Only if there isn't a point after which no further improvement is possible. I hope there isn't. But even if there is, it might just be enough to make such calculations take a few orders of magnitude less than the age of Universe.:-)
...kind of makes your original argument a bit moot, considering the proposed mathematical model really only applies to QM interactions
Not really. Consider: the best way to algorithmically test an airplane, if by best we mean the most accurate, would be to simulate every particle of it dealing with other particles. That's not the best in terms of speed though, so that we have useful simplifications in the form of the laws of aerodynamics. Those aren't as accurate as the former, but enough so. Similarly, once we really understand how cognition works (again, it doesn't necessarily require for us to go down into QM to find it out) we most probably will be able to get simplifications of cognitive laws that'll make reimplementing it in the form of AIs much easier.
There's also the possibility for some genius theoretical researcher to come about having figured out the simplified laws of cognition way before we have managed to build simulations of cognitive systems. Such an outrageous achievement isn't unheard of. Einstein fully figured out General Relativity without as much as looking at a solar eclipse. Hadn't he existed we'd have figured it out eventually for sure, but it'd have taken massive crunching of untold amounts of raw data not quite fitting previous theories until down the line someone got to the fundamental equations. Lacking an Einstein of cognitive sciences though our best hope is to go for the massive amounts of data and try painstakingly figuring things out from those.
The only difference in my view is that your copy has a different PID, or in other words, a different soul.
Technically you're inverting some of the classical distinctions between soul and substance. The notion was the the soul was the defining aspect, equivalent to a piece of code or data; the matter was the usually undifferentiated whatever that'd be shaped into something specific by a soul, roughly equivalent to a hardware not running anything; and a substance would be both joined. I guess your PID analogy would be equivalent to the substance then, not the soul.
I'm afraid the burden of proof rests on you, my friend. At this time nobody has done any of those things successfully.
Actually, it has been done for some very small brains, in the order of a few dozen neurons. It's a matter of scale and raw processing power more than anything else. But yeah, to reach the level of complexity of the human brain it'll take a long time.
You miss the point: until we can violate the current model of relativity (i.e. somebody invents a Tachyon-based computer), or something equally as wild, there is no way to "shortcut" our way into simulating interactions at that level with that kind of scale without spending orders of magnitude more, in this case, space-time.
It doesn't have to be at the same scale neither in size nor in time. We need a simulated brain to understand in details how a brain works. Maybe said simulation needs to be be so detailed as to require emulation all the particles in their Quantum interactions, but that's the worst case scenario, in all likelihood a simulation at a higher level of complexity will suffice. Even so, supposing we'll need a down-to-the-particles simulation, nothing requires us to build one with the same size of an actual brain or running at the same speed. It could take a whole city-sized computer running at 1/1,000,000 the actual speed of the brain and we'd still be able to gather enough information from it over the years to reduce the original complexity to fundamental algorithms and equations, and then reduce these further, until we managed to gather a comprehensive theory of human cognition.
At this point reimplementing the devised high level operations in more efficient ways would be a walk in the park, the original supercomputer not needed anymore.
If you're right, neither of us will ever know it. If I'm right, I get to taunt you about being wrong for all of eternity.
No problem, I'll enjoy our future arguments.;-)
Besides, I trust you'll see me suffering and, feeling merciful, will gently ask God to let me go. What request He, being also very merciful, will most certainly grant. Right?
if **everything** is a machine then the distinction is **pointless**
Nope. Saying that everything is a machine is a shorthand for saying that everything that happens can be reduced to the operations (machinations) of its components. But I get your point. What you call a machine I'd call an artifact, meaning a machine developed by human efforts, as opposed to, let's say, other living beings developed through the trial and error of natural selection, or the non-living beings produced by the raw nature laws. In other words, my scheme is this:
That somewhat Platonic scheme means that by figuring out math laws we also figure out all that can be known about physical laws and by extension about everything in the universe. "By definition." So, yeah.:-)
AI is incapable of doing anything a human didn't decide for it first programtically in the code.
