Inferior Creators

[sithum]

I've been thinking about the concept of inferior creators recently. It seems that when full human A.I. is achieved (which according to some sources will happen within the next 50 years) the scalability of the product will be such that human beings will eventually create a being that is superior to itself. Now, I realize that I'm being extremely ad-hoc here with the use of the terms inferior and superior. However if we assume that we can reduce humanity to a thinking machine then the A.I. that we eventually create will over time, through refinement, be a faster intelligence with a seemingly infinite memory.

Therefore we will have reached a situation where human beings have, through a process of technological improvements over time created a being that is superior to itself.

Now if we sort of extend this process backwards in time and consider all the creation stories that humanity has been bombarded with since the beginning of written history, is it reasonable to postulate the idea that we were created by a being(s) which was inferior to us?

All the creation stories I have ever encountered paint a picture of a creator who is omnipotent or in the very least extremely powerful. Is this assumption of a superior creator(i.e. a creator which is superior to its creations) consistent throughout all creation myths? Or are there creation stories out there that portray a creator as an entity which is 'less' powerful(or inferior) than that which the creator creates?

I was wondering also if this issue has been discussed at all in the mainstream yet?

Finally: has it ever been postulated by a creationist that our creator was inferior to us?

(for the record I consider this to be a total philisophical tangent and I do not mean to in any shape or form endorse the idea of a creator/creation story in the first place).

[Tucson_Tim]

It seems that when full human A.I. is achieved (which according to some sources will happen within the next 50 years)

(my bold)

And who are these sources?

I have my doubts about a "full human A.I." ever being developed, but if so, 50 years sounds very optimistic.

[lomiller1]

(my bold)

And who are these sources?

I have my doubts about a "full human A.I." ever being developed, but if so, 50 years sounds very optimistic.

At the rate of advancement seen over the last 50 years computers we will be able to build computers with as much raw processing power as the human brain within another 50.

First problem is that that advancement isn't sustainable over the next 50 years.

Second is that we don't really have sufficient understanding of what intelligence or consciousness is to write the code to reproduce it even if we had a computer with enough power to run the software.

Someone may be clever enough to write a program that could pass a Turing test even before this but the question about weather this is truly an AI will remain IMO.

[Rhaedas]

If you mean inferior as being less intelligent or self-aware, then yes, the process of evolution would qualify.

[Ara Pacis]

Now if we sort of extend this process backwards in time and consider all the creation stories that humanity has been bombarded with since the beginning of written history, is it reasonable to postulate the idea that we were created by a being(s) which was inferior to us?

Parents?

Okay, farther back? The parent's parents? Australopithecus?

[tashirosgt]

An interesting question for an AI created being, is what is a "being"? For example, a person can have a brain injury and lose control of a limb and also lose the feeling that the limb is part of his body. His doctors regard the appendage as part of him, but he does not. For an AI "being", what would it's opinion of its own extent be? And would that match its constructors opinions?

I tend to think of myself as the collection of things that I can immediately affect and control. But I have components that I can't consciously control - according to the opinion of the world. The parts that I can control, can only be controlled to a limited extent. For example, there is a limit to the precision that I can have when drawing something on paper. Opinions about what constitute "immediate" and what constitutes "control" can vary. Suppose an AI being has a very large time scale and a very lax definition of control. Since it can exert some influence on large set of things, it might regard them all as part of itself.

[Ronald Brak]

And who are these sources?

I have my doubts about a "full human A.I." ever being developed, but if so, 50 years sounds very optimistic.

Ray Kursweil and friends. And I'd be willing to bet fifty bucks that a computer will pass a Turing test within fifty years. Not that I know much about computing, but I base this conclusion on what I know of human intelligence.

[Tucson_Tim]

And I'd be willing to bet fifty bucks . . .

Sounds like this could be listed as one of the great scientific wagers of all time:

http://en.wikipedia.org/wiki/Scientific_wager

[Ronald Brak]

Finally: has it ever been postulated by a creationist that our creator was inferior to us?

Plenty of creation myths tell of the creation of people by various animals and/or forces that aren't necessarily intelligent, although they are generally regarded as powerful, though not necessarily bright. A case of unintelligent design.

[sithum]

Plenty of creation myths tell of the creation of people by various animals and/or forces that aren't necessarily intelligent, although they are generally regarded as powerful, though not necessarily bright. A case of unintelligent design.

