Hacker News Reader

Comment by Leon | Hacker News Reader

His sticking point is that the brain or mind could be a physical system that is not computable or equivalent to a Turing machine. If we take the assumption that the mind is a physical system, then it would still only be a matter of time before we found the reason for the difference in computational ability (maybe it would be like Penrose's quantum arguments for the mind), but those systems would still be within reach of implementation within a computer system. The brain is still a combination of proteins, amino acids, and a few other chemicals. We would just have to move to using chemical computers (which are already being researched).

Even following his arguments, it would only be a matter of time before sentient machines are created, equivalent to our own minds.

His arguments against AI are mostly 'just because' and has little proof.

View on HN

No, his arguments are not 'just because', this is a pretty smart guy and from what I get out of it his reasoning goes like this:

If the mind is a physical system then we can attempt to recreate it by computer but this is no guarantee that such an attempt will be successful.

For an example of such a situation, imagine that it would take all the resources of a planet in order to create a single AI then it might be theoretically possible but in practice it would fail anyway.

And that is assuming that it is possible, for which there is no proof until it is done. If the mind turns out to depend on mechanisms that we can not model accurately enough to get the desired effect then the whole deal is off.

There can be no 'proof' that AI is impossible, just as much as there can be no 'proof' that green elephants do not exist.

The only proof that would be acceptable is for someone to produce an AI.

Nobody needs to present proof that it is not possible.

Well, that's not entirely true. What is true is that if the mind is more powerful that a Universal Turing Machine, then we have no idea how to approach building one. However, we know that the resources required to build one are small -- humans do it all the time in a relatively cheap process called "sex." That is, the lower bound for creating AI is very small. Of course, the danger is that our brains have certain limitations for a reason -- these could be the property of this super-Universal Turing Machine. Therefore, a true AI may look nothing like our modern computer. The GP is correct in that we should at least be able to build a simulacrum of our own mind -- even if that is simply an artificially constructed duplicate using biology, genetics, and biochemistry. I would say that this may be fairly far away, because we have yet to truly understand many of the processes and structures in the human brain. This is why I believe AI is much farther off than most people think: to understand AI we must first understand ourselves, which is much more complex than at first take.

Now, this entire hypothesis rests on the idea that our brain is super-Universal. That is, it cannot be phrased as a computable function (by the Church-Turing thesis). This is a fairly unpopular view in the AI community, but valid until we prove otherwise. Up till now there has been no evidence that such a Turing machine exists in theory, much less in practice, which is a minor blow. In addition, many of our psychological structures and chemical activities seem to suggest a Universal Turing machine.

Edit: I just realized that I implied that a super-Universal Turing machine is a Turing machine. This is obviously untrue, by definition. Consider my terminology to be a replacement for whatever name we would give to this noncomputable function solver (since we have no category for that at the moment).

Edit2: Not that we don't have a category for noncomputable functions, only that we don't know which category they would fall into. I guess the catch-all "hypercomputable" is suitable. Thanks, naveesunder

>>This is why I believe AI is much farther off than most people think: to understand AI we must first understand ourselves, which is much more complex than at first take.

I saw an interview with the director of the Blue Brain project. He projected 10 years, before they can emulate a brain...

http://nextbigfuture.com/2009/08/henry-markam-project-direct...

"Computational Neuroscience over the past 50 years is a theory driven science, while Blue Brain is biology driven. When we reach the human brain, human-like perceptions and motor actions should emerge automatically. The model brains will be able to learn to do what we human’s can learn to do, perform complex decision-making, manifest emotions, intelligence and personalities. We see these all as straightforward emergent properties of the brain. Self-awareness and consciousness may also emerge if this phenomenon depends on neuronal, synaptic and/or molecular interactions. Anything that depends on the physical elements that can be measured in the brain should emerge if we are successful in building it accurately enough. If we inject some theories of brain while we build it and ignore the biology, then we are back to square one with computational neuroscience and we will almost certainly fail."

As I said before, he's on the hook for having said that.

I think that was a very foolish statement to make, now the clock is ticking.

But I'll be very happy when on August 24th 2019 mr. Markham will reveal the artificial brain. For now all he's got is a great way to get a lot of funding. More power to him.

>>For now all he's got is a great way to get a lot of funding.

You really need to motivate that... The project has afaik finished the first phase with good results. (See the interview I referenced.)

Also, you don't have to wait until 2019 to see if you are right and the researchers are dishonest and/or idiots. The Blue Brain are going to work on emulating animals with much smaller brains at first.