Wrong. That'd mean that human beings are incapable of doing anything evolution didn't put into us by trial and error. That simply isn't true. As we managed to escape the grips of natural selection starting our very own artificial selection by way of reflection, AIs will escape the grips of our explicit instructions once we tell them how to reprogram themselves. If we do things right they'll keep carrying some of our values because they so which, much like we ourselves carry on evolution's "desire" for us to reproduce, but that'll be it.
simulating particles with particles is, well, just plain redundant, and might as well just be done experimentally.
Not quite. The advantage is that you can insert probes in any part of the code and watch how sets of things happen, then develop generalizations from these deeply granular observations that would otherwise be very difficult to come about by observing the real system working. I bet neuroscientists would give four limbs to be able to study an interruptible, debuggable fully simulated brain.
But then there are ethical questions involved too, as a fully simulated brain would be a real person after all. Studying it this deeply might be construed as a very sick form of torture. Even so, having the ability to simply log all the activity of all the molecules in a simulated brain, even if you don't touch it and let it be its own person (even if in slow motion), would be extraordinarily useful.
If a problem would take many times the age of the universe to calculate, then dividing that time by a small factor will not help much.
True enough. However, the fact that actual proteins manage to do it in a microsecond or less suggests there's a very easy way we just aren't aware of yet. Consider: it can be the case that the development of new mathematical tools works as a kind of Moore's Law for scientific research, making thing exponentially simpler with every step. This new mathematical tool by itself isn't much, but if the next improves the previous one by 1000 times, then next by another 1000 times, and so on and so forth, at some point we might reach usable speeds even not taking into account future Quantum computers.
Evidently the rate at which mathematical breakthroughs happens is by no mean predictable, so this might well be wishful thinking.:-)
Your mind manipulates meta-physical things all day long such as concepts, numbers, ideas, time etc.
Yep, but that doesn't mean those don't are efficiently causal in relation to us rather than formally causal. There are formalizations of causality that remove the need for a time component so that pre-space-time abstract (for us) causes can indirectly influence us without really being from a separate domain than our physicality. Additionally, it's evident that much of what we take to be metaphysical entities are effects from the fact that our brain, being relatively small in relation to the whole of reality, must out of necessity employ compression techniques so that a single pack of neurons and synapses can refer back to a ton of stuff out there. It'll take a lot of effort to eventually differentiate what's really metaphysical in all we consider so from what's just a bunch of compression artifacts, and then to identify the proper causal sources of our awareness of said actual metaphysical entities.
The magical indestructible personality-carrying dust naive religious persons believe in.
But yeah, I can also discuss non-religious, technical concepts of soul if you wish. Which one do you prefer? Plato's? Aristotle's? Plotinus's? Augustine's? Thomas Aquinas's? Duns Scotus's? Descartes's? Spinoza's? Leibniz's? Kant's? Some other? Those are all thankfully outside the minor considerations I made, and personally I'm quite fond of Aristotle's as it's remarkably compatible with modern computational models, but that's only in principle, as in practice I'm really more in the Buddhist anatman ("no-self") camp.
by definition, that's false.
I love "by definiton" replies! The next step is quoting a dictionary!
You also make the false assumption that AI = singularity. It does not. For some interpretation of singularity to do not need AI. I don't know which definition you are using.
Exponential intelligence explosion via self-upgrading general AIs.
If you think you will upload your brain and then YOU will be in the computer, you are wrong. A copy of you will exist on the system. One that diverges from the meat bag version immediately. Lets say you just uploaded you memories and responses into a machine, are you telling me the next day you don't mind if someone puts a gun in you mouth and pulls the trigger?
I'd mind because I have lizard brain instincts that trigger fight or flight responses. In the non-idiot part of my mind, however, I'll know another branch of me is safe, so I won't consciously worry too much. It's better however if by that point we managed to have merging abilities too so that my other selves can update this me and I those mes.
A middle ground solution I hope will happen before that will be the possibility of a gradual digitalization of the brain by replacing blocks of neurons or even individual ones in a gradual fashion until the whole brain becomes a parallel CPU. That'd be psychologically comfortable in the beginning due to the cognitive defects that cause us to not internalize that the Ship of Theseus is a false paradox. Followed then by the excision of useless cruft such as the aforementioned lizard-brain impulses, after which doing basic stuff such as backing up ourselves, instantiating new copies, halting and deleting other, exchanging information between them etc. will all become non-issues.