So perhaps humanities pursuit of A.I. is unique in the sense that we will eventually create something that is more intelligent and more powerful than ourselves. The creator will be humbled by his or her creation.

Of course the A.I. could theoretically keep creating new and better A.I. infinitely. Each level of A.I. would be lesser than the next step in complexity. You would have an infinite number of inferior creators as the creation would get better and better with each step. Would a maximal intelligence ever be reached in this line of creation? Do we (humanity) have the ability to even conceive of what this maximal intelligence would be like?

I suppose it's easy to see that the types of numbers the A.I. could handle would get larger and larger and would be processed faster and faster..so from the standpoint of pure calculation it's easy to conceive how the A.I. could keep improving...

but how about abstraction.. how does one measure a machines ability to execute abstract reasoning? Is there a limit to how good abstract reasoning can be? Or is it infinite like it is with straightforward number crunching?

[pzkpfw]

So perhaps humanities pursuit of A.I. is unique in the sense that we will eventually create something that is more intelligent and more powerful than ourselves. The creator will be humbled by his or her creation.

I've read science fiction where this is essentially equated to continuation of evolution; that our current form essentially becomes a living fossil and the new form is the future of "humanity" (whatever that is...).

Interesting thoughts.

[sithum]

I've read science fiction where this is essentially equated to continuation of evolution; that our current form essentially becomes a living fossil and the new form is the future of "humanity" (whatever that is...).

Interesting thoughts.

Another question that I had was: will the survival instinct of the 'new form'(i.e. ith iteration of A.I.) look upon the 'old form'(i.e. the wetware human form.. which could eventually be the ‘living fossil’ as you say..) as a threat to its survival or a necessary asset? More generally, assuming that it is the (i-1)th iteration of A.I. which creates or gives rise to the ith iteration of A.I., will the previous iterations of A.I. be looked upon as a threat to survival or as a necessary asset by the subsequent iteration of A.I. ?

I suppose it could be a bit of both.

I mean, from my perspective, I think it would always be useful to have a few original wetware humans walking around because they would provide a guaranteed restore point if something went wrong with the A.I.'s hardware [EMP shockwaves come to mind].. then again.. I may be inherently biased at some gut level in the sense that I want the current form of my species to survive..I think that there is at least the potential that the new A.I. will see our existence as useful or necessary to its existence. To abstract this even more I’m going to say that it is possible that the (ith) iteration of A.I. could find the existence of the ([i-1]th) iteration of A.I. useful. [I’m not sure if ‘it’ is the right pronoun to use for an A.I. machine..].

On the other hand though, I wonder if 'fear of one’s creator would be part of the instinct of each new A.I. iteration' (It certainly seems that being intimidated by ones creator is a part of all of humanities creation stories so far...).. in this sense would the ith iteration of A.I. look upon previous iterations (including the original wetware) with fear?

It’s interesting because each A.I. iteration would theoretically have an infinite lifespan and therefore it is conceivable that the ith iteration would know about the existence of all previous iterations… therefore each iteration would be ‘aware’ of a finite series of creators of lessening complexity going back to the original wetware… but the link between iteration 1 (i.e. the first iteration of A.I. created by wetware) and the original wetware would be the least established assuming there were no advancements in cybernetics to go along with the development in A.I. .. so perhaps the wetware would always be looked upon as ‘special’…. Or possibly ‘enviable’…

I also wonder about the transference of the survival instinct from one iteration of A.I. to the next. Could this survival instinct get in the way of the generation of new iterations of A.I. from old one’s? Perhaps massive storage technology make the survival of old A.I. iterations trivial and easy to implement.. or perhaps the generation of a new A.I. iteration will take place gradually from the perspective of the old iteration of A.I. and therefore perhaps all this talk about iterations of A.I. is pointless.. but I think it might be relevant to examine this transference of survival instinct and how the new survival instinct will affect the A.I.’s ability to replace itself with new versions (since replacement can be viewed as destroying something old and creating something new.. perhaps programming the original A.I. with slight suicidal tendencies would be a good thing to ensure that the new A.I. wouldn’t have too much reservation destroying itself in order to upgrade itself…conversely it may be a good thing if subsequent iterations of A.I. were instinctively slightly hostile or even homicidal to previous iterations of A.I. so that upgrades could be achieved without any rejection from the system)… I wonder if all this could be implemented without an increasingly stronger survival instinct for each new iteration.. a survival instinct which increased with each new iteration could have disastrous consequences.