Simulating, not emulating.

Yes, the first phase has been successful, maybe even later phases will be successful.

But it all depends on how you define success.

Let's pretend we're living on an island and we are looking up in to the sky and we see aircraft flying over. Instead of going the 'cargo cult' route to building an aircraft we are a bit smarter and we realize it must be their forward speed that somehow holds them up.

So, we build a windtunnel, aim it into the prevailing winds and we make little wooden planes to test our theories of aircraft.

Sometimes the winds are strong enough to give us a meaningful result.

The shapes will check out, all kinds of stuff will be learned that is useful for making airplanes.

And that's where we get stuck, in spite of all our fantastic work in early 'phases' we do not have a functioning aircraft. For that we need other technology that we do not have and that we will not be able to simply acquire by looking at an airplane or even taking apart a crashed one. You need access to a functioning airplane for that, and preferably someone that can explain what technologies went in to building it and how important the various tolerances are. There is more to making an airplane than just the outside view of it. But there are many ways of making things fly, such as a balloon, a rocket or even a glider. Any one of those might be valid for our purpose of making an aircraft and by studying physics (the brain analogy breaks down here) you would learn what really goes in to making an aircraft.

For me there is only one phase here that you could claim to be successful or not with respect to the project as a whole, which is to have the thing become sentient, learning a spoken language or so and then to start asking critical questions.

All the other 'phases' are mechanistic and so will all be achieved for sure.

The last one is the only one that matters.

I'm sure we will learn a ton about how 'brains work' at a mechanical/chemical/electrochemical level. But the input data is the wiring diagram + the chemical composition of some of the ingredients.

If it would be that easy then it should be possible to restore dead people as well, but it seems that something crucial is broken when a brain is without oxygen for even a very short period. What that is is a big mystery. Religious people call it 'soul', technical people call it 'state'.

If you compare the brain with an Eniac style machine where the program is stored in the connections between the neurons then you have to also take into account that the brain is rewiring itself on a continuous basis.

But it might be more than that. What if the state is more like 'RAM', and it needs the software running on it now to keep it going. Once you unplug it the whole thing goes dead, and even when you plug it in again it stays dead. Just like a computer that only has a front panel and no bootrom. The bootstrapping is what it is all about.

A simulated brain might be fully capable of sentience but remain comatose. We'd have no way of knowing, since we do not sufficiently understand the state of comatose people as it is today. We end up feeding them intravenously for years sometimes hoping that they one day will 'wake up'.

To me this whole simulation business is just one small step on the way to 'building a brain', it may help you to get some of the details right but to boldly claim that in 10 years you will have a functioning human brain is not clever.

It makes the same mistake as the AI people in the past have made, to assume that because the first steps are easy all the steps are going to be equally easy.

As far as I am concerned this is open ended research, it may take decades, possibly centuries to achieve the last stage IF it can be done.

To put your career on the line with a hard claim like that either means that you have a crystal ball or that you are overly sure about yourself.

Scientists announce findings of fact, they do not predict the future in cases as complicated as this one.

Anyway, I'm getting seriously out of my depth here, I hope they will succeed with as many 'phases' as possible and I'm happy they have a charismatic project leader that has been able to get the funding required to do this research but I really would have been much happier if he had not named a hard date. That was an unfortunate decision.

I don't think they are dishonest or idiots at all, merely underestimating the scope of the problem.

>>For me there is only one phase here that you could claim to be successful or not with respect to the project as a whole, which is to have the thing become sentient, learning a spoken language or so and then to start asking critical questions.

Uhm, you argue here that it would be without value if you don't solve the hardest problem of all in ten years.

Have the Blue Brain project claimed anywhere that they aim for full human behaviour in 10 years?!

It would be wonderful if they can simulate a rodent's brain with nerve inputs and get it learn to run in a maze and do similar things.

>>If it would be that easy then it should be possible to restore dead people as well, but it seems that something crucial is broken when a brain is without oxygen for even a very short period. What that is is a big mystery. Religious people call it 'soul', technical people call it 'state'.

I thought the technical term was "biological damage"?

You really seems to argue that anyone with some biological brain damage from oxygen deprivation loses their "soul"?!

Either you're drunk, trolling me or you don't know anything at all about biology and damage to the brain and loss of partial function and therapy.

>>What if the state is more like 'RAM', and it needs the software running on it now to keep it going.

I think that argument was killed by electric shock therapy.