Unless you think 'who you are' can be moved from your brain into the machine becasue it's all part of an incorporeal being. You don't believe in incorporeal beings, do you?
Yep! In my code! Which can be run on any number of hardwares and instances if only we manage to copy it right!
The carefree manner in which you laid out the plan for out-dating our current wetware could be the way you come off as a singularity nut;)
Except I haven't laid out anything, I just made a small remark.:-) But the thing is, unless something is physically impossible, saying it'll happen isn't nuttiness. Defending the possibility of FTL travel is probably nuts, but general AI? Far from it. As the old saying goes, if it happened once it can certainly happen again. There's nothing inherently special about consciousness if something as (literally) brain dead as natural selection managed to produce it. A few more decades or, worst case, centuries, and we'll have it figured out.
There's this technical sense yes, but the far more common usage of the expression is as "there's a bunch of complex stuff going on that results in 'x', I'll say that's an 'emergent property', be done with it and not care at all in reducing 'x' to its individual components". This is pretty common in AI, were lots of researches play with throwing together neural networks to see what "emerges" from them (usually nothing) rather than going for the far more onerous route of step-by-step deconstructing 'x' into its single individual components so as to then become able to programmatically reconstruct 'x' or any variation desired, and having it work as expected. In short, it's usually an alternative name for what in another field certain folk call "irreducible complexity".
The human concept of "a soul" is an emergent property of high order intelligence.
You know that the "emergent property" expression is technobabble, right? It's the same kind of other contentless words such as "soul" and "magic", i.e., something that seems to be an answer to a question but isn't and in fact has the only purpose of nudging you into stopping asking the difficult questions.
this probably doesn't cut approaching that at all though. it might be a needed stepping block for making it possible at all though...
Certainly. As with other technologies we start by copying nature up to the point we understand the subject matter well enough to leave the natural solution behind. Airplanes don't flap wings after all.
even then that 100 years by moore's law guess was flawed.
Ah, of course. It's just the worst case scenario: emulating every particle interaction in a fully simulated version of a physical brain. The real solution is probably going to be orders of magnitude simpler, and that before we start improving things further.
Are you a singularity nut or just a misguided computationalist?
A singularity nut. Your brain is a machine. It can be understood, decompiled, analyzed, improved and reimplemented. You're already an AI running on appropriate (and at some point in future becoming outdated) hardware.
Unless you believe in souls. Do you believe in souls?
No idea what that means, but doesn't it sound cool?
I wonder how much this math simplification (?) will allow us to accelerate physical systems emulation, not to mention approaching general AI. The upper bound for emulation of the human brain in the absolute worst case scenario, the one that supposes consciousness arises from individual particles interacting rather than from a higher level of organization such as synapses signaling, was thought to become computable by the old method in about 100 years (if Moore's law holds until then). I'd love to know how many performance doublings this one will allow us to cut from that. The Singularity is one step closer!:-)
Also, computing proteins folding is probably going to get a serious performance boost too. If this proves to really work genetic engineering is going to enter a new phase.
It looks like Wolfram was onto something in A New Kind of Science with his approach to replacing complex equations with simple rules.
I'd say Plato (perhaps Pythagoras) was onto something when he basically said that math is the fundamental everything of everything. Yep, the guy was wrong on the details, but what damn fine intuitions he managed to have 2400 years ago. No matter what we do we always end up referring back to him...
Slashdot Mirror
This would basically mean that P=NP
Only if there isn't a point after which no further improvement is possible. I hope there isn't. But even if there is, it might just be enough to make such calculations take a few orders of magnitude less than the age of Universe. :-)
...kind of makes your original argument a bit moot, considering the proposed mathematical model really only applies to QM interactions
Not really. Consider: the best way to algorithmically test an airplane, if by best we mean the most accurate, would be to simulate every particle of it dealing with other particles. That's not the best in terms of speed though, so that we have useful simplifications in the form of the laws of aerodynamics. Those aren't as accurate as the former, but enough so. Similarly, once we really understand how cognition works (again, it doesn't necessarily require for us to go down into QM to find it out) we most probably will be able to get simplifications of cognitive laws that'll make reimplementing it in the form of AIs much easier.
