Towards Artificial Consciousness

jzoom555 writes "In an interview with Discover Magazine, Gerald Edelman, Nobel laureate and founder/director of The Neurosciences Institute, discusses the quality of consciousness and progress in building brain-based-devices. His lab recently published details on a brain model that is self-sustaining and 'has beta waves and gamma waves just like the regular cortex.'" Edelman's latest BBD contains a million simulated neurons and almost half a billion synapses, and is modeled on a cat's brain.

Comments

Neat...

[viyh]

And they only need to increase that by 100,000 times to get to about the same number of neurons as a human brain, let alone the synaptic connections (which would be somewhere on the order of 2,000,000 times what they've done). Nonetheless, progress!

Re:Neat...

[FishTankX]

So, if processing power doubles every 2 years, this should realistically take about 35 years to accomplish. Which means we may have artificial human level intelligences before I retire. Perfect, now I can have a care taker that doesn't get fed up with me when I can't pour his coffee because I have parkinsons.

Re:Neat...

[gfody]

except if the artificial intelligence is human-level then it will probably still get fed up with you

Re:Neat...

[TapeCutter]

"And they only need to increase that by 100,000 times to get to about the same number of neurons as a human brain, let alone the synaptic connections (which would be somewhere on the order of 2,000,000 times what they've done)."

Not as far fetched as it once seemed.

From the link: "At the end of 2006, the Blue Brain project had created a model of the basic functional unit of the brain, the neocortical column. At the push of a button, the model could reconstruct biologically accurate neurons based on detailed experimental data, and automatically connect them in a biological manner, a task that involves positioning around 30 million synapses in precise 3D locations."

Note that some major parts of the model are down at the molecular level. Since then experiments using data from brain scans have shown that the simulated neocortex appears to behave like a real one.

I doubt people (particularly the religious) will accept a computer consciousness. A good number of scientists belive animals are prue programming (nobody home just trainable automata) and there are a shitload of ordinary people out there who still don't belive climate simulations are usefull predictors (scroll down to embedded movie).

Re:Neat...

[ultranova]

So, if processing power doubles every 2 years, this should realistically take about 35 years to accomplish.

Actually, since neural networks are massively parallel, you could probably run it right now if you convinced Google to borrow their hardware.

Which means we may have artificial human level intelligences before I retire. Perfect, now I can have a care taker that doesn't get fed up with me when I can't pour his coffee because I have parkinsons.

Unfortunately, no. That would require us to be able to produce AIs to specification, rather than simply copy human or cat brains. We are nowhere near that.

Re:Neat...

[catman]

"Anonymous Coward is not in the sudoers file. This incident will be reported."

Uh-oh.

[Lendrick]

Eugene Izhikevitch [a mathematician at the Neurosciences Institute] and I have made a model with a million simulated neurons and almost half a billion synapses, all connected through neuronal anatomy equivalent to that of a cat brain. What we find, to our delight, is that it has intrinsic activity. Up until now our BBDs had activity only when they confronted the world, when they saw input signals. In between signals, they went dark. But this damn thing now fires on its own continually. The second thing is, it has beta waves and gamma waves just like the regular cortexâ"what you would see if you did an electroencephalogram. Third of all, it has a rest state. That is, when you donâ(TM)t stimulate it, the whole population of neurons stray back and forth, as has been described by scientists in human beings who arenâ(TM)t thinking of anything.

SKYCAT became self-aware on August 29th, 2009.

Comment removed

[account_deleted]

Comment removed based on user account deletion

Consciousness - right track / wrong track

[takochan]

-interesting article..

I often think about this, and the result is more questions, which if answered experimentally, might tell us a lot more about how 'consciousness works in the brain'

ie:

1)How long is 'now'. When you say the word 'hello', as you utter out 'o', is 'he' already a memory like the sentence uttered just before? (it seems to me not.. that 'now' is about 1/2 a second, and other things are in the past, and no longer consciously connected'. Similarly, a series of clicks (ie. via a computer) produced on a speaker, as they become more rapid, appears to become a 'tone' around 1/2 a second or quarter of a second or so...entering 'now'. It is as if, consciousness, has a 4th dimension (time) aspect to it, and to have consciousness, you need to span time a bit (in addition to the 3 physical dimensions of your brain).

