconnected, and simulating them, it seems like this ought to be possible in a few decades, maybe even in a century, with only incremental progress and no expectation of revolutionary discovery. Looking at the "let's understand what 'intelligence' really is before we go trying to simulate it" approach, how do you even form an estimate? Inherently, this approach does rely on some revolutionary discovery being made. Perhaps it's just me being a pessimist, but I don't believe we'll ever have an answer for that (forever is a long time, but I say "we" meaning humans-as-they-exist-today), let alone within the next century. While I don't dismiss the value of seeking to gain a fundamental understanding of intelligence, I don't think the goal itself is attainable -- how can we hope to understand something that is many orders of magnitude more complex than anything we've managed to understand before?

But C. Elegans? with only 300 neurons, 7000 synapses? That's old news (though still not quite "complete" in some sense). Though we don't yet have a complete connectome for D melanogaster (it's in the works [sorry for the shitty citation]) nor a complete model for its neurons, but simulation work on its 100000 node CNS is underway regardless. Obviously we won't see virtual flies until this connectome is fully diagrammed and more experimental data about the neurons is available and computers get a bit faster, so indeed it is true that it is too soon to expect these projects to be fully completed. Probably much too soon. But that doesn't mean it's too soon to start, and $0.5B is peanuts when you consider how much a truly successful simulation project is likely to cost.

It's not just the compute power that's a serious limiting factor, but also the availability of imaging technology that would enable us to develop a complete connectome for the brains we seek to simulate. The most immediate hurdle for projects like these isn't our lack of fundamental understanding (although to say that such understanding "would help" would be a huge understatement), it's the combination of insufficient computing resources (though if past trends continue, this won't be an issue for long) combined with insufficient knowledge of the brain's structure (which can be remedied by continuing to advance the state of medical imaging technology, and dfMRI seems very promising recently). Getting more experimental data to develop accurate neuron models for various animals is simply labor-intensive and not really waiting on anything except more funding. However, these are well-defined engineering or funding problems, and steady incremental progress is being made on all of these fronts. I expected this infusion of $0.5B to simply help that along.

We have the technology to invasively/destructively map brains and develop accurate neuronal models today. We have the technology to simulate interconnected neurons today. Progress will continue, but only at the rate at which we fund improvements in these underlying technologies as well as projects that seek to pull together these technologies to shoot for ambitious brain simulations. While I don't doubt that there are other worthy causes to spend this money on (even within neuroscience), I think it's still unfortunate if this particular approach doesn't get the green light because of the actions of one seemingly power-hungry individual.

Re:WHOOSHHH!!!!

[Giant Electronic Bra]

Well, we have the ability to map a very small number of neurons. Actually do we even have that? We still don't EXACTLY understand how the potentiation of each synaptic junction works. There are something like 700 TRILLION of them in a human brain too, so we're not even remotely close, even assuming we can determine the weight for each one individually, to scanning even a tiny fraction of an actual brain. We couldn't even do that fruit fly, so even assuming we know where the synapses are that's like having just the hardware but not the software. And if you don't know how the things work at some higher level then you're going to actually have to measure the weight of each of those 700 trillion junctions to create a human brain sim. I'm guessing, but I think we're a LOT more than 30 years from being able to do that. Just think of it in terms of being adatabase problem. 700 trillion records? Yeah, good luck! In every respect our capabilities obviously fall far short, and even things like mice, heck even fish, are WAY beyond our reach there. So I think there's a huge amount of basic work that comes before even worrying about actual brain sim itself. At least that's how I see it, and that's STILL assuming that brute sim is a better investment than figuring out the principles and building more abstract models. By the time you get your brute force sim to work I may well be able to implement a more abstract one with 1 millionth of the effort.

But of course we don't know. I agree with you on this, Markram's way will cost a boatload more than EUR 500 million. I'm not so sure its WORTH that boatload, and I would want to know before going down that road at all, at least very far. With simply CNS sims there's almost surely things to be learned, but it is much less clear there's a real value proposition with higher organisms.

Re:and it already proved itself

[fyngyrz]

Oh, the troll, of course. Any other questions?

Singularity after all?

[blue9steel]

NUMBER OF NEURONS
Rat Brain 200,000,000
Human Brain 86,000,000,000

If a Moore's law type progression occurs in this field we'll get human brain simulation within a decade. If so, watch out because everything will change. Of course the researcher is only simulating a portion of the rat brain and technology rarely moves in straight lines, so let's be reasonable and call it twenty years.

Can we buy a hyphen or two?

[gstoddart]

Complex Living Brain Simulation

So much disambiguation needed.

Complex-Living Brain Simulation ... the brain simulation lives in a complex.

Complex Living-Brain Simulation ... it's complex and simulates a "living brain".

Complex-Living-Brain Simulation ... a simulation of brains which live in complexes.

I'd like to say I expect better from the Guardian. I'd like to, but I can't.

Soma?

[MenThal]

Great, now we have sentient rats on the ARK. What devastating plagues will they bring along this time?!?