There's also the possibility for some genius theoretical researcher to come about having figured out the simplified laws of cognition way before we have managed to build simulations of cognitive systems. Such an outrageous achievement isn't unheard of. Einstein fully figured out General Relativity without as much as looking at a solar eclipse. Hadn't he existed we'd have figured it out eventually for sure, but it'd have taken massive crunching of untold amounts of raw data not quite fitting previous theories until down the line someone got to the fundamental equations. Lacking an Einstein of cognitive sciences though our best hope is to go for the massive amounts of data and try painstakingly figuring things out from those.
in the end it just instructions executed on a box made by humans... doing nothing humans did not tell it to do
Oh, UNpossibly beautiful! Let's rephase it then: AI is all an expression of natural selection pressures! Like you!
Glad you agree!
LOL!
The only difference in my view is that your copy has a different PID, or in other words, a different soul.
Technically you're inverting some of the classical distinctions between soul and substance. The notion was the the soul was the defining aspect, equivalent to a piece of code or data; the matter was the usually undifferentiated whatever that'd be shaped into something specific by a soul, roughly equivalent to a hardware not running anything; and a substance would be both joined. I guess your PID analogy would be equivalent to the substance then, not the soul.
I'm afraid the burden of proof rests on you, my friend. At this time nobody has done any of those things successfully.
Actually, it has been done for some very small brains, in the order of a few dozen neurons. It's a matter of scale and raw processing power more than anything else. But yeah, to reach the level of complexity of the human brain it'll take a long time.
You miss the point: until we can violate the current model of relativity (i.e. somebody invents a Tachyon-based computer), or something equally as wild, there is no way to "shortcut" our way into simulating interactions at that level with that kind of scale without spending orders of magnitude more, in this case, space-time.
It doesn't have to be at the same scale neither in size nor in time. We need a simulated brain to understand in details how a brain works. Maybe said simulation needs to be be so detailed as to require emulation all the particles in their Quantum interactions, but that's the worst case scenario, in all likelihood a simulation at a higher level of complexity will suffice. Even so, supposing we'll need a down-to-the-particles simulation, nothing requires us to build one with the same size of an actual brain or running at the same speed. It could take a whole city-sized computer running at 1/1,000,000 the actual speed of the brain and we'd still be able to gather enough information from it over the years to reduce the original complexity to fundamental algorithms and equations, and then reduce these further, until we managed to gather a comprehensive theory of human cognition.
At this point reimplementing the devised high level operations in more efficient ways would be a walk in the park, the original supercomputer not needed anymore.
If you're right, neither of us will ever know it. If I'm right, I get to taunt you about being wrong for all of eternity.
No problem, I'll enjoy our future arguments. ;-)
Besides, I trust you'll see me suffering and, feeling merciful, will gently ask God to let me go. What request He, being also very merciful, will most certainly grant. Right?
if **everything** is a machine then the distinction is **pointless**
Nope. Saying that everything is a machine is a shorthand for saying that everything that happens can be reduced to the operations (machinations) of its components. But I get your point. What you call a machine I'd call an artifact, meaning a machine developed by human efforts, as opposed to, let's say, other living beings developed through the trial and error of natural selection, or the non-living beings produced by the raw nature laws. In other words, my scheme is this:
(machination?) -> physical laws -> (machination) -> evolution -> (machination) -> human beings -> (machination) -> artifacts -> (machination)
And here it stops for now, until we manage to develop general AIs, at which point their own machination will lead to something else.
2nd, the universe is more complex than mathmatics can describe by definition...
"By definition" is always the 2nd worst kind of answer. The worst kind of quoting dictionaries.
Suppose, for an hypothesis, that the questioning "machination?" above comes as this:
(machination?) -> math laws -> (machination) -> physical laws -> (...)
That somewhat Platonic scheme means that by figuring out math laws we also figure out all that can be known about physical laws and by extension about everything in the universe. "By definition." So, yeah. :-)
AI is incapable of doing anything a human didn't decide for it first programtically in the code.