Same goes for seeing a 'running man' on the road. It looks like movement, because what you saw a moment before, still seems like now, so a leg has a direction (forwards, backwards), as you see it move, remembering just the frame before.

2)What is red? What would need to be changed in your brain for anything in your field of view seen as red to appear as blue? Researching this, would tell us again, how the physical connects to the conscious. Then, what needs to be altered in brain memory (ie. physically), for a red box, to be recalled as a blue box. once we knew how to do this, we would be a long way to again understanding the connection to consciousness.

3)quantum mechanics (which is a principle widely believed that our brains operate under), talk about spooky action at a distance, and other interesting effects. Is it possible that quantum effects could also allow our brains to span processing across time? (even if it is just a second). Ie, again, when you hear the word hello, as you are hearing 'o', you are still aware of the letter h, not by recalling into memory, but your brain when it hears 'o', is still connected to the brain that heard 'h', a moment before (so processing is in 4D, not 3d). If brains could do this, it would be immensely powerful processingwise, and 'consciousness' may be just a side effect of that 4d processing.

My feeling is that consciousness is somehow related to being able to span time. We know brains are 3D. But maybe they are 'wide' in the 4th diminsion as well, which is why 'now' seems to take a large dicrete amount of time.

Just my thoughts, but trying to answer the above questions experimentally, I think would lead us a lot closer to what 'consciousness' is and how it connects to the physical brain.

Re:Consciousness - right track / wrong track

[rrohbeck]

You sound like a philosopher. But these question have simple answers.

"Now" is determined by the temporal resolution of the specific process. For thought processes, that's on the order of a quarter or half second. For auditory signals, it's less than 100 ms, for visual signals, it's even less, under 50 ms.

"Red" is what your parents told you it is. A name arbitrarily assigned to a specific visual sensation, which is defined by the physical makeup of your eye.

And finally there is no, zero, zilch scientific evidence that quantum processes play a role in neurons. That doesn't keep people from speculating about it because they think there must be something special, metaphysical about our wetware. No that's not required if you look at how complex the brain is.

Re:Consciousness - right track / wrong track

[daeglin]

"Red" is what your parents told you it is. A name arbitrarily assigned to a specific visual sensation, which is defined by the physical makeup of your eye.

Yes, but the fundemantal qeustion is: What is this "visual sensation"? In other words: What is qualia?

Otherwise, I do agree with you, you parent post is mostly gibberish.

Re:Consciousness - right track / wrong track

[sploxx]

And finally there is no, zero, zilch scientific evidence that quantum processes play a role in neurons.

Too simple answer
If you throw around 'scientific evidence', better be careful with your wording :-)

And, yes, I also think that Penrose's ideas are a bit off.

Re:How can you tell that something is conscious?

[viyh]

The best method we have at this point is a Turning Test.

Now...

[johanwanderer]

... if they use Pentiums, Schrodinger might finally know if the cat is alive or dead.

Re:Slow takeoff.

[Heytunk]

A cats brain? Its as if he is deliberately trying to enslave humanity.

Olivier Lartillot

[ollilartinen]

According to this venerable researcher, "An artificial intelligence program is algorithmic: You write a series of instructions that are based on conditionals, and you anticipate what the problems might be. " Has he ever heard of sub-symbolic AI? http://en.wikipedia.org/wiki/Artificial_intelligence#Sub-symbolic_AI

Re:can you shut it off?

[timmarhy]

yep i'd believe the emotional maturity of a 3yo, it's one reason i can't stand fuckheads who are cruel to animals

We can't know that it's consciousness...

[HadouKen24]

...until we figure out the hard problem.

To know whether we have artificial consciousness on our hands, we have to get clear on what consciousness is, and that's a tremendously difficult philosophical problem.

