The Blue Brain project

[Xuenay]

Video lecture: Anders Sandberg on whole-brain emulation

[Akkleptos]

Video lecture: Anders Sandberg on whole-brain emulation

Sorry, no. It won't extend your life indefinitely, unless you can come to terms that it would be a "brain emulation"... Which is not necessarelity the same person on a diferent medium.


If you listen closely, objections like the ones posed by present physics amd common stupid interpretations of what naturely-plausible things. But reach out for nationwide news.

Other than that more God-like technology is necessary to get to the level where the mind contents of a brain can be readable by machines.

For more detail, have a look at Forum_Troll's posts before you post a pointless ambitious yet worthless....

[Sarevok]

While existing people may not be "transferred" to digital media the emulated copies are still a type of person. They really are immortal and superhumanly intelligent while retaining memories of life as a flesh and blood human. I wonder sometime if I ever had a digital twin like that how differently he would see the world. Afterall he does not get sick, need to eat or sleep and lives so long he can seriously contemplate a trip to a nearby star system for a vacation journey. Yet at same he has my memories. So would he care about all my human friends, relationships, hobbies etc ?

[Starglider]

They really are immortal and superhumanly intelligent while retaining memories of life as a flesh and blood human.

Conventional brain simulations are not superhumanly intelligent. They may think (much) faster or slower than a human depending on the efficiency of the simulation and the capability of the hardware, but an accurate simulation by definition has human-level intelligence. To get superhuman intelligence you have to start modifying the structure of the simulation, beyond the original biological template. This will be an unreliable and somewhat risky process if you don't know exactly what you're doing (quite likely; one of the major attractions of brain simulation is that it can work without a genuine understanding of high level brain function).

So would he care about all my human friends, relationships, hobbies etc ?

Initially yes. There will be a divergence over time which is really no different to the change that you'd experience if dropped into a new city/culture/etc, just faster.

[Junghalli]

Sorry, no. It won't extend your life indefinitely, unless you can come to terms that it would be a "brain emulation"... Which is not necessarelity the same person on a diferent medium.

Even if we leave aside the whole "is a perfect copy you or not" philosophical question, we could get around that whole issue by replacing the brain function piecemeal instead of all at once and keeping the subject conscious throughout the procedure. It would probably be more technically challenging, but it would spare us that whole headache very neatly.

Conventional brain simulations are not superhumanly intelligent. They may think (much) faster or slower than a human depending on the efficiency of the simulation and the capability of the hardware, but an accurate simulation by definition has human-level intelligence. To get superhuman intelligence you have to start modifying the structure of the simulation, beyond the original biological template.

Here's something I'm curious about. Is there not a fairly well-accepted theory in computer science that any Turing machine can emulate the operation of any other Turing machine given sufficient time? And wouldn't this indicate that intelligence is fundamentally a matter of processing speed?

[Starglider]

Is there not a fairly well-accepted theory in computer science that any Turing machine can emulate the operation of any other Turing machine given sufficient time?

Yes

And wouldn't this indicate that intelligence is fundamentally a matter of processing speed?

No. A Turing machine does not do anything useful (on its own). Specific programs do useful things; some programs are much more useful than others, and some do the same things but with widely varying efficiencies. Right now we're still trying to build any sort of general AI, and structure is far more important than computing power.

Computing power (not the same thing as 'processing speed', which is starting to get into serial/parallel distinctions) will only correlate closely with intelligence once AI design approaches theoretical optimums. The kind of systems we can currently make near-optimal (e.g. individual algorithms) are several orders of magnitude less complex than even a minimalist general AI. Programs require storage as well as processing of course, but for most AI work memory size is not the primary hardware limitation.

[Akkleptos]

Even if we leave aside the whole "is a perfect copy you or not" philosophical question, we could get around that whole issue by replacing the brain function piecemeal instead of all at once and keeping the subject conscious throughout the procedure. It would probably be more technically challenging, but it would spare us that whole headache very neatly.

Nice. Nevertheless, how is the difference between you and a chocolate shake the same as the one between me and an example our latest organic/inorganic food substitute?
Would you like to provide part of the material-stuff for such a substitute?

[Starglider]

Is there not a fairly well-accepted theory in computer science that any Turing machine can emulate the operation of any other Turing machine given sufficient time?

Yes

Not relevant to your question, but I should note that this is only true for a universal Turing machine with an infinite tape. A computing system is 'Turing complete' if it can function as a UTM, though of course all real computers are restricted to finite tapes. People often just say 'Turing machine' when they mean 'universal Turing machine'.

Nice. Nevertheless, how is the difference between you and a chocolate shake the same as the one between me and an example our latest organic/inorganic food substitute? Would you like to provide part of the material-stuff for such a substitute?

[Jules] English, motherfucker! Do you speak it? [/Jules]

[Sea Skimmer]

One major issue with implants or nanotech like this, is heat dispersion. This massively limits what you can do computationly wise, both in complexity and magnitude.