Wrong. That'd mean that human beings are incapable of doing anything evolution didn't put into us by trial and error. That simply isn't true. As we managed to escape the grips of natural selection starting our very own artificial selection by way of reflection, AIs will escape the grips of our explicit instructions once we tell them how to reprogram themselves. If we do things right they'll keep carrying some of our values because they so which, much like we ourselves carry on evolution's "desire" for us to reproduce, but that'll be it.
simulating particles with particles is, well, just plain redundant, and might as well just be done experimentally.
Not quite. The advantage is that you can insert probes in any part of the code and watch how sets of things happen, then develop generalizations from these deeply granular observations that would otherwise be very difficult to come about by observing the real system working. I bet neuroscientists would give four limbs to be able to study an interruptible, debuggable fully simulated brain.
But then there are ethical questions involved too, as a fully simulated brain would be a real person after all. Studying it this deeply might be construed as a very sick form of torture. Even so, having the ability to simply log all the activity of all the molecules in a simulated brain, even if you don't touch it and let it be its own person (even if in slow motion), would be extraordinarily useful.
If a problem would take many times the age of the universe to calculate, then dividing that time by a small factor will not help much.
True enough. However, the fact that actual proteins manage to do it in a microsecond or less suggests there's a very easy way we just aren't aware of yet. Consider: it can be the case that the development of new mathematical tools works as a kind of Moore's Law for scientific research, making thing exponentially simpler with every step. This new mathematical tool by itself isn't much, but if the next improves the previous one by 1000 times, then next by another 1000 times, and so on and so forth, at some point we might reach usable speeds even not taking into account future Quantum computers.
Evidently the rate at which mathematical breakthroughs happens is by no mean predictable, so this might well be wishful thinking. :-)
Your mind manipulates meta-physical things all day long such as concepts, numbers, ideas, time etc.
Yep, but that doesn't mean those don't are efficiently causal in relation to us rather than formally causal. There are formalizations of causality that remove the need for a time component so that pre-space-time abstract (for us) causes can indirectly influence us without really being from a separate domain than our physicality. Additionally, it's evident that much of what we take to be metaphysical entities are effects from the fact that our brain, being relatively small in relation to the whole of reality, must out of necessity employ compression techniques so that a single pack of neurons and synapses can refer back to a ton of stuff out there. It'll take a lot of effort to eventually differentiate what's really metaphysical in all we consider so from what's just a bunch of compression artifacts, and then to identify the proper causal sources of our awareness of said actual metaphysical entities.
The brain is not a machine, it's a chemical factory. If you can't even tell that difference then you should probably shut up.
LOL! Warning, warning! Self-righteousness detected!
define soul.
The magical indestructible personality-carrying dust naive religious persons believe in.
But yeah, I can also discuss non-religious, technical concepts of soul if you wish. Which one do you prefer? Plato's? Aristotle's? Plotinus's? Augustine's? Thomas Aquinas's? Duns Scotus's? Descartes's? Spinoza's? Leibniz's? Kant's? Some other? Those are all thankfully outside the minor considerations I made, and personally I'm quite fond of Aristotle's as it's remarkably compatible with modern computational models, but that's only in principle, as in practice I'm really more in the Buddhist anatman ("no-self") camp.
by definition, that's false.
I love "by definiton" replies! The next step is quoting a dictionary!
You also make the false assumption that AI = singularity. It does not. For some interpretation of singularity to do not need AI. I don't know which definition you are using.
Exponential intelligence explosion via self-upgrading general AIs.
If you think you will upload your brain and then YOU will be in the computer, you are wrong. A copy of you will exist on the system. One that diverges from the meat bag version immediately. Lets say you just uploaded you memories and responses into a machine, are you telling me the next day you don't mind if someone puts a gun in you mouth and pulls the trigger?
I'd mind because I have lizard brain instincts that trigger fight or flight responses. In the non-idiot part of my mind, however, I'll know another branch of me is safe, so I won't consciously worry too much. It's better however if by that point we managed to have merging abilities too so that my other selves can update this me and I those mes.
