Von Neumann Machines, Nanotech and AI

[Hustler]

I was reading an article the other day about space exploration and the search for extraterrestrial life. In doing so, I got pulled into a web of links about self-replicating spacecraft, called von Neumann probes. The idea is that a probe would be built and sent to a nearby star system. It would then assemble components to build replicas of itself, sending those to other nearby star systems. These, in turn, would continue the chain. In time, with the growth rate such a system of machine reproduction could produce, an intelligent civilization could use such probes to cover the galaxy, gaining information about all star systems contained within. These probes are an example of machines known as self-replicating machines, or universal assemblers who, besides their primary function, are also constructed with the ability to create more of themselves.

In 1981, Frank Tipler used the idea of von Neumann probes as the basis for extraterrestrial intelligence not existing. It was his argument that, given a moderate rate of machine exploration and reproduction, von Neumann probes should be common throughout the galaxy and we should have encountered one or more by now. A compelling argument, if we assume such probes would be used by an extraterrestrial intelligence for space exploration. This hypothesis was later rebutted by Carl Sagan and William Newman, wherein they pointed out that, had some intelligent civilization created such probes in the galaxy's past, the machines would have consumed most of the mass of the galaxy by now, which would have given any intelligent race pause before creating a device that would destroy the entire galaxy. Thus, they would not have been used.

Many variants on this theme have emerged in the circles of speculative intellectuals since. One is a death machine that is designed to seek out and destroy life elsewhere in the galaxy. Another is a sentinel, which halts reproduction upon encountering a civilization and simply observes it from an undetectable vantage. An objection to the sentinel, though, is that it would take but one machine malfunctioning and creating endless replicas before we're back to the same situation as before, where the machines have consumed most of the mass of the galaxy. And this got me thinking...

Suppose you created a nano-von Neumann machine. Suppose it was tiny in scale, perhaps around the same size as a virus or bacteria. Suppose you placed it in an isolated ecosystem with material enough to reproduce billions and billions, or even trillions of times (not hard to do, considering a typical virus has a mass of under a femtogram). What would happen? In the arguments above, where people were discussing malfunctions leading to a sentinel machine reverting to a standard von Neumann probe, there is the implication that a malfunction would serve much as a mutation does in biological organisms. Would malfunctions be inevitable and would they lead to evolution? Surely some malfunctions would prove beneficial to the machine and, upon replicating itself, it's own branch of the machine kingdom would gain an edge against the original species. Is such a machine really all that different from the simplest of biological organisms, such as nanoarchaeum or viruses or the entities that straddle the line between life and non-life, such as nanobes or nanobacteria?

If we take this a step further, if we consider this to be the creation of a silicon-based system of living creatures (as opposed to our own, carbon-based system), would we expect intelligence to be an eventual evolutionary product of this machine kingdom? Is this the true path to AI; is machine evolution the real answer, where malfunctions or bugs in the machine's code (it's DNA, if you will) lead to an evolutionary progression toward intelligence, or is the method of tinkering with neural nets still the best option? If we hoped to spawn an entire kingdom of machines, what are some of the parameters we should impose on the initial machine race? How are they powered?

[MacGuffin]

So... how can we prove humans are not the result of nanoprobes gone wrong?

[Hustler]

So... how can we prove humans are not the result of nanoprobes gone wrong?

How can we teach you about the folly of disproving negatives?

[MacGuffin]

How can we teach you about the folly of disproving negatives?
Via corporal punishment?



If you really wanted to spawn a new kingdom of machines, why impose any limits at all? Let them figure out (if "figure" is the correct term) what they need to propagate. Of course, that probably means signing the death warrant for your own species.

Better to isolate them in some bubble where they can't get to you... hopefully.

Hell, create a whole new dimension for them... to... play... in...




Are you there God? It's me, MacGuffin.

[Hustler]

If you really wanted to spawn a new kingdom of machines, why impose any limits at all? Let them figure out (if "figure" is the correct term) what they need to propagate. Of course, that probably means signing the death warrant for your own species.
Some parameters do have to be set.


Better to isolate them in some bubble where they can't get to you... hopefully.
Like what? If they get more advanced than we do, they can presumably get out.


Hell, create a whole new dimension for them... to... play... in...
I don't think we have that capability just yet and, if I had to guess, I'd say nanomachines are much sooner to come along than dimension creation. And, the same holds true here as for the bubble... if they get more advanced than we are, they can get out of that dimension.


Are you there God? It's me, MacGuffin.[/QUOTE]

[Karl]

I think Robotic Evolution would be very limited, although whether this would or would not be evolution is very hazy. I have this book that I think briefly mentions this idea, let me look at it...

