How much better than a human can an artificial one be?

[Humphnaegal]

I was watching the video of Boston Dynamics bipedal robot: Petman, and I started to think about the potential of robots, and how far it could go in emulating humans... and then bettering what evolution managed to do with limited resources, time, and mishap.

How smart, fast, and strong will the laws of physics ultimately allow humans 2.0 to be?

[Sarevok]

I don't think "artificial humans" are possible. You can never make anything near as good as the human mind with software.

[Simon_Jester]

I don't think this question really has an answer. Too much depends on technological limits we can't predict: what will be the most powerful computers we ever build? What will be the most efficient algorithms we ever develop? What kind of compact power sources or storage devices can exist?

[MrDakka]

That question would probably be better answered after the technological singularity (if you believe in such a thing) which is based on a lot of disruptive technologies that could drive us into extinction before we can get a handle on them.

However as Simon Jester pointed out, at this point, its wild ass guesses for now. As you stated yourself, its ultimately the laws of physics that confine us. If in the far future you want to be an AI god, all you need to is breach several toposophic levels and get yourself a Matrioshka brain.

If you want to speculate about the future of humanity, head over to Orions Arm Universe Project. They have the whole human 2.0 and up thing figured out at least in terms of the laws of physics.

[Skgoa]

The problem lies in the question itself: humans are very very bad at many things. We just like to ignore our limits and think of ourselves as the pinnacle of life. There is no reason an artificially created intelligence/life-form/mech-body to cary our brains around should be anything like we are right now. So what you are really asking is "can you build something that has the exact same downsides as a regular human body but better?" And that's a contradiction.

e/ Having said that, we can also look at it from a psychological/sociological angle. I.e. it may very well be that humans in the future don't want to make to obvious changes to their outward appearance. This introduces an upper bound, since there would be no market for improvements that go wildly beyond what human bodies could achieve naturally. Better memory, better imune system, and better senses, should be less affected by this, but e.g. higher strength then you would need to lift a washing machine up the stairs would not give an ROI for average people. Augmentations would be highly regulated and probably have legal limits, anyways, since they could become dangerous to non-augmented humans.

[Skgoa]

gettoedit: Actually, I would expect an anti-augmentation backlash. Just think about how emberassing it is right now to be found out to have fake hair or having to take Viagra. And religious people couldn't very well explain why they think God did a bad job.

[cosmicalstorm]

Our brain is very limited, it trucks info around at 150 meters/s, the speed of a photon is 299.792.458 meters/s...

[someone_else]

You can never make anything near as good as the human mind with software.

I'd like to know the exact reasoning behind this. (emphasis mine)
I mean, when you know where every cell is and what all connections and interactions are, you can do the fuck you want to make it better or emulate it in a computer.

And I don't think thare is an upper limit in what can be discovered. Maybe won't happen tomorrow, but there are no time limits in the OP.

Actually, I would expect an anti-augmentation backlash.

I would expect the augmented to do some shit and end up as ruling class. Not necessarily noticed by the idiots or the religious.
There is obvious pressure to not make the augmentation obvious. And I'm ready to bet augmentations are "inherited" (i.e. chosen by the parents for their future children) and not "removable" (as all radical modifications), so even when an augmented guy is discovered what the hell you want to do with him? He is like that due to his parent's choice and you cannot remove the augmentation without killing him.

[Scrib]

I think that if we can create artificial intelligence or augment human consciousness we'd probably get rid of a bunch of psychological leftovers, if you will,that were useful in our past but worthless today. Racism, our brains searching for positive patterns to reinforce our beliefs, the unreliability of memory, restlessness in children during extended periods of learning, sociopathy, lack of empathy for people outside our range of vision, whatever malevolent force makes people in groups lose their individuality and being fucking crazy, lack of forward thinking and impulse decisions, etc.Those alone would be more effective in making a "better" human being than any purely physical alteration. Of course, this doesn't bode well for anyone who doesn't want to get these changes, since it could be considered good for society as a whole.

Of course, I'm an eternal optimist. More likely we get a bunch of people trying to "perfect" the human race and fucking it up on the level of those forced sterilisations.

[Humphnaegal]

Our brain is very limited, it trucks info around at 150 meters/s, the speed of a photon is 299.792.458 meters/s...