>>I don't think they are dishonest or idiots at all, merely underestimating the scope of the problem.

Uhm, (i) WHAT have the Blue Brain project really claimed? (ii) I'm not an expert (I'll read up on brain science in a few years, when it has stabilized... the field seems to move faster than light, right now) and the little I know is mostly on cellular level, but you really don't know anything at all about biology?

> Have the Blue Brain project claimed anywhere that they aim for full human behaviour in 10 years?!

They predicted that an artificial human brain can be built within 10 years.

So, yes.

> It would be wonderful if they can simulate a rodent's brain with nerve inputs and get it learn to run in a maze and do similar things.

Agreed, that would be amazing.

But I can already do that with relatively simple electronics, that does not even require software. What would be even more amazing if this simulated creature showed initiative of its own.

> I thought the technical term was "biological damage"?

Not neccesarily. A computer that is running a stored program in ram is indistinguishable on an atomic level from one that has just been rebooted but is hanging at the boot prompt waiting for a key press because there is no network to boot from.

The one is a useful tool, the other one a door stop.

Damage did not enter in to it, the computer is fine, the bootstrap sequence is what is holding it up.

> You really seems to argue that anyone with some biological brain damage from oxygen deprivation loses their "soul"?!

You are trying to make me sound ridiculous because I tried to draw an analogy between how religious people (and I'm not religious) see the situation and how an electrician or a scientist would see it. State is information, the information is what can be lost, even though you still have all the physical components.

> Either you're drunk,

thank you for that.

> trolling me

Apparently not, I don't have time to waste in spite of spending a considerable amount of time on answering you.

I may be mistaken, and I'm certainly open to learning.

> or you don't know anything at all about biology

I've worked my way through a university grade course in Genetics, but do not have any formal education in the field, in fact I have very little formal education at all. That hasn't stopped me from learning though.

> and damage to the brain and loss of partial function and therapy.

There seem to be a multitude of failure modes, not all of them are well understood.

>>What if the state is more like 'RAM', and it needs the software running on it now to keep it going.

> I think that argument was killed by electric shock therapy.

Could you explain that ?

>> I don't think they are dishonest or idiots at all, merely underestimating the scope of the problem.

> Uhm, (i) WHAT have the Blue Brain project really claimed?

To be able to simulate a full brain within 10 years.

> (ii) I'm not an expert

Neither am I, but I do know that if you make a bold claim like that you have a problem if you fail to deliver. That would be a pity because I think that if there ever is going to be an answer to these questions that answer is going to come from a project like this. By overselling it they are damaging their long term prospects.

The same happened to 'regular AI'.

> (I'll read up on brain science in a few years, when it has stabilized...

I read as much as I can, several hours a day on lots of different subjects, including this one, I highly doubt that they have the trajectory planned out to the point where can state that they will have a functioning artificial brain in 10 years.

> the field seems to move faster than light, right now) and the little I know is mostly on cellular level, but you really don't know anything at all about biology?

That is your assumption to make, I don't think you are right.

But I will not resort to calling you either a drunk or trolling me as you just did, I appreciate the time you took to write your answer.

Not everybody that you find yourself disagreeing with is drunk or a troll.

I wrote in GP: "Have the Blue Brain project claimed anywhere that they aim for full human behaviour in 10 years?!"

Answer: >>They predicted that an artificial human brain can be built within 10 years. So, yes.

Note the difference between building a simulation of a brain and getting the full human behavior -- which is a totally different can of worms; just consider the lacking critical sensory inputs while "growing up"...

Please check the term on wikipedia, or something: Cerebral hypoxia

Damage from lack of oxygen to the brain is well documented and gives different results depending on in which part the damage is. (This is well known because of blood vessels problems can cut off specific parts and then different times to get back blood flow. Neuronal damage and its reasons has been researched, too.)

That is a basic part of common knowledge, which you contradicted when you wrote:

I am sorry I took that as trolling.

(The point about ECT was that it "resets" electrical state. I don't really know, but suspect the same could be said for many grand mal attacks. As far as I know, there is very little support for your thoughts on a need for consistent state, since the chemical properties and other hardware changes can rebuild the state.)

Essentially the Blue Brain project is a study in emergent behaviour, simulate something and its behaviour should emerge. That is exactly what the project leader claims in the article linked, I don't think even they expect it to be human behaviour because that is limited to beings that are humans.

The problem with the approach is that a simulation, by definition is not the real thing and that it will always end up being an approximation.