[MacGuffin]

I don't think we have that capability just yet and, if I had to guess, I'd say nanomachines are much sooner to come along than dimension creation. And, the same holds true here as for the bubble... if they get more advanced than we are, they can get out of that dimension.

To be honest, considering how hard space travel is, it is more likely they loop back and eradicate us.

You could set parameters, but what is that saying from Jurassic Park? "Nature finds a way". And I certainly don't think humans have the foresight to plan for every possible contingency.

Like I speculated in my first post, maybe we are the result of silicon creatures deciding to try carbon-based nanomachines.

[jittus rye]

Man, this sub forum is annoying. I had to alter my profile just to post here.

Life is just a collection of tools that interact with an environment. Unfortunately I read "The Emotion Machine" recently. It is available for free online, you should check it out.

I am a bit busy atm so I guess I will talk about something else on the subject later.

When a process malfunctions in such a way that it accomplishes something different, and it manages to replicate, it isn't much of anything special. The nature of 'infinite' randomn variance is such that it only allows for a relatively small amount of possibilities to take place due to things like natural selection, and vestigial products. Poo on neural networks, goooooooo ok whatever.

[Karl]

Hm, yeah, what this is saying is that there's a limited number of things that could go wrong, and it's hard to imagine a program as complex as an organism. The most we could do would be to try to program ways for it to respond to changing environments, but even then it would be very boring evolution.

The malfunction idea seems very limited, from my perspective. It's possible to create a Robot that does not malfunction in certain ways. Making a sentinel probe that would self destruct before reverting to a nano-von Neumann probe is very believable.

Edit: I just realize I apparently capitalize "Robot" but not "god".

[Karl]

Sorry for the double post, but I guess what I'm getting is that this Robotic Evolution requires a program that can change itself and move from less complex to more complex. I don't quite see how we can do this. If we can, some sort of I Robot reminiscent stuff is likely, with machines taking the lead over humanity, although that wouldn't really bother me.

[Anonymous]

If we take this a step further, if we consider this to be the creation of a silicon-based system of living creatures (as opposed to our own, carbon-based system), would we expect intelligence to be an eventual evolutionary product of this machine kingdom? Is this the true path to AI; is machine evolution the real answer, where malfunctions or bugs in the machine's code (it's DNA, if you will) lead to an evolutionary progression toward intelligence, or is the method of tinkering with neural nets still the best option?

Possibly a mixture. Simply going by the first option would most likely take to long. As it is, I don't see humanity as a race with very much promise in terms of longevity without some type of aid.


If we hoped to spawn an entire kingdom of machines, what are some of the parameters we should impose on the initial machine race?

Ideally, it would be something to prevent them from harming humanity. At the same time, though, more parameters would probably mean less intelligence. Plus, keeping the base format simple enough to not cause complications might not provide much room for permanent parameters in the first place.

[garak]

I'd say that any creation that is meant to go out into the "wild", reproduce, and spread, has to be just as fit as any lifeform that would want to take on that task. Are we really capable of creating something that advanced? Look at how hard it is for us to just get ourselves off of this rock, let alone to other planets, or landing on them, or any of that stuff. And to create a robot that can do all of that on its own? I wonder if we'll ever be that advanced.

[Hustler]

Are we really capable of creating something that advanced? Look at how hard it is for us to just get ourselves off of this rock, let alone to other planets, or landing on them, or any of that stuff. And to create a robot that can do all of that on its own? I wonder if we'll ever be that advanced.

I think nanotechnology is right around the corner. Especially compared to space travel, which is not what I'm suggesting our creations do. All I'm proposing is the creation of nanomachines that can replicate themselves. Presumably, this would require the machines to be very simple, so that enabling them to reproduce wouldn't be an intractable problem for us.

[Oso Mocoso]

I think nanotechnology is right around the corner. Especially compared to space travel, which is not what I'm suggesting our creations do.

Nanotech products are already on store shelves. I own several products whose manufacture was possible due to nanotech material processes. I have a shirt that doesn't need to be ironed and remains wrinkle free because of a nanotech process applied to the cotton. I also own a suit of foul weather gear that doesn't absorb body odor, which is nice if you're on a long sailing voyage.


All I'm proposing is the creation of nanomachines that can replicate themselves. Presumably, this would require the machines to be very simple, so that enabling them to reproduce wouldn't be an intractable problem for us.

Self-replicating nanomachines are another thing entirely. I think that'll be a while.

--Oso

[Karl]

I think nanotechnology is right around the corner. Especially compared to space travel, which is not what I'm suggesting our creations do. All I'm proposing is the creation of nanomachines that can replicate themselves. Presumably, this would require the machines to be very simple, so that enabling them to reproduce wouldn't be an intractable problem for us.