This is one of those physics fundamental limits I'm thinking of. When you copy the pattern prediction algorithms of the human brain in the future and have this massively parallel computer with the smallest transistors possible, you should get something able to do everything we can brain wise, but almost 2 million times faster. Is that the limit? (at least that can be fit in a human head per the topic)

As for physical capabilities, would a future robotic human be comic book superhero like in its capabilities? Probably not. There seems to be a limit in terms of chemical bond energies, so humans 2.0 won't be clearing buildings in a single bound.

Are there some things I'm missing here?

[Ariphaos]

No.

The human brain is built as a hundred billion nodes which are each able to tell whether or not a given set of firings falls on one side of a threshold or another, up to a few hundred times per second. That sort of thing is extremely powerful for certain tasks - you'll be lucky to see a tenfold increase in capacity with the same amount of heat output, much less a millionfold. On the other hand, for some things, like say, typical math, it's a Rube Goldberg device. A millionfold increase would be rather pathetic.

A surprising amount of evolution has selected not for better performance, but rather, better heat dispersal. This sort of thing is why it's so hard to judge just how much better an artificial human or human augment could be. There are some areas where you can get a quantifiable answer to some extent (replace skeleton with titanium, replace the portion of our brain that does math with an actual cpu, general anti-aging stuff, upper limits to vision and other senses, etc.) but outside of visual resolution (something like 20/6 or so for the human eye, depending on wavelength), there are no easy equations.

[Ziggy Stardust]

It really depends on how you define "better," and what you actually consider to be an improvement. If you mean better at "being human," whatever that nebulous concept is, the answer is probably no.

At this point in time, our understanding of the human brain's mechanisms is relatively primitive compared to what you would need to actually replicate it mechnically/digitally/whatever. We simply don't know enough. And our current technology isn't advanced enough to do so even if we had an idea what direction to go in.

Is it technically possible? Well, yes. In the same sense that any science fiction concept is technically possible. Given no limits technological and scientific progress it will likely happen. Is that a reasonable thing to predict? The jury is out.

[K. A. Pital]

We made a software simulation of a rat brain neural network. Given we can simulate a neural network in principle, the creation of a human-like simulation is a matter of computing power and correct architecture, and not much more. Moore's law takes care of the first and brains take care of the second.

[Ziggy Stardust]

We made a software simulation of a rat brain neural network.

I assume you are referring to the Blue Brain Project? If so, this is not entirely accurate. The simulation is of a neocortical column, the smallest and most simplistic functional unit of neural architecture. The simply scaled their model of the column up by a magnitude of 100 (or 1000, can't remember the exact numbers). It is not clear that a column is really representative of the entire brain (and, in fact, there is plenty of evidence that parts of the brain involved with language, higher reasoning, and theory of mind are significantly more complicated in the way they connections between and across columns operate). And this is just on the cellular level, the project hasn't begun to look at the molecular/gene expression level. When it comes right down to it, the architecture of parts of the human brain is still not entirely understood.

[PeZook]

It's obviously physically possible to make a human brain, though, so it's just a matter of understanding and technology to recreate it. We will, eventually, barring a civilization-destroying cataclysm.

[K. A. Pital]

Yeah, but shouldn't we actually center on simulating the cellular level, since that might be well enough to create functioning virtual neural networks? I'm not saying that we understand the architecture right now, but once we find it out (and we will), the question will more or less be reduced to sheer computing power.

[someone_else]

the point is that all stuff going on in synapses is a chemical affair, and that neurons do interact with hormones and other chemicals. And how a cell reacts to chemical stimuli is determined by what receptors it has and how they work. So you also have to add the molecular and genetics understanding to get something working.

It's much easier to start from scratch if you want a neural network. This way you avoid having to understand all the pointless crap evolution left in us.

[Sarevok]

You can never make anything near as good as the human mind with software.

I'd like to know the exact reasoning behind this. (emphasis mine)
I mean, when you know where every cell is and what all connections and interactions are, you can do the fuck you want to make it better or emulate it in a computer.

And I don't think thare is an upper limit in what can be discovered. Maybe

I will believe it when I see it. Has anyone managed to do it so far ? No. Untill then it is just fanxiful speculation like. Warp drives and time travel.