Our most powerful computers could not simulate all the interactions between atoms in a 1x1x1 cm volume of solid matter, let alone all the matter in a chemical soup as complicated as the brain, and expect it to behave like one.

So, this is also an approximation, only a very detailed one in spite of the researchers claims.

They are still simulating neurons, one brain weighs about 1.5 Kg, it contains roughly on the order of 100 billion neurons, but it contains much more than just the neurons, and other cells in the brain could easily be crucial for its functioning.

For example there are the 'glial' cells, which at least double the number of cells in any simulation that wants to take on the whole brain, the glial cells are part and parcel of brains all brains that we've looked at have them, they may be only there in a supporting role or they may end up being crucial.

Recent studies suggest that these cells are not as simple and passive as we used to think and that they may have a more active role.

That doesn't say it can't be done, it simply says that to stick a deadline on it is unwise. You should not make bold claims without the evidence to back it up.

Essentially they are saying they are only 5 orders of magnitude away from simulating a brain in real time:

The currently achieved milestone stands at 100 million neurons with 100 billion synapses 100 times slower than real-time, and that is assuming that the simulation model is indeed accurate enough to eventually lead to emergent behaviour, which to date has not happened. (And that would be the biggest news since the moon landings in terms of human achievement).

Especially the growing up (which already starts in womb) is where the problems may be, we have absolutely no idea what it takes to bootstrap a brain.

There may be more to it than we know about, in fact there probably is.

Some theorists postulate the existence of firmware like code present in the DNA, there are many other even weirder theories, one of them may well be right. That would mean that we are simulating a bunch of hardware here, completely forgetting about the software required to run that hardware.

Thanks for the explanation on hypoxia, I did not realize that the damage was always irreversible, even when the period without oxygen is short.

Regarding the electro shock therapy, yes, that is true but it starts out assuming that the brains state is 'electrical', which it doesn't have to be at all.

It could be chemical, it could be based on standing waves in the neural tissue, there are a great many ways to store state.

For instance, one of the first computers that ever existed used standing waves in tubes filled with mercury to store its state.

So until we know what that state is and how it is stored we will not know how to reset it reliably, nor will we know if it is possible to restore such a state once it has been 'lost'.

'Grand mal attacks' or seizures seem to have something to do with feedback loops within the brain, cycles where the end result sets the whole thing off once again.

By necessity the brain has to contain many such potential loops and under normal conditions the feedback factor is kept in check, but with the right trigger pattern a series of self re-inforcing patterns can be established.

edit: have a look here: http://www.its.caltech.edu/~dadgrp/research/poc/index.html

there is a lot of interesting information in there, even though it is already old (10 years), the numerical data is still accurate.

>>Essentially the Blue Brain project is a study in emergent behaviour, simulate something and its behaviour should emerge. That is exactly what the project leader claims in the article linked, I don't think even they expect it to be human behaviour because that is limited to beings that are humans.

First, the neuron models will probably still be adjusted in a century. (For instance, I've seen papers just a few years old which discussed how water really worked; that doesn't mean the understanding of water isn't very, very good.)

Second, the brain researchers claim the research using new tools are going incredibly fast; see the claim in the interview of more information every year -- than in all of the 20th century! (So the article from ten years ago seems a bit outdated -- also, membrane receptors is a hard problem which just recently has even been touched. The Blue Brain people claims that they started with a molecular level emulation of the brain and then optimized it).

(Also note that the Blue Brains researchers are aware of differences between individual cells because of history, etc. They aren't idiots.)

>>The problem with the approach is that a simulation, by definition is not the real thing and that it will always end up being an approximation.

Third, I have a hard time thinking of a better way to learn than research groups like Blue Brain encoding the present knowledge in computer models -- to compare against the real systems. No one expects the models to be perfect at once.

Fourth, I trust the reviewers of the papers the Blue Brain research group writes to do basic reality checks, e.g. glial cell influences (when/if they are found to have a large influence). That would just be a factor of two in complexity, for parallel models like this, Moore's law works well...

Fifth, chemistry is generally slow -- there can't be too much happening in thought/reaction, so most stuff in a neuron's work is "prebuilt"; no, not with memcached but with structural changes while learning. (There is heavy evolutional pressure to optimize for speed.)

How good the emulations will be after refinement etc, is a much later question. But the researchers are evaluating (quite) modular systems "from below" -- they can't be too complex, because there just aren't enough building blocks (i.e. not enough genes).