We can do that. We can even program in some things that change from generation to generation. However, malfunctions leading to more effective and more complex machines doesn't seem likely.

[Hustler]

We can do that. We can even program in some things that change from generation to generation. However, malfunctions leading to more effective and more complex machines doesn't seem likely.
It doesn't have to be likely if you have billions or trillions of machines.

[Dunearhp]

Read this book recently. Sci-Fi, it touches on a few of the ideas you are talking about.

http://www.amazon.com/Recursion-Tony-Ballantyne/dp/0330426990

[Dunearhp]

One of the biggest limitations of von Neumann machines is that, like bacteria, they are limited by the resources of the environment in which they find themselves. I do not consider the Grey Goo (http://en.wikipedia.org/wiki/Grey_goo) scenario plausible, or even possible. It may be possible to make nano-machines that can operate in several different ecological chemistries, using the basic carbon and water chemistry somewhere like earth (acting like little more than custom bacteria), or other types of chemistries in other environments (where water is absent, never liquid etc.). A universal replicator is another matter entirely.

Most likely we would find these machines competing with their biological counterparts in the given environment.

[Ferrus]

Silicon based life has been considered impossible for many years.

[kuranes]

This is partly what this non-fiction book ( that I have only gisted ) is about -
http://www.wolframscience.com/nksonline/toc.html
http://www.math.usf.edu/~eclark/ANKOS_reviews.html
http://www.kurzweilai.net/articles/art0464.html?printable=1

Early game based on cellular automata
http://en.wikipedia.org/wiki/Conway%27s_Game_of_Life

Interesting novel touching on this -

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

[Hustler]

Silicon based life has been considered impossible for many years.

Dietary fiber was considered to be of no importance to human digestion or well-being for many years.

[HilbertSpace]

We do already have artificial life in silico - software simulations that, while much simpler than any living organism, do run around inside simulated ecosystems evolving and interacting with the environment and each other. John Holland did some real formative work there, and had several books on his work, both the general genetic algorithm and artificial life applications.

The question of bridging the gap from a simulation to a real artificial organism is a bit harder. Venter, the human genome sequencing guy, is finishing up the creation of what might be called the first true artificial organism (http://www.guardian.co.uk/science/2007/oct/06/genetics.climatechange). It's a reverse engineering process in which the scientists involved tried to pare down a genome to the bare minimum.

I suspect that this sort of thing is what artificial life will ultimately look like - at least near term. The technology is almost there, and the applications (not to mention the pure research) will be potentially huge. The atomic-scale robots that get Bill Joy feeling like a Luddite are much further off, I think - but I suspect that even there we might tend to adopt paradigms already in the natural world, and bacterial scale is certainly easier to handle for us within the next 10 - 20 years. The chemical computation approach works well, and it's enough to keep us busy for a while.

Remember that viruses are insufficient complex to sustain themselves as independent living organisms. There's not enough moving parts. They do evolve, and they are sort of living in that they reproduce, but it's a different order of magnitude of complexity.

[Deckard]

If we take this a step further, if we consider this to be the creation of a silicon-based system of living creatures (as opposed to our own, carbon-based system), would we expect intelligence to be an eventual evolutionary product of this machine kingdom? Is this the true path to AI; is machine evolution the real answer, where malfunctions or bugs in the machine's code (it's DNA, if you will) lead to an evolutionary progression toward intelligence, or is the method of tinkering with neural nets still the best option? If we hoped to spawn an entire kingdom of machines, what are some of the parameters we should impose on the initial machine race? How are they powered?
Assuming we could create a silicon-based self-replicating analogue for carbon-based single cell organisms, I don't think we can assume intelligence will arise. Maybe if we assume ideal conditions, we could predict a fairly high probability of it. As far as considering this realistically...do we really want to wait around for a few billion years to see if our experiment produces intelligent life?

[slacker]

Any theoretical VN probe worth its salt would destroy its home planet in seconds. If not, they will fail at their mission.

The requirements for a VN probe are a fatal design flaw. Actually, I might even posit that all AI with "artificial" goals are, relatively, evolutionary unfit. VN probe evolution will quickly select against space exploration.

And I second Dunearhp's comment about limited environmental resources.

[joft]

would we expect intelligence to be an eventual evolutionary product of this machine kingdom?
Unless you accept intelligent emergent behavior of massive numbers of nanobots, they're going to have to evolve out of the nano-scale to a scale large enough for a single one of them to be intelligent.

