Artificial Neural Networks Demonstrate the Evolution of Human Intelligence

samazon writes "Ph.D. students at Trinity College in Dublin have constructed an artificial neural network model to demonstrate the Machiavellian intelligence theory — that human intelligence evolved based on the need for social teamwork and indexing a variety of social relationships and statuses. (Abstract) The experiment involved programming a base group of 50 simulated 'brains' which were required to participate one of two classical game theory dilemmas — the Prisoner's Dilemma or the Snowdrift game. Upon completion of either game, each 'brain' produced 'offspring' asexually, with 'brains' that made more advantageous choices during the games programmed to have a better chance to reproduce. A potential random mutation during each generation changed the 'brain's structure, number of neurons, or the strengths of the connections between those neurons,' simulating the evolution of the social brain. After 50,000 generations, the model showed that as cooperation increased, so did the intelligence of the programmed brains." The full paper is available.

Comments

Now...

[Like2Byte]

Now *THAT's* intelligent design!

$Ducks

Re:Now...

[PatDev]

I know you made a joke, but this right here is why believing in intelligent design and evolution etc are not necessarily incompatible with each other.

No, it is not. They are incompatible.

Intelligent Design comes in two forms. The first is when we admit that it is just a euphemism for creationism. In this case, the theory of evolution (as well as most of the field of archaeology) clearly contradicts the story of Genesis, thus rendering the two incompatible.

The second is the form in which ID, in an attempt to distance itself from religion, rests upon the principle of irreducible complexity. The basic idea is that certain constructs represented in nature today (the human eye is an oft-used example) would have been useless in a less-complex or less specific form, and thus these traits would not have evolved (a half-formed eye is an evolutionary disadvantage, a being is better off not wasting the calories keeping that useless tissue alive). Since these traits could not develop through incremental changes, some traits must not evolve, but must have been put there by some intelligent agent.

This second form is not so much a scientific theory as it is a fundamental misunderstanding of stochastic processes and the field of mathematical optimization. This form of ID is basically the claim that evolutionary optimization can never escape local optima to discover global optima - something a competent applied mathematician knows to be false.

Re:Now...

[Baloroth]

There is a third form you missed entirely, to which (I think) parent is referring. A situation where an intelligence creates the initial conditions necessary for life (in the case of the universe, the laws and parameters that govern it, or in a more local scale, the materials and conditions on the Earth that would bring about life in the end) which results in a "designed" life evolving on its own, as a consequence of those initial conditions, much like how this experiment outlines certain specific parameters that it hopes will bring about more advanced "brains".

Well, I think that is a possibility under ID anyways, I'm certainly not an expert on it.

Re:Now...

[PatDev]

That is a distinct viewpoint known as Deism - also commonly discussed as a "watchmaker God". It is a means of reconciling belief in a deity with the apparent lack of evidence for one. However, Deism directly contradicts intelligent design - the two are as irreconcilable as evolution and intelligent design.

Intelligent Design is the proposition that "certain features of the universe and of living things are best explained by an intelligent cause, not an undirected process such as natural selection" http://en.wikipedia.org/wiki/Intelligent_design . The very undirected process a hypothetical Deist god would set in motion (evolution) is specifically what Intelligent Design claims does not work.

It's not that evolution and religion cannot coexist - if I'm not mistaken, evolution even has the Papal seal of approval. They can. But intelligent design is not religion - it's a dogma pretending to be science. Only the form of pseudo-science they chose to make their defining point is so clearly refutable that they wind up with less credibility than if they had just gone with "faith" as their explanation.

Re:Now...

[tnk1]

This IS a possibility under what I would call "Weak Intelligent Design".

However, it is important to note that, in that context, Weak ID is not a direct competitor to evolution, it is a cause of it. Therefore, evolution becomes the only *scientific* mechanism for actual development of species (although a hypothetical deity could still decide to intervene directly). A Weak ID person may well then oppose this law because there is no scientific evidence for any other theory that describes how species emerge. To their mind, any "Strong ID" belongs more properly in a philosophy classroom, not in a science classroom, because it describes something that is possible, but impossible to use science to prove.

It does bear mentioning that it is possible that a deity could have created the Earth and Universe in 6006BC, and then simply created those entities in such as way as to make them look like they were 15 billion years old. There is nothing that prevents a creator from creating current conditions "in place" and then laughing as we scamper around trying to figure out what happened. And that is why I usually don't get spun up about this stuff. Thinking that way is entirely useless as a way of making scientific discoveries, but there is nothing that says science has to have the answers to *everything*.

Re:Now...

[holmstar]

The universe is absurdly huge. Even something with a tiny statistical likelihood is almost guaranteed to happen at some point, somewhere, and probably in many variations.

Re:Now...

[jc42]

The ID propenents are on very solid ground in their belief that something as complex as an eye, a flagellum or the blood clotting cascade could not evolve given that the partially formed proto-systems are useless.

Actually, their argument really reduces to "I don't see how the intermediate stages could be adaptive, so they weren't." But an interesting example appeared in the biological literature about a decade ago: A group of starfish called "brittle stars" (because of their hard surface made of silicate crystals) are in the very early stages of evolving a compound eye, and it's quite adaptive.

The critical part of this discovery is that their hard crystalline surface contains scattered lenses that focus incoming light on light-sensitive cells in the underlying skin surface. It is estimated that the resulting proto-eyes has an angular resolution of only about 1 degree, which isn't very good. It's about twice the angular diameter of the sun and moon, so these important light sources would be only around 1 pixel to them. But this does give them the ability to locate light sources and shadows, and to spot moving objects that are in good contrast to their background.

These critters' eyes are nowhere near as accurate as ours, or (to make a more relevant comparison) the eyes of crustacea or insects. But they are apparently beneficial to the stars, and are at a very early stage of development. In fact, there have been a number of comments on the high optical quality of their little lenses, comparable to the best lenses that humans can manufacture.

Interesting consequences

[Covalent]

The paper suggests that evolution favors cooperation but that it also favors low-cost solutions (i.e. lots of little dumb brains (ants) vs. singular powerful brains (humans)). Perhaps this explains the Fermi Paradox: Aliens are all over the place on other worlds, but they're mostly the former kind of cooperative rather than the latter.

Re:Interesting consequences

[CrimsonAvenger]

The paper suggests that evolution favors cooperation

The experiment in question picked "games" that require cooperation to achieve best results. So naturally the paper would suggest that evolution favors cooperation.

Linking reproduction to cooperation might be a