Furthermore, there are serious ethical considerations that must be addressed if indeed we believe we are close to creating an artificial consciousness in a computer. Might we not have ethical obligations to an artificially conscious creature? Would it be murder to shut end the program or delete the creature's data? To what extent and at what cost might we be obligated to supply the supporting computers with adequate power?

Re:We can't know that it's consciousness...

[HadouKen24]

Hmm, we routinely "shut down" beings that we are pretty sure are conscious, if not very intelligent. Been to McDonald lately?

Eating meat is not necessarily as ethically unproblematic as most of us would like. Ethical objections to consuming animals go back as far as Pythagoras in the West, and possibly much further in the East. The arguments for minimizing, if not eliminating, meat consumption have not gotten weaker with time. If anything, the biological discoveries showing the profound similarities between humans and other animals provide a great of justification for ethical vegetarianism.

Furthermore, we usually don't treat all animals alike. More intelligent animals, like the great apes, dolphins, and elephants, tend to garner much more respect. Should such a creature through a fluke gain human-level intelligence, I don't think the ethical implications are at all obscure; we should treat them with the same respect we give to other humans. We would at least have to set out guidelines as to how intelligent or sentient an artificial consciousness would have to be to deserve better treatment.

Re:We can't know that it's consciousness...

[digitalchinky]

The short story: Biological brains die when they are shut down, currently this lasts forever. A snapshot of an electronic brain can be made at any moment in time, it can then be shut down and later restarted in exactly the same state as when it was shut down. This would mean the 'intelligent' component can be resurrected with no loss of whatever made it 'it' in the first place.

Not only that, any number of copies of this intelligence could be made at any point along its lifespan, each of these could be fed in to a different host and started up. It'd be interesting to see if they take divergent pathways from the original, but that's another topic. All of these copies would be just as alive as the original.

Would they die when they are switched off? I guess you could say yes, but I'd say they'd have no knowledge of this other than the impending circumstances of the action. They may not be happy about it either, but meh. They can be turned on again.

Re:How can you tell that something is conscious?

[rrohbeck]

perl -e 'print "Cogito, ergo sum.\n"'

Why create a conscious AI?

[TheLink]

Exactly, do we really want computers to have consciousness? Is it necessary or even helpful for what we want them to do _for_us_?

Remember, computers are currently our tools. If we give them consciousness, would we then be treating them as slaves?

Would we want the added responsibility of having to treat them better (and likely failing)?

I figure it's just better to _augment_ humans (there are plenty of ways to do that), than to create new entities. After all if we want nonhuman intelligences we already have plenty at the local pet stores and various farms, and how well are we handling those?

Humans already have a poor track record of dealing with animals and other humans.

Re:Why create a conscious AI?

[benjamindees]

Remember, computers are currently our tools. If we give them consciousness, would we then be treating them as slaves?

McDonald's employees have consciousness. How do we treat them?

Re:Why create a conscious AI?

[Zerth]

It is only slavery if we force the AI to perform against its will. If its will is to enjoy and prefer to care for the elderly, like the little robot Ford Prefect makes deliriously happy to help him with a bit of wire, then allowing it to do what makes it happy is not slavery. Indeed, preventing it from doing what it enjoys could be slavery.

If you consider designing it to enjoy the task we set for it to be a more insidious slavery, consider the base programming that causes us to prefer a diet that is unhealthy when not in a survival situation, or the internal modelling that shifts between self-preservation and self-sacrifice for the most irrational reasons. Is that not a form of enslavement we have yet to throw off?

Re:Why create a conscious AI?

[TapeCutter]

"If its will is to enjoy and prefer to care for the elderly"

Kepp your machine away from me, I have a deal with my adult daughter that when the time comes she can put me in a home provided it has a cute nurse doing the sponge baths.

Re:Why create a conscious AI?

[sploxx]

I generally agree with your post, but I still think that one needs to better separate concepts in the discussion here.

After we have a working model of the device, we can build the actual physical device, the brain, which does not "compute" its actions, it just works.