Then the question becomes: would there ever be a great enough selective advantage for these things to become larger? I suppose a theoretical biologist might have some suggestions for how we carbon based organisms did it.

But would synthetic life follow roughly the same evolutionary path? Is it possible that the materials involved would push the energy threshold for evolving toward the macro scale impossibly high?

I think the constraints of the real world would slow this program down a lot, and simulated software replicators are likely to become "intelligent" first. If we start from scratch at the nanoscale and wait for them to evolve into large organisms with central nervous systems, it might take thousands or even billions of years. But with software, we're already close to things like visual recognition, natural language processing, etc.

[Karl]

It doesn't have to be likely if you have billions or trillions of machines.

What you're saying is that with trillions of machines, it would, in fact, be likely. You are contradicting me when I say it's unlikely. I am saying it is unlikely, with ten machines, or trillions.

I maintain that "malfunctions leading to more effective and more complex machines doesn't seem likely." The changes that could occur are too limited, and are likely to lead to simplicity and not complexity, just like your sentinal to von neumann, where its capabilities decreased.

[Hustler]

As far as considering this realistically...do we really want to wait around for a few billion years to see if our experiment produces intelligent life?
That's a whole different question for nanotechnology. Can it, in combination with medical science, make me live for a billion years? That'd be nice.


Any theoretical VN probe worth its salt would destroy its home planet in seconds. If not, they will fail at their mission.
That doesn't really make any sense and, moreover, it seems like you didn't read the second half of the post.


I think the constraints of the real world would slow this program down a lot, and simulated software replicators are likely to become "intelligent" first. If we start from scratch at the nanoscale and wait for them to evolve into large organisms with central nervous systems, it might take thousands or even billions of years. But with software, we're already close to things like visual recognition, natural language processing, etc.
Perhaps the influence of the real world, in the form of a body to go along with the mind, is necessary for the emergence of intelligence. Giving a nanomachine no imperative except to replicate seems quite similar to what biological organisms on the nano-scale are all about. Programming large computers to do specific tasks, even if we pile on the tasks and make the programs more and more robust, doesn't seem to me like a path toward legitimate intelligence. Computers today are no more intelligent than they were 30 years ago, but they can certainly perform far more specific tasks.


What you're saying is that with trillions of machines, it would, in fact, be likely. You are contradicting me when I say it's unlikely. I am saying it is unlikely, with ten machines, or trillions.
Then your statement is unfounded. Real world machines malfunction all the time; it doesn't take a population of trillions of them to start seeing things go wrong.


I maintain that "malfunctions leading to more effective and more complex machines doesn't seem likely." The changes that could occur are too limited, and are likely to lead to simplicity and not complexity, just like your sentinal to von neumann, where its capabilities decreased.
That depends on how you view it. Did its ability to self-replicate decrease? No, it increased. It's conformance to the initial specifications decreased and, as such, it's performance from a human-helper standpoint, but it's objective capabilities did not. I could say they increased and be just as correct as you when you say they decreased. And that's just one example. Consider a sentinel which goes haywire and instead of camping out and watching a civilization from afar when it encounters one, flies right down into the midst of said civilization. The possibilities for minor malfunctions are limitless, some more complex, some simpler, some of equal complexity.

[HilbertSpace]

I do think the evolution of multicellularity would be relatively likely with anything complex enough to be called "alive," whether it is natural or technological in origin. When the agents (complex systems in and of themselves) communicate with each other, the resulting system can itself become complex, and start exhibiting emergent properties. We can see this sort of thing in bacterial colonies today - there's a researcher in Israel who has done some very interesting work modeling information flow dynamics in colonial formations.

We can also look at the simplest instances of multicellularity - the slime mold being one big favorite. Slime molds live as single celled creatures for a large part of their lives, but when they go into starvation mode they gather themselves into a multicellular structure complete with specialization among individual cells based around cell signaling dynamics that might be thought of as analogous to (but less complex than) the processes responsible for creating a multicellular organism from a single cell during ontogeny.

The question of inevitability, however, is always tricky. Single celled existence is extremely viable. In a multicellular organism, the vast majority of cells have to give up their own reproduction and instead become essentially slaves to the larger organism. It certainly can happen, and indeed has happened at several levels of scale (think about eusocial insects, for instance - the vast majority of ants in a colony do not reproduce). Overall, given a sufficiently complex environment, I think that we might expect to see something analogous to multicellularity evolve, but I don't think I would say it is inevitable.

[Hustler]

We can also look at the simplest instances of multicellularity - the slime mold being one big favorite. Slime molds live as single celled creatures for a large part of their lives, but when they go into starvation mode they gather themselves into a multicellular structure complete with specialization among individual cells based around cell signaling dynamics that might be thought of as analogous to (but less complex than) the processes responsible for creating a multicellular organism from a single cell during ontogeny.