I also think it would be useful to speculate as to how fast the new A.I. would surpass our intelligence. If it did so gradually, perhaps the wetware version of humanity could grow and adapt with the machines for a time.. until the limits of the wetware architecture were reached.. of course this train of thought seems to lead naturally to eugenics.. darn..

[Ara Pacis]

Of course the A.I. could theoretically keep creating new and better A.I. infinitely. Each level of A.I. would be lesser than the next step in complexity. You would have an infinite number of inferior creators as the creation would get better and better with each step. Would a maximal intelligence ever be reached in this line of creation? Do we (humanity) have the ability to even conceive of what this maximal intelligence would be like?

[emphasis mine]

Yes, according to some religions. If you accept the postulate of the universe as simulation, it may even make logical sense.

[eburacum45]

Since the human brain exists and seems to work very well, I have no real doubt that one day we will not only reproduce it artificially, but will build devices which are more capable.

The problems with creating AI are considerable, however, and simple calculations like the ones Ray Kurzweil like to make, using Moore's Law to predict when computers will match the brain in processing power, are not adequate. Our brains, and the minds that are inside them, are the product of hundreds of millions of years of vertebrate evolution. To reproduce something which has taken so long to emerge might take hundreds of years of development -or even longer.

However I am pretty sure that very smart AI that are not based on the human model will emerge much sooner than that. Will they be self-aware and conscious? Probably not. But they may well have other useful attributes.

[sithum]

Since the human brain exists and seems to work very well, I have no real doubt that one day we will not only reproduce it artificially, but will build devices which are more capable.

The problems with creating AI are considerable, however, and simple calculations like the ones Ray Kurzweil like to make, using Moore's Law to predict when computers will match the brain in processing power, are not adequate. Our brains, and the minds that are inside them, are the product of hundreds of millions of years of vertebrate evolution. To reproduce something which has taken so long to emerge might take hundreds of years of development -or even longer.

However I am pretty sure that very smart AI that are not based on the human model will emerge much sooner than that. Will they be self-aware and conscious? Probably not. But they may well have other useful attributes.

It would be interesting to try to see if there are 'principles' or 'laws' of evolution much in the same way that there are laws of physics. This would probably all be discovered at the biochemical level.

If there are 'principles' of evolution that can be discovered perhaps they can be simulated (and possibly sped up with super computers). It would be a real feat of biochemistry to discover what these principles were.

(Definitely agree with your analysis that the Non-Human derived A.I. will have useful attributes)

[HenrikOlsen]

It would be interesting to try to see if there are 'principles' or 'laws' of evolution much in the same way that there are laws of physics. This would probably all be discovered at the biochemical level.

If there are 'principles' of evolution that can be discovered perhaps they can be simulated (and possibly sped up with super computers). It would be a real feat of biochemistry to discover what these principles were.

(Definitely agree with your analysis that the Non-Human derived A.I. will have useful attributes)

There seems to be, as there are stories that crop up repeatedly.

One such law is that optimization happens at the local rather than global level, so species develops in the direction of short term gains.

One example is the arms race you get when a prey animal hits upon growing armor as a defense against a predator and the predator hits on bigger weapons as a result, you get an arms race where both species are locked in a dead end competition for ever thicker hide and ever larger teeth and claws until both become so encumbered that the predator can't catch anything except it's opponent and it's opponent can't drag itself to new food fast enough at which point both go extinct.
It takes about 5 million years and has happened at least five times so far.

One consequence of the local optimization law is that if multiple solutions exist for a problem and the benefits of the initial steps of all of them are pretty much the same, then no matter the eventual benefits of one solution over another it's random which one is picked.

Another is that of there is only a few general solutions, dictated by physics,, it will be used repeatedly. (streamlining, eyes with lenses)

[sithum]

There seems to be, as there are stories that crop up repeatedly.

One such law is that optimization happens at the local rather than global level, so species develops in the direction of short term gains.

One example is the arms race you get when a prey animal hits upon growing armor as a defense against a predator and the predator hits on bigger weapons as a result, you get an arms race where both species are locked in a dead end competition for ever thicker hide and ever larger teeth and claws until both become so encumbered that the predator can't catch anything except it's opponent and it's opponent can't drag itself to new food fast enough at which point both go extinct.
It takes about 5 million years and has happened at least five times so far.