It is a hard problem, but I don't see a reason why it should be impossible to reach good models. That doesn't matter, what is more important is that the researchers think so, too.

>>Regarding the electro shock therapy, yes, that is true but it starts out assuming that the brains state is 'electrical', which it doesn't have to be at all.

Well, give an alternative mechanism that isn't reset by electrical shocks or cooling down to low temperature (as with some surviving drowning victims in cold water or some operations.) (The point of Grand mal attacks was that I wondered if they were reset mechanisms.)

Note that changes in protein expression would be quite easy to find. And that human nerve cells aren't that different from other animals, which has been experimented quite heavily on. I can't have any ideas, here (also note the "fifth" point above.) Do you have anything except conjecture?

I don't have the time to keep discussing, sorry.

Edit: Cleaned up and was too hurried to mark it up carefully above. Sorry, if anyone ever reads this. :-)

we should probably take this off-line, or we'll end up finding the limits on HNs indent algorithm. You don't have an email address in your profile, so I can't mail you directly, mine is j@ww.com , drop me a line if you're interested discussing this further.

Thanks, but I'd better not. Lack of time and I don't really have that much insightful to say.

For fun I studied some chemistry and biochemistry a couple of years (I followed the subject before). My girl friend follows e.g. brain research in a similar way.

I'm mostly a fan boy in cellular biology and (for a few years more) less than that in brain research. :-)

> relatively cheap

And who said HN has a shortage of humor :)

We know that intelligence can beget other intelligence, just like we know that life can beget life. But so far nobody to my knowledge has succeeded in creating artificial life without starting from something else that was already alive.

Intelligence might pose similar challenges.

> to understand AI we must first understand ourselves,

And therein lies the rub. We may get the mechanics down to every last nut and bolt and we may still not understand how intelligence arises. Just like we can not bring a recently deceased human back to life again. We have all the pieces but we do not know how to put the spark back in.

Compare it with a computer analogy, broken for sure, like all analogies, but it may bring the point across:

If an intelligence is a product of both software and hardware, and we seem to transfer the software through our genes and use that to build hardware that comes 'pre-programmed' then that program might not just run on any hardware. It may make lots of assumptions that we can not easily simulate without starting with a live brain and that would be cheating.

If we can not reliably bring people out of a coma, all the pieces are still there and appear to be 'plugged in' then how can we expect to generate an AI for which we do not even have the pieces ?

If I have a computer running an os in ram that I keyed in one fine sunday morning using the bootstrap panel, and which I then destroyed, next switch the machine off while it is working great you would have to know how I got the initial program in there before you could attempt to duplicate it.

Call it the case of the missing scaffolding, the last coupld of billion of years of evolution put that program where it sits, and from what we can see living creatures can pass it to the next generation in one form or another.

But it seems to be terribly difficult to switch the machine back on once it is of.

To me strong AI and life are quite strongly intertwined, I doubt you can have AI without artificial life, and that in and of itself may be quite enough of a challenge.

There is a category for super Turing computation - hypercomputation. http://en.wikipedia.org/wiki/Hypercomputation. There is a whole hierarchy of such problems/machines http://en.wikipedia.org/wiki/Arithmetical_hierarchy

True. My impression was that the machines were not well defined though. That is, hypercomputation is mostly defined as things not computable by a Turing machine. In addition, I guess I was saying more that we don't know what category our mind would fall into if it is indeed hypercomputable, not that we don't have mathematical categories for noncomputable (not general recursive) problems. I edited the parent post for clarity, as well.

"However, we know that the resources required to build one are small -- humans do it all the time in a relatively cheap process called "sex." That is, the lower bound for creating AI is very small. "

This is false. Sex produces the DNA. A whole lot more energy/resources go(es) into making an adult brain.

A system that is "not computable or equivalent to a Turing machine" is not "within reach of implementation within a computer system". (Unless, of course, the Church-Turing thesis is false. Do you have any evidence of this?)

If we take the assumption that the mind is a physical system, then it would still only be a matter of time before we found the reason for the difference in computational ability

I don't see how the latter follows from the former. The physicality of the mind says nothing of its computability. You're assuming the computationalist axiom that reality is completely computable, which is quite an assumption.

The brain is still a combination of proteins, amino acids, and a few other chemicals. We would just have to move to using chemical computers (which are already being researched)

We already produce hundreds of thousands of them every day. The process is rather fun. As some point the definition of success has been stretched so far as to be meaningless.

Reply on news.ycombinator.com