Is it closer to being a multicellular organism or a community of single-celled organisms that each take on specific functions to assist in the survival of the community? Where is the line between a multicellular organism and a community of independent organisms, like ants? In the case of the slime, do they all have the same DNA?

[HilbertSpace]

Perhaps the influence of the real world, in the form of a body to go along with the mind, is necessary for the emergence of intelligence. Giving a nanomachine no imperative except to replicate seems quite similar to what biological organisms on the nano-scale are all about.


I very much agree with this. I think we first need to define intelligence. The one I like to use is that intelligence is the ability to detect and integrate patterns in the environment.This covers everything from the functioning of giant neural nets like the human brain, to simple viable organisms, to the evolutionary process itself as a genetic algorithm. Intelligence of this sort is inevitable - it's part of the definition of being alive. Artificial life forms, whether they look like cells or like tiny automotive robots, will need to sense their environment and adapt to changes, both in real time and in evolutionary time.



Computers today are no more intelligent than they were 30 years ago, but they can certainly perform far more specific tasks.


I think we have made some progress here. One of the possible obstacles has been the idea that intelligence necessarily means human-like intelligence rather than that exhibited by all living organisms. But degree intelligence is always a matter of informational complexity.

Evolutionary computation, artificial life, simulated artificial neural networks, and the like do exhibit subtle behaviors, and have advanced over the past 30 years. Whether or not we need to embed something in the real world in order to get "real intelligence" out of it is an open question. The answer you get will often vary with the department of the person you ask - robotics guys say probably yes, software guys say probably no.

In my opinion, it's really about informational complexity. The real world has high complexity. Using it gives us ready access to a system that requires sophisticated problem solving capabilities. However, the world of pure information within a computer is also becoming more complex - it's still far simpler than the real world, but we are beginning to encounter situations in which the information required to solve problems within a computer (e.g., large scale semantic databases, or databases constructed from real world observations) do require the complex, emergent problem solving capabilities we associate with intelligence.

[HilbertSpace]

Is it closer to being a multicellular organism or a community of single-celled organisms that each take on specific functions to assist in the survival of the community? Where is the line between a multicellular organism and a community of independent organisms, like ants? In the case of the slime, do they all have the same DNA?

A colony has the same DNA, because they are all clones of the starting cell. The question of drawing a firm line between a community of independent organisms and a single multicellular organisms, though is pretty tricky. Rather than seeing a line, I think we have to see it as a gradient.

The gradient measures what I call coherence - it is a measure of the tightness of information integration between two systems. It covers both the built in information (the "hardware" information infrastructure, like the genome, that sort of defines the accessible state space for the agent), and the communication, or real time information - the degree to which states can be communicated and coordinated in a shared environment.

If cells within a multicellular organism are highly coherent, and cells of independent (e.g., different species in different environments) of bacteria are non-coherent, eusocial ants would occupy a middle ground. People have started calling ant colonies "super-organisms" because the colony itself has to be thought of as an organism, with the individual ants acting like cells in a body rather than as organisms.

If we think about it in terms of informational complexity contained within and processed by an organism, we might say that it increases with regard to the entity as a whole when coherence increases, but can (and perhaps must) decrease with regard to any individual subunit. It is sort of analogous to what happens in human societies (and I'd argue it happens to people for the same reason it happens to ants) where specializations develop in response to complexities, and which in turn increases the complexity of the overall system.

[Deckard]

The question of inevitability, however, is always tricky. Single celled existence is extremely viable. In a multicellular organism, the vast majority of cells have to give up their own reproduction and instead become essentially slaves to the larger organism. It certainly can happen, and indeed has happened at several levels of scale (think about eusocial insects, for instance - the vast majority of ants in a colony do not reproduce). Overall, given a sufficiently complex environment, I think that we might expect to see something analogous to multicellularity evolve, but I don't think I would say it is inevitable.
Given a sufficiently complex environment and enough time, wouldn't the probability of multi-cellular organisms evolving begin to approach 1? And I suppose one could extend that argument to the probability of 'intelligent' life emerging.

So, while it might be likely or even inevitable that these features emerge in an artificial life experiment, the time frames involved are prohibitive. The practical question then is: How do we hurry the evolutionary process along? Maybe biological engineering can help us here, so we end up with a hybrid of design and evolution.

[Hustler]

A colony has the same DNA, because they are all clones of the starting cell.

What if you take cells from two or more different slime mold colonies, put them together, and then subject them to the nutrient deprivation conditions you described before that get them to start acting in concert. Will all the cells join forces, or will the cells only work with others with identical DNA?