One consequence of the local optimization law is that if multiple solutions exist for a problem and the benefits of the initial steps of all of them are pretty much the same, then no matter the eventual benefits of one solution over another it's random which one is picked.

Another is that of there is only a few general solutions, dictated by physics,, it will be used repeatedly. (streamlining, eyes with lenses)

It would be interesting to see if the local optimization law could be confirmed using a random number seed in some of these experiments..

I was hoping that there would be a simulation that could be run with virtual grey matter in which the stimulus leading to a beneficial change in grey matter (i.e. higher intelligence) could also be simulated. In this way all one would need is to model a really primitive virtual brain and then allow the program along with the aforementioned laws of evolution to run at a really fast rate. This way the virtual brain would evolve to what we wanted it to (i.e. a fully functioning virtualized human brain) from some really simple 'base case'. Hopefully with super computers running in parallel it could be done quickly.

It may not be necessary to understand how the human brain works to build a fully functioning virtual brain on a computer. I think it may be possible, and perhaps simpler, to derive how the brain evolves according to a set of principles (and/or perhaps how the brain grows from fetus to adult again according to a set of principles) and then run a simulation on a piece of virtual grey matter and then let the virtual brain you want grow and develop on its own.

Trying to build a fully functioning human brain from the get go would be an enormous feat. But if we could let the brain build itself based on perhaps a DNA blueprint or a series of simulated evolutionary changes then we may get the result more quickly... even though we don't understand all the nuances of why the brain works the way it does... the area of brain semantic implementation: how memories are stored etc. is proving to be a very tough nut to crack. Maybe we can circumvent this by starting with a simple brain and then try to simulate what evolution would do to this brain over billions of years given billions of virtual selection pressures.

[eburacum45]

It would be interesting to try to see if there are 'principles' or 'laws' of evolution much in the same way that there are laws of physics. This would probably all be discovered at the biochemical level.

If there are 'principles' of evolution that can be discovered perhaps they can be simulated (and possibly sped up with super computers). It would be a real feat of biochemistry to discover what these principles were.

I think that programs involving a-life capable of evolution are one way of attempting to discover these 'laws'.
http://en.wikipedia.org/wiki/ALife
Alife seems likely to be a very interesting area of investigation, and some sort of artificial intelligence might emerge from rapid simulated evolution of this kind.
.
.

[sithum]

I think that programs involving a-life capable of evolution are one way of attempting to discover these 'laws'.
http://en.wikipedia.org/wiki/ALife
Alife seems likely to be a very interesting area of investigation, and some sort of artificial intelligence might emerge from rapid simulated evolution of this kind.
.
.

Very interesting area of study.

I was also wondering if there is a way to grow a brain in-vitro (perhaps starting with few stem cells) and record each step in the brain's formation (i.e. the structure of each neuron, its neural connections, the spacial orientation and position of each neuron). I know that MRI gives the overall structure of a brain, but what I'm after is a table of values from which a complete 3-D computer model of the brain could be produced where every single neuron and connection could be resolved. The model could then be made functional. I think that such a model could form the perfect basis for simulating a brain on a computer. Obviously it would be preferable, at least in the beginning, to do this with a brain which is relatively simple.

I'm sure most people have seen the rat cyborg that was built last year: http://www.youtube.com/watch?v=1QPiF4-iu6g

It would be cool if a functional 3-D computer model of the rats brain could be built using a method similar to the above. Then we could see if the simulated brain could be attached to a robot as the organic brain tissue was in this video.

[Crazy Tom]

I believe the Orion's Arm crowd uses "e" as the pronoun to refer to non-organic lifeforms. Ex: e's computer, e assured me, etc.

It's unlikely that each successive generation would view the past ones as threatening, as it's level of understanding would be so far above theirs, rather like comparing Homo Sapiens Sapiens to Australopithecus. More likely the higher AIs would use the lower ones for tasks that for the higher AIs might be mindless and boring, while for the lower AIs might be quite challenging. And the higher AIs would focus on what they considered challenging.

A possible symbiotic relationship?

[Ronald Brak]

Someone may be clever enough to write a program that could pass a Turing test even before this but the question about weather this is truly an AI will remain IMO.

I imagine that, like us, it will write itself. But anything that can pass a Turing test would definitely be AI and it would be strong AI, unless you are using a very unusual definition of strong AI. Strong AI is comparable to human level intelligence, although not necessarily the same as human intelligence. We have AI now and most people on this board probably use it every day.

[trinitree88]

I imagine that, like us, it will write itself. But anything that can pass a Turing test would definitely be AI and it would be strong AI, unless you are using a very unusual definition of strong AI. Strong AI is comparable to human level intelligence, although not necessarily the same as human intelligence. We have AI now and most people on this board probably use it every day.

Ronald Brak. I don't knowingly use any AI, but have a distinct memory of the aeronautics lab at MIT using the most advanced aerodynamics equations of the time, prove that with it's tiny little wings and big fat body....that the bumblebee could not fly. I doubt they will ever supercede a decent human brain, and there are an awful lot of them around. pete

[sithum]

I imagine that, like us, it will write itself. But anything that can pass a Turing test would definitely be AI and it would be strong AI, unless you are using a very unusual definition of strong AI. Strong AI is comparable to human level intelligence, although not necessarily the same as human intelligence. We have AI now and most people on this board probably use it every day.

We are on a slight digression here.. but since you mentioned the Turing test I have to ask: do you think that MSN and other chat programs make it easier for machines to pass the turning test? I always thought that really simple A.I. could be developed in order to simulate say a stripper that didn't speak English very well and was always drunk....many of the mistakes could be passed off due to knowing english as a second language and being inebriated.. you could also introduce skips in her black and white video feed that could be explained away due to the bad internet service in her country--this would allow flexibility in video sequences.. the revenue from this project could then be used to fund more complicated A.I. experiments.. how likely do you think it would be, given current A.I. technology, that the aforementioned stripper would pass the Turing test?

[novaderrik]

why is every big scientific advance always due to come "in the next 50 years"?
nuclear fusion has been that far off for that long now, and permanent human bases on other planets are always about that far off, too..
if we are going to make up arbitrary numbers for things that may or may not happen in the future, then why doesn't anyone ever use an oddball number- just once, i'd like to hear of something definitely happening within the next 37.428 years.

[Ara Pacis]

why is every big scientific advance always due to come "in the next 50 years"?
nuclear fusion has been that far off for that long now, and permanent human bases on other planets are always about that far off, too..
if we are going to make up arbitrary numbers for things that may or may not happen in the future, then why doesn't anyone ever use an oddball number- just once, i'd like to hear of something definitely happening within the next 37.428 years.

It's called "Orders of Approximation".

[Ronald Brak]

Ronald Brak. I don't knowingly use any AI, but have a distinct memory of the aeronautics lab at MIT using the most advanced aerodynamics equations of the time, prove that with it's tiny little wings and big fat body....that the bumblebee could not fly. I doubt they will ever supercede a decent human brain, and there are an awful lot of them around. pete

If you use search engines you're using AI.

[m1omg]

If you use search engines you're using AI.

He is just retelling an old, 20th century urban myth:
http://en.wikipedia.org/wiki/Bumblebee#Bumblebee_myths

Best MIT computer algorythms...yeah, we ABSOLUTELY trust that

[m1omg]

If you use search engines you're using AI.

He is just retelling an old, 20th century urban myth:
http://en.wikipedia.org/wiki/Bumblebee#Bumblebee_myths

Best MIT computer algorythms...yeah, we ABSOLUTELY trust that

[Ronald Brak]

We are on a slight digression here.. but since you mentioned the Turing test I have to ask: do you think that MSN and other chat programs make it easier for machines to pass the turning test? I always thought that really simple A.I. could be developed in order to simulate say a stripper that didn't speak English very well and was always drunk....many of the mistakes could be passed off due to knowing english as a second language and being inebriated.. you could also introduce skips in her black and white video feed that could be explained away due to the bad internet service in her country--this would allow flexibility in video sequences.. the revenue from this project could then be used to fund more complicated A.I. experiments.. how likely do you think it would be, given current A.I. technology, that the aforementioned stripper would pass the Turing test?

It has been suggested that a computer could learn to simulate a human by hanging out in chat rooms and making conversation.

As for the program that simulates a drunk paint remover who can't speak much Engish, a sucessful Turing test would only indicate that people couldn't tell the difference between the program and a real drunk paint remover who can't speak much English. Kind of like how Monty Python demonstrated that penguins are as intelligent as people who can't speak Engish. (Neither they or the penguins successfully answered any questions.)