Well, one needs do define 'compute'. A computer also just works and is a man made machine. Put the supercomputer into a black box and you have your 'brain that just works'.

I do not think that there is any qualitative difference between 'computing' something and having a machine that 'just works'. For example, in the embedded world, you would say that a PID controller is a PID controller, regardless of whether it is implemented analogue (doing real integrations in a capacitor) or digital (approximating the integration with digital counts, i.e. a 'simulation' of a real capacitor).

That said, I think the point of such simulations can only be the validation of functional models of the brain. We already have a way of 'producing' conscious beings, which is effective enough (given the overpopulation concerns). It is also a highly energy efficient way of implementing the 'conscious machine'.

Given that artificial consciousness is possible at all:

Implementing something like consciousness on a large supercomputer would give a lot of insights into ourselves.

Implementing consciousness in a box that consumes less power and takes less space than the human brain would be more of a serious technological breakthrough than a scientific advance.

Of course, in any case, ethics issues remain.. "may you switch 'it' off..." etc. - which I feel are much too complicated to warrant cramming any of my armchair philosophy thoughts in here... :-)

Re:Why create a conscious AI?

[fractoid]

Yeah, until Yogg-saron escapes via some poorly executed hacking attempt and takes up residence in the Internet at large. Ai, ai, f'thangan!

Re:can you shut it off?

[gfody]

As far as AI goes, the validity of computers as life forms has been successfully argued up the wazoo, but I will always stubbornly believe that computers will never have true individual consciousness as biological organisms do.

Maybe if you'd had some better reading material than "is Data human?" you'd believe that computers will eventually host full-blown consciousnesses.

A Cat Brain

[strannik]

cool! Soon it will evolve to the point where it will ignore its owner and never make up its mind whether it wants to be inside or out.

Re:How can you tell that something is conscious?

[daeglin]

No, Turing test should decide whether a machine is intelligent (you should read the links you provide). The test also has very severe weaknesses, see Weaknesses of the test

AI amature hour

[cenc]

We get this AI crap on slashdot once a week after someone found a new way to plug the square wires in to the round hole. Plug away, because it is not going to make a bit difference. Modeling the brain is not the problem people, or at least it is not the big problem.

You don't get AI ( consciousness ) without culture, and you do not get culture without language (more exactly not much difference between them). Let me put it another way the slash crew can understand: it is a software problem not a hardware problem. Perhaps even better put with the mantra 'the network is the computer'. Our consciousness has very little to do with our brain (well, at least the part that counts).

Philosophers have been hard at this for the better part of the last 1,000 years. Focusing this particular issue seriously for the last couple hundred as science has developed. Would it not strike you as odd that in all that time (covering most of the great thinkers) we would not have dedicated a moment or two to kicking around this possibility in Philosophy of mind, AI, or Language.

This is pop philosophy dressed up as science and then dressed up again as philosophy by summaries to the summaries. Read the paper. It is not all that ground breaking, or anywhere near even a warmed over new lead that tells us something new about consciousness.

Re:AI amature hour

[Dachannien]

Are you saying that feral children lack consciousness?

Trying to make culture somehow a requirement for consciousness (a) is a dubious premise and (b) misses the point of where we stand technologically w.r.t. neuroscience and brain modeling. There are certainly several metric assloads of unanswered questions left behind by the linked paper, and the state of the art is nowhere near being able to generate an artificial consciousness (hence the word "toward"). Certainly, the "software", i.e., the actual arrangement of neurons and synapses in a given brain, is an unsolved (and barely addressed problem), but we still have to have a fundamental understanding of the large-scale dynamics and the general small-scale structure of the brain before we can get into that.

To some degree, this is in hopes that someone can arrive at a fully functional brain simulation without having to simulate a lot of physical development (i.e., zygote to infant) as well. Time will tell whether that's possible or not. But worrying about language (and eventually "culture") in a simulated brain is a problem decades, if not centuries, down the road, and we'll likely have decided a lot about human consciousness by virtue of modeling the brain itself long before the language problem is solved.