[HilbertSpace]

What if you take cells from two or more different slime mold colonies, put them together, and then subject them to the nutrient deprivation conditions you described before that get them to start acting in concert. Will all the cells join forces, or will the cells only work with others with identical DNA?

As long as they're from different colonies of the same species, I think that the behavior would be unchanged. The signals they use for communication should all be identical, and I don't believe they would be able to tell the difference.

[HilbertSpace]

Given a sufficiently complex environment and enough time, wouldn't the probability of multi-cellular organisms evolving begin to approach 1? And I suppose one could extend that argument to the probability of 'intelligent' life emerging.


That might be the case for multicellularity. I'm less confident about it being true for human-style intelligence. It appeared pretty late in the game, and isn't exactly ubiquitous.

If you were to rewind, say, 10 million years, you might still get something like what we have now. If you were to go back a billion or two, though, things might look very different. I don't think you'd be able to count on seeing anything you'd recognize, including human-like intelligence.



So, while it might be likely or even inevitable that these features emerge in an artificial life experiment, the time frames involved are prohibitive. The practical question then is: How do we hurry the evolutionary process along? Maybe biological engineering can help us here, so we end up with a hybrid of design and evolution.

That's a good point. We can use artificial selection as well as genetic engineering to intervene, and use those tools to create dedicated specialists for specific tasks, while still retaining the robustness we see in biological systems. From a single cell kind of system, getting to a HAL-like intelligence would still be a stretch, I think, but I could certainly see bio-engineered creations doing information processing, exploration, or carrying out complex engineering tasks.

[Ferrus]

That might be the case for multicellularity. I'm less confident about it being true for human-style intelligence. It appeared pretty late in the game, and isn't exactly ubiquitous.
Although from an exobiological perspective, would it not be interesting to simultaneously create millions, perhaps billions of artificial, expedited evolutionary chains, and determine from this rough probabilities regarding the formation of such entities as multicelluarity, sentience, and perhaps any other number of processes known and and unknown? Which could, surely, to some extent be used as a key to determining the likelihood of the selfsame processes emerging in the universe at large (ignoring abiogensis), or at least with lifeforms such as may form in an enviroment equivalent to that generated artificially?

Out of interest, how would you conceive of self-consciousness, and how does celluar communication, or indeed perhaps communication within a digital neural network result in an entity that comes to regards itself as being an entity rather being simply a collection of particles interacting with other particles in a specific manner?

[Deckard]

That might be the case for multicellularity. I'm less confident about it being true for human-style intelligence. It appeared pretty late in the game, and isn't exactly ubiquitous.

If you were to rewind, say, 10 million years, you might still get something like what we have now. If you were to go back a billion or two, though, things might look very different. I don't think you'd be able to count on seeing anything you'd recognize, including human-like intelligence.
Yeah...I see your point, but it seems to me that certain traits or abilities are almost evolutionary inevitabilities, simply because they offer such a significant selective advantage. Things like communication ability, ability to learn, ability to plan. Or at a lower level, manoeuvrability, flexible diet, robustness to environmental change, etc. This doesn't really take into account evolutionary dead-ends, which I'm presuming might occur (i.e. where a species might not be able to progress further in intelligence for whatever reason). However one would think that the diversity of the phylogenetic tree would compensate for the possibility of such dead-ends.

I'm not suggesting it's inevitable exactly. But just going by the principle that greater intelligence is a selective advantage once the organism is able to support such a mutation, it would appear that evolutionary momentum would usually be in the direction of greater intelligence.


That's a good point. We can use artificial selection as well as genetic engineering to intervene, and use those tools to create dedicated specialists for specific tasks, while still retaining the robustness we see in biological systems. From a single cell kind of system, getting to a HAL-like intelligence would still be a stretch, I think, but I could certainly see bio-engineered creations doing information processing, exploration, or carrying out complex engineering tasks.
Yeah, I'd agree with this. Bio-engineering is very cool and has a ton of applications, but I don't see it going the full cycle from single cell to human-like intelligence. Conversely, I can see the designed/engineered computer-simulated approach eventually being able to produce human-like intelligence, however this approach is less capable of simulating the extremely complex workings & interactions of living cells. The brain is kind of a computer which happens to be made of alive stuff, so it's far better suited to modelling in abstract simulation. With the brain, what matters is the function of an element (say, a neuron), rather than its physical implementation. So we can create an abstract functional model without having to simulate all the biology, and that's far easier computationally. Or at even greater levels of abstraction, we can design simulated modules that perform the same or similar function to the analogous parts of the brain, such as the memory system, but with a completely different design.