The human brain just isn't design to radiate more than a few of watts, possibily tens of watts.

What if you just slightly artificially cooled the blood flowing through the brain to keep temperatures in balance? I’d imagine localized overheating could still be trouble (a job for nanoscale heat pipes maybe), but it’d be a huge improvement over relying on natural radiation. I’m thinking a system which cools bloodflow at several points too, so you don’t get too cold at the blood inlet side. Placing the extra holes in the skull should be no big deal if you’ve already reached the point of implanting such a complicated computer system.

[Starglider]

What if you just slightly artificially cooled the blood flowing through the brain to keep temperatures in balance?

The feedback loops regulating temperature, pressure and osmotic balance in the brain are really complicated and delicate, even moreso than the rest of the body. I don't think you'd want to mess with them like that when it really isn't necessary. Putting minimal hardware in the brain (i.e. the neural interface electrode array) and the computer and power bricks in some less critical location (e.g. the abdominal cavity, connected by a fibre optic cable) is highly preferable. Not only do you greatly simplify the cooling issue, you can also replace the main components much more easily when they malfunction, get damaged, become obsolete or suffer from implant rejection.

[Junghalli]

Nice. Nevertheless, how is the difference between you and a chocolate shake the same as the one between me and an example our latest organic/inorganic food substitute?
Would you like to provide part of the material-stuff for such a substitute?

If you're talking about what I think you're talking about this is an engineering issue of getting a sufficiently good brain simulation, not a fundamental problem with the concept itself. Unless you believe in some sort of magic that keeps a human mind from being run on a computer no matter how good you simulate the operations of the brain in the computer.

As for the second part, I'm not sure whether I'd like to be the first human test subject for a destructive consciousness-continuous upload (it would depend on how much confidence I had that the researchers had been able to work out all the serious bugs before going to human testing), but if it was a reasonably reliable technology I wouldn't have a problem with it. Though I personally would prefer the bio-immortality route to the immortality by uploading route if I had a choice; I'm rather attached to my physical humanity.

[Junghalli]

Double-posting because the edit time ran out.

No. A Turing machine does not do anything useful (on its own). Specific programs do useful things; some programs are much more useful than others, and some do the same things but with widely varying efficiencies. Right now we're still trying to build any sort of general AI, and structure is far more important than computing power.

Ah, OK. Here's another question:

When you say an upload might start out running faster than the original biological brain but by definition would start out with the same intelligence, how exactly are you defining intelligence? I would think that an increase in clock speed would by itself constitute an increase in intelligence, because it would allow the entity to process data and solve problems faster.

[Starglider]

I would think that an increase in clock speed would by itself constitute an increase in intelligence, because it would allow the entity to process data and solve problems faster.

A simple increase in clock speed increases the number of problems that can be solved in a given time, but there will be very rapidly diminishing gains in the quality of the solutions or the maximum difficultly of the problems that can be tackled. Consider being locked in a room and trying to write a novel, for either a month, year, decade and century. Your ouput given a year will likely be much better than if given only a month, but a whole decade won't produce anything like as much improvement, and there's only a limited amount of polishing you can do in another 90 years. Sure, you could try to teach yourself to be a better novelist, but that's limited by both natural talent and lack of ability to interact with the external world. Another way of using the computing power is creating a few thousand clones of you, getting them all to write novels, and picking the best output, but again; rapidly diminishing gains. There are ways around these problems but the solutions are more complicated than 'ramp up the clock speed'. Ultimately the human brain just isn't designed to be scalable, so a radical architectural change is needed to get efficient use out of very fast computing hardware.

[Junghalli]

Consider being locked in a room and trying to write a novel, for either a month, year, decade and century. Your ouput given a year will likely be much better than if given only a month, but a whole decade won't produce anything like as much improvement, and there's only a limited amount of polishing you can do in another 90 years.

But with a higher clock speed you could get more work done in the same amount of time. If you were a writer you could write more novels, if you were an engineer you could complete more projects, if you were a mathematician you could solve more problems. Wouldn't the ability to solve more problems in the same amount of time be an increase in intelligence?

There are ways around these problems but the solutions are more complicated than 'ramp up the clock speed'. Ultimately the human brain just isn't designed to be scalable, so a radical architectural change is needed to get efficient use out of very fast computing hardware.

Yes, but this just means that increasing the clock speed by itself is not the optimal means of increasing intelligence. It doesn't mean that being able to solve more problems in the same amount of time isn't in itself an increase in intelligence.

Or are you defining intelligence as something detached from the simple ability to solve X problems in Y time?

[Starglider]

Wouldn't the ability to solve more problems in the same amount of time be an increase in intelligence?