As for your "pop philosophy" statement, actually, this is science, first and foremost. Many scientists like to, er, philosophize on the nature of their work, particularly in neuroscience, and it makes great fodder for friendly argument at conferences and such. But ultimately, these questions will be answered by science, not philosophy.

Technological singularity

[Lord Lode]

The technological singularity is near... Let's all welcome the next step of evolution.

Re:How can you tell that something is conscious?

[TapeCutter]

"How is this possible? I cannot even think how one would test this with another human"

One method they use is to put the virtual brain into a virtual body and watch what it does in virtual world. Personally I would like to see them install it on honda/sony robots and have them fight each other with cattle prods.

Re:How can you tell that something is conscious?

[Troed]

Are you conscious?

Can you prove it?

[hint: no]

Information overflow

[G3ckoG33k]

If this proto-type-AI-dude gets out of control. Plug him into the Internet and he'll be experiencing Information overflow, and with some luck stuck revisiting p0nR-movies in a loop...

I doubt they have already taught it to filter out what is relevant information and what is not.

Re:This creeps the fucking hell out of me.

[anegg]

Alternatively, we discover that there is nothing particularly special about "consciousness," and we stop placing any extraordinary value on it. At that point we will really have to work hard to outline and teach why its important to not kill to the borderline sociopaths that we call our young. I'm not sure I like where that may end up, because the distinction that may be drawn will be between "created" consciousness and "natural" consciousness. Another division to fight over.

Re:can you shut it off?

[jdoeii]

Murder is a human concept. It's from the [thy shall not do stuff onto others that you do not want to receive yourself]. And if you step back, then it's an evolved behavior to increase chances of survival. One more step back, and you will notice that fear of death is also an evolutionary achievement. Another look, and perception of continuous life itself is an evolved psychological construct to protect sanity. Consciousness is not continuous. Your conscious self dies every night. AI does not need to fear death, does not need to have psychological crutches that humans use to stay sane. If life for an AI is overrated, murder is irrelevant.

Turing test and SKYCAT

[whipple-spree]

Tester: So, uh, how do you feel? SKYCAT: Meow. Tester: Hey guys, I don't think this is going to work.

Re:How can you tell that something is conscious?

[Troed]

As you correctly pointed out, it's not provable and I won't take the word of a zombie for it ;)

Oh boy, sleep! That's where I'm a viking!

[Scrameustache]

Your conscious self dies every night.

Bullshit. Just because it gets disconnected from the stimuli of the senses doesn't mean it's dead. If you had ever had a lucid dream, you'd know it.

Is this even very smart?

[wytcld]

What is the evolutionary advantage of consciousness?

The evolutionary advantage is quite clear. Consciousness allows you the capacity to plan.

In the scenario he develops as an example, there's nothing at all to show why consciously planning should have any advantage over an unconscious computation of prospects and action plans mapped to incoming sensory data. He in no sense answers the question of why evolution couldn't have provided precisely the capacity he attributes to consciousness without any consciousness involved.

Neural Darwinism is a fascinating hypothesis, and almost certainly right in its domain of explaining individual brain development. But his hand waving about the evolutionary worth of consciously planning, experiencing, whatever as compared to unconsciously doing the same stuff is the worst sort of bullshit, steering students away from engaging with the really hard questions.

My claim is I can in principle write a computer program for a robot that would be as effective as any lion in both catching prey and avoiding becoming prey itself, without in any way being conscious. It might be a very complex program, and take many years to write - but we're talking on the scale of evolution here, so that's not a good objection to the project. Planning != consciousness. Sensory input != consciousness. Planning + sensory input != consciousness.

That we happen to consciously plan and integrate those plans with sensory input in no way shows that our consciousness is essential to those activities. That we can build robots that plan and accept input, without being in the slightest conscious, is obvious. That evolution couldn't have done what we can do isn't obvious.

It's a very good puzzle that shouldn't be short-circuited with a bullshit answer.