What i'm trying to say is, the two approaches (real world evolved vs. simulated designed) both have limitations. Though i think that computer simulations will have increased relevance to bio-engineering, as our hardware gets better at simulating complex cells & their interactions.

[HilbertSpace]

Although from an exobiological perspective, would it not be interesting to simultaneously create millions, perhaps billions of artificial, expedited evolutionary chains, and determine from this rough probabilities regarding the formation of such entities as multicelluarity, sentience, and perhaps any other number of processes known and and unknown? Which could, surely, to some extent be used as a key to determining the likelihood of the selfsame processes emerging in the universe at large (ignoring abiogensis), or at least with lifeforms such as may form in an enviroment equivalent to that generated artificially?


A big part of the reason for doing the simulations is to try to really work out how biological evolution actually works. It's not so much a matter of trying to work out probabilities - we can already talk about the probability for a mutation occurring (which isn't too difficult to measure, or at least estimate). It's being able to determine the way in which mutations at the gene level really affect the course of evolution. Moving from genotype to phenotype is a difficult issue because they're at different levels of realization - different levels of the hierarchy that make up a living organism.

If we could find life independently evolved on another planet, it would help to answer some of those questions more easily, but there really is still a big chunk of theory to be developed.



Out of interest, how would you conceive of self-consciousness, and how does celluar communication, or indeed perhaps communication within a digital neural network result in an entity that comes to regards itself as being an entity rather being simply a collection of particles interacting with other particles in a specific manner?

I really don't know. On one level, I think that the idea of self as independent entity is a little artificial. The self is fuzzy around the edges, especially from an information perspective. It's obviously not entirely artificial as a concept - a simple thing like tissue rejection argues that - but in my opinion it's really only part of the picture, and one which we sometimes tend to concentrate on too much in that it can obscure some important perspectives on encapsulating systems.

Like with intelligence, I'm most comfortable with an idea of selfness that transcends the anthropocentric concept. Even simple organisms can display the more pragmatic aspects of selfhood, like self-preservation. More complex animals can recognize their own territorial markings versus those of others, and can also usually recognize their own conspecifics.

[HilbertSpace]

Yeah...I see your point, but it seems to me that certain traits or abilities are almost evolutionary inevitabilities, simply because they offer such a significant selective advantage. Things like communication ability, ability to learn, ability to plan. Or at a lower level, manoeuvrability, flexible diet, robustness to environmental change, etc. This doesn't really take into account evolutionary dead-ends, which I'm presuming might occur (i.e. where a species might not be able to progress further in intelligence for whatever reason). However one would think that the diversity of the phylogenetic tree would compensate for the possibility of such dead-ends.



I think these things are not only inevitable; they are the sine qua non of life. But they can and must occur first at the chemical level, and later in real time. Even bacteria learn in evolutionary time, and they can even be said to have some form of planning. Getting back to the slime molds, they 'plan' to migrate (via spore formation) to get around local starvation conditions.

I always see the anthropocentric models of intelligence as problematic, in a way, because we tend to concentrate on them and lose the idea that what we do in concept space is just a very rapid (and admittedly highly complex) specific realization of what everything does all the time in biochemical and gene space.

My point is that I don't think we can say that organisms evolve toward higher intelligence - it's much less directional than that. There's a lot of examples of loss of complexity and loss of function (Gould was big on this sort of thing, because he felt that you can't even say organisms evolve toward higher complexity, at least not monotonically). I don't entirely agree with him there - I think that the complexity of the organism is partly a function of the complexity of the environment, and that environmental complexity increases with proliferation of organisms.



The brain is kind of a computer which happens to be made of alive stuff, so it's far better suited to modelling in abstract simulation. With the brain, what matters is the function of an element (say, a neuron), rather than its physical implementation. So we can create an abstract functional model without having to simulate all the biology, and that's far easier computationally. Or at even greater levels of abstraction, we can design simulated modules that perform the same or similar function to the analogous parts of the brain, such as the memory system, but with a completely different design.


I agree with all of this. I think that our most promising approach even to the science of cellular physiology is to realize that the cell itself is basically a highly complex state machine, with micro and macrostates. Biochemical and neural networks are very similar in some ways in that they're both complex information processing entities that depend on emergent properties to perform computation.

It's that dependence on emergence, though, that makes it a little dangerous to perform simplifications from the top down while attempting to replicate an organic system (brain or cell). I think the society of mind sort of approach to AI is too well grounded in psychology and not enough in biology. The neural network attempts, in my opinion, hold a lot more promise for creating intelligent behavior. It's just that my own tendency is toward evolutionary themes, because they seem to be closer to the raw metal of the universal process of moving through an informationally rich environment in an adaptive manner.