A middle ground solution I hope will happen before that will be the possibility of a gradual digitalization of the brain by replacing blocks of neurons or even individual ones in a gradual fashion until the whole brain becomes a parallel CPU. That'd be psychologically comfortable in the beginning due to the cognitive defects that cause us to not internalize that the Ship of Theseus is a false paradox. Followed then by the excision of useless cruft such as the aforementioned lizard-brain impulses, after which doing basic stuff such as backing up ourselves, instantiating new copies, halting and deleting other, exchanging information between them etc. will all become non-issues.
Unless you think 'who you are' can be moved from your brain into the machine becasue it's all part of an incorporeal being. You don't believe in incorporeal beings, do you?
Yep! In my code! Which can be run on any number of hardwares and instances if only we manage to copy it right!
Do you believe someone assembled us?
Yes: natural selection.
The carefree manner in which you laid out the plan for out-dating our current wetware could be the way you come off as a singularity nut ;)
Except I haven't laid out anything, I just made a small remark. :-) But the thing is, unless something is physically impossible, saying it'll happen isn't nuttiness. Defending the possibility of FTL travel is probably nuts, but general AI? Far from it. As the old saying goes, if it happened once it can certainly happen again. There's nothing inherently special about consciousness if something as (literally) brain dead as natural selection managed to produce it. A few more decades or, worst case, centuries, and we'll have it figured out.
See here for my reply to a similar response.
See here for my reply to a similar response.
No technobabble needed.
There's this technical sense yes, but the far more common usage of the expression is as "there's a bunch of complex stuff going on that results in 'x', I'll say that's an 'emergent property', be done with it and not care at all in reducing 'x' to its individual components". This is pretty common in AI, were lots of researches play with throwing together neural networks to see what "emerges" from them (usually nothing) rather than going for the far more onerous route of step-by-step deconstructing 'x' into its single individual components so as to then become able to programmatically reconstruct 'x' or any variation desired, and having it work as expected. In short, it's usually an alternative name for what in another field certain folk call "irreducible complexity".
"I can only think of these two possibilities. That must mean those are the only two! I reject one, therefore, the other one. I'm a genius!"
LOL at such "deeply wise" non-answers. :D
The human concept of "a soul" is an emergent property of high order intelligence.
You know that the "emergent property" expression is technobabble, right? It's the same kind of other contentless words such as "soul" and "magic", i.e., something that seems to be an answer to a question but isn't and in fact has the only purpose of nudging you into stopping asking the difficult questions.
this probably doesn't cut approaching that at all though. it might be a needed stepping block for making it possible at all though...
Certainly. As with other technologies we start by copying nature up to the point we understand the subject matter well enough to leave the natural solution behind. Airplanes don't flap wings after all.
even then that 100 years by moore's law guess was flawed.
Ah, of course. It's just the worst case scenario: emulating every particle interaction in a fully simulated version of a physical brain. The real solution is probably going to be orders of magnitude simpler, and that before we start improving things further.
Are you a singularity nut or just a misguided computationalist?
A singularity nut. Your brain is a machine. It can be understood, decompiled, analyzed, improved and reimplemented. You're already an AI running on appropriate (and at some point in future becoming outdated) hardware.
Unless you believe in souls. Do you believe in souls?
No idea what that means, but doesn't it sound cool?
I wonder how much this math simplification (?) will allow us to accelerate physical systems emulation, not to mention approaching general AI. The upper bound for emulation of the human brain in the absolute worst case scenario, the one that supposes consciousness arises from individual particles interacting rather than from a higher level of organization such as synapses signaling, was thought to become computable by the old method in about 100 years (if Moore's law holds until then). I'd love to know how many performance doublings this one will allow us to cut from that. The Singularity is one step closer! :-)
Also, computing proteins folding is probably going to get a serious performance boost too. If this proves to really work genetic engineering is going to enter a new phase.
It looks like Wolfram was onto something in A New Kind of Science with his approach to replacing complex equations with simple rules.
I'd say Plato (perhaps Pythagoras) was onto something when he basically said that math is the fundamental everything of everything. Yep, the guy was wrong on the details, but what damn fine intuitions he managed to have 2400 years ago. No matter what we do we always end up referring back to him...
There will always be criminals, sociopaths and crazy people in any society therefore there is a need to NOT restrict gun ownership
FTFY.
Are you addicted to guns? No it's not the same thing.
Criminals are.