'Intelligence' is a very fuzzy word. It is true that more intelligent people can usually solve any given problem faster. However, we don't consider existing computer programs intelligent despite the fact that they can do their assigned tasks millions of times faster than a human, nor would we normally consider an idiot-savant with a gift for factoring numbers in their head to be more intelligent than a pure mathematician, who needs a calculator to factor large numbers but can understand and invent sophisticated theorems. AI got burned with equating raw speed to intelligence early on (e.g. thinking that being able to brute-force chess meant that general AI was nearly solved), so usually we equate intelligence to the hardest task a system can perform (in some reasonable time), not the throughput of easy tasks per second. After all, algorithms are the hard bit, throwing more computing power at the problem is just a budget issue.

It doesn't mean that being able to solve more problems in the same amount of time isn't in itself an increase in intelligence.

Thinking really fast is a transformative capability, there is no doubt about that, but it doesn't compromise our ability to understand what the system is doing. Qualitative improvements in intelligence are something else entirely. A cyber-chimp that thinks like a normal chimp but 100 times as fast will not be writing poems or designing oil refineries or doing anything else that a normal chimp would not be able to do, eventually. A human is a qualitatively different thing. Similarly, improvements in the structure of intelligence will produce qualitatively better reasoning than humans are capable of. I would think that this site would find that particularly easy to comprehend given how often you all (well, DW at least) bitch about humans being moronic irrational emotional prejudiced wastes of oxygen.

[Junghalli]

'Intelligence' is a very fuzzy word. It is true that more intelligent people can usually solve any given problem faster. However, we don't consider existing computer programs intelligent despite the fact that they can do their assigned tasks millions of times faster than a human, nor would we normally consider an idiot-savant with a gift for factoring numbers in their head to be more intelligent than a pure mathematician, who needs a calculator to factor large numbers but can understand and invent sophisticated theorems.

Yes, but increased clock speed would result in an across the board increase in problem-solving capacity, not just in a handful of areas like you get with an idiot savant or a narrowly specialized computer program. Wouldn't superior across the board problem solving be superior intelligence?

Thinking really fast is a transformative capability, there is no doubt about that, but it doesn't compromise our ability to understand what the system is doing. Qualitative improvements in intelligence are something else entirely. A cyber-chimp that thinks like a normal chimp but 100 times as fast will not be writing poems or designing oil refineries or doing anything else that a normal chimp would not be able to do, eventually. A human is a qualitatively different thing. Similarly, improvements in the structure of intelligence will produce qualitatively better reasoning than humans are capable of. I would think that this site would find that particularly easy to comprehend given how often you all (well, DW at least) bitch about humans being moronic irrational emotional prejudiced wastes of oxygen.

Well, yes, there's more to increasing intelligence than just increasing clock speed, but nevertheless, if you had two minds that were identical except for one thinking 10X the speed, wouldn't the 10X faster one be more intelligent simply because it could process data and solve problems more quickly? That's what I've been trying to get at.

This does remind me now of a debate that happened on SB concerning OA's Archai and the claim of the thought processes of these extremely powerful AI being incomprehensible to humans, specifically the question under discussion was whether they were incomprehensible to us in the same way that our motivations would probably be incomprehensible to a dog, or whether the thought process of a very powerful AI could theoretically be understood by a human, albeit if it was very powerful working through its chain of reasoning might well take totally impractical amounts of time (i.e. millions of years to simulate a few seconds worth of thought). I'd be interested to hear what you think of that question.

Hmm, maybe we should move this to the AI FAQ thread.

[Starglider]

Yes, but increased clock speed would result in an across the board increase in problem-solving capacity

'Capacity' is a vague combination of two distinct parameters, 'speed' and 'difficulty' (where difficulty is of course a complex property), you cannot conflate them or easily trade one for the other.

Well, yes, there's more to increasing intelligence than just increasing clock speed, but nevertheless, if you had two minds that were identical except for one thinking 10X the speed, wouldn't the 10X faster one be more intelligent simply because it could process data and solve problems more quickly?

Yes, but there is only a limited amount you can do with more speed, compared to a more sophisticated reasoning capability. A lot of the off-hand dismissals of the power of superintelligence on this board come from the notion that a Culture mind (say) is just a human speeded up a million times relative to the external universe. Speed is a big deal in social impact terms, but it's small beans compared to the real meat of the matter.

whether the thought process of a very powerful AI could theoretically be understood by a human, albeit if it was very powerful working through its chain of reasoning might well take totally impractical amounts of time (i.e. millions of years to simulate a few seconds worth of thought).

A human can function as a UTM - see the Chinese Room Argument - so given n trillion years and an indefinite amount of pens and paper it is always possible for a human to simulate a computer (of course the error rate will be hideous). However the human does not have the 'understanding' that the simulated system does any more than the arithmetic unit in your CPU understands the physics model of your FPS game. Humans would not be able to have the understanding that such an AI does because our brains would not have the capacity to hold it, and I suspect that even with a complete dump and indefinite time we couldn't get enough of the formal system into our heads at once to even make our own simplified summary of it. We'd just have to settle for asking the AI to summarise its logic for us.