[PeZook]

I will believe it when I see it. Has anyone managed to do it so far ? No. Untill then it is just fanxiful speculation like. Warp drives and time travel.

What?

Dude, 138 million human brains are produced annually.

[Humphnaegal]

Yeah, warp drives and time travel are not known to be possible*, but brains actually exist and can be studied, and are being studied in ongoing programs right now.

I find it a stretch to imagine we won't have machines with the mental capabilities of humans by 100 years from now. I just wonder how much further it can progress after that, and what sort of fundamental limits to reality would hold it back, but human level is quite possible... since it already exists.


*Warp drives are only theoretically possible providing you are able to possess something not known to exist - negative mass matter. Time travel is only possible in the sense that you can change at a different rate due to relativity.

[cosmicalstorm]

Yeah, warp drives and time travel are not known to be possible*, but brains actually exist and can be studied, and are being studied in ongoing programs right now.

I find it a stretch to imagine we won't have machines with the mental capabilities of humans by 100 years from now. I just wonder how much further it can progress after that, and what sort of fundamental limits to reality would hold it back, but human level is quite possible... since it already exists.


*Warp drives are only theoretically possible providing you are able to possess something not known to exist - negative mass matter. Time travel is only possible in the sense that you can change at a different rate due to relativity.

Going by the principle of mediocrity, it would be massively surprising if our brain, by a accident of nature happened to be the end point of intelligence. More likely, we occupy just a very small part of the total parameter space of possible of mind architectures. We know our brains are modular. We already have many algorithms that are clearly super-human in their ability to perform various tasks.
If you had root access to a brain, that seems almost certain to create a powerful feedback loop.
The fact that so many people, even famous scientists go about as if human mind architecture was the apex of creation, is really strange to me.

[PeZook]

It is an apex of evolution, in a sense: while birds, for example, have more efficient brains, we're the only civilization-building tool users on the planet.

So while we can't do math for shit, we still can do it way better than any other living organism. And not just math, but solve a lot of various problems from all disciplines.

The human brain is a very generalist tool that doesn't truly excel at anything, but can handle a staggeringly wide spectrum of wildly different problems with decent competence. It also has many faults, like energy use and heat dispersal (wonder how much improvement would be gained by simply making it possible to use larger portions of the brain at the same time...) which could almost definitely be elliminated in any actually engineered system, since the thing's ALSO optimized for working with limited energy supplies.

But it's not, strictly speaking, wrong to marvel at the brain. So far we're still below it, technologically.

[Skgoa]

We made a software simulation of a rat brain neural network. Given we can simulate a neural network in principle, the creation of a human-like simulation is a matter of computing power and correct architecture, and not much more. Moore's law takes care of the first and brains take care of the second.

And this, ladies and gentlemen, is why we don't let non-CS people make judgements about computer science topics. There is no reason why the leap you made should hold true, since complexity doesn't work as linearly as you think. (We also did not simulate a rat brain, but a neural net that had as many nodes as a rat brain has but it did not correctly model the internal state of the nodes.)

[Ariphaos]

It didn't even model as many nodes as the rat brain had, nor did it do so at the same speed.

Someone on Kuro5hin commented a decade ago, I think in response to the same story, that he would be impressed when a neural net could properly model a bee.

[Ziggy Stardust]

There is no reason why the leap you made should hold true, since complexity doesn't work as linearly as you think.

Not to get too off-topic, but this point is a good one. One thing that has always bothered me about the Ray Kurzweil school of transhumanist thought is the incredibly simplistic assumptions about the development of complex biological, psychological, and technological systems.

(We also did not simulate a rat brain, but a neural net that had as many nodes as a rat brain has but it did not correctly model the internal state of the nodes.)

To build on this, the model they used was based purely on sensory neurons (IIRC neurons specifically associated with input from the whiskers), which may not at all operate the same way as those associated with "higher" brain functions. Furthermore, from what I've read about the project (I could be wrong about this point, though), they haven't made much effort in modeling how clusters of neurons interact, focusing instead on a model centered around the idea that a single specific stimulus induces a single specific response in a single specific neuron, when any neuroscientist worth their salt can tell you is not an accurate representation of neural function.