[NoahFence]

It was his argument that, given a moderate rate of machine exploration and reproduction, von Neumann probes should be common throughout the galaxy and we should have encountered one or more by now.

The second statement does not follow. If von Neumann probes are common throughout the galaxy, it's possible we can't detect them. Stealth probes? Or just using something we can't pick out somehow. Hell, that could even explain dark matter...they've already eaten 90% of the universe!

[Hustler]

There has been some talk about multicellular emergence in this thread, and I think that's an interesting point when discussing nanomachines. One could argue that the transition from single-celled to multicellular life was the biggest step that evolution ever took (one could argue that about other things too, of course). It also took a very long time before we saw the first multicellular organisms, after countless billions of generations of single-celled life. Likewise, it took a long time before eukaryote cells emerged from prokaryote cells. These major, yet microscopic changes, were among the biggest steps that paved the way for human life as we know it today to emerge from the simplest primordial lifeforms.

What are some of the properties of these steps in evolution that make them so time-consuming, and what design features could be implemented in a nanomachine to see it move past these evolutionary hurdles more quickly? Deckard mentioned before that achieving emergent AI via machine evolution would perhaps take billions of years. What, then, could we do to speed it along? Furthermore, is there any evidence or reasoning to suggest that machine evolution would be faster or slower than biological evolution?

[HilbertSpace]

There has been some talk about multicellular emergence in this thread, and I think that's an interesting point when discussing nanomachines. One could argue that the transition from single-celled to multicellular life was the biggest step that evolution ever took (one could argue that about other things too, of course). It also took a very long time before we saw the first multicellular organisms, after countless billions of generations of single-celled life. Likewise, it took a long time before eukaryote cells emerged from prokaryote cells. These major, yet microscopic changes, were among the biggest steps that paved the way for human life as we know it today to emerge from the simplest primordial lifeforms.


These are all very true. Perhaps the largest innovation we (that is, Earth's organisms) discovered was memory - the division of an organism into a blueprint (which is now locked in as DNA) and the organism. The discovery of a very simply way of encoding instructions to build and run an organism, including everything from finding food to reproduction, was the first, and probably the most important, major innovation. Another big one was the origin of sexual reproduction.



What are some of the properties of these steps in evolution that make them so time-consuming, and what design features could be implemented in a nanomachine to see it move past these evolutionary hurdles more quickly? Deckard mentioned before that achieving emergent AI via machine evolution would perhaps take billions of years. What, then, could we do to speed it along? Furthermore, is there any evidence or reasoning to suggest that machine evolution would be faster or slower than biological evolution?

One of the more interesting ways I've heard of discussing the difference between genotype and phenotype is this - picture that you are standing on a map. Make it some random location somewhere in the north of England. At every time slice, you take a step in a random direction. If you are in England after taking that step, you remain in Phenotype A (English). The microstate of your DNA has changed - you're in a different spot - but the macrostate of your phenotype remained the same. If you're in a border region, crossing over to Phenotype B (Scottish) might be very probable. If you're far from the border, though, you might never wind up in Scotland.

Artificial selection (which is just another form of natural selection, if you take a step further out and remember that humans are also part of the biosphere) tends to work faster than natural selection, which is why we've been able to create so many varieties of domesticated plants and animals over the course of thousands of years. I can think of at least two reasons for that - culling in artificial selection can be much more brutal than in natural selection, and specific characteristics can be chosen and concentrated upon without the distraction of other, random events. Furthermore, we create a dedicated environment that meets the needs of the newly created organisms that might otherwise be at a disadvantage in a potentially hostile environment (imagine a pack of Lhasa Apsos roaming the Serengeti).

If we can better master the genotype -> phenotype process as a process for realizing an emergent phenomenon (whether it is DNA or some other structure, organisms of sufficient complexity to be called life would need something along those lines), we might be able custom design lifeforms that were both well adapted and easily adaptable to new situations. We might specifically couple mutation rate to overall breeding pool survival (if everyone around you is dying off, pump up mutation rate), and we might bias mutations to changing some parts of the DNA more than others. We might have reproduction occur with zero, one, five, or some other number of partners, depending on what we're looking to do in terms of evolutionary dynamics.

We could also bite the bullet and short circuit the evolutionary process by designing onto the first generation many of the properties we want - we lose some part of having a completely evolutionary design, but we do start from a point further along the lines we need. This is sort of Venter's approach - starting with a viable, naturally occurring bacteria and deleting genes to make it into a simpler, minimalistic organism which could then in theory become a starting point for making a new, dedicated organism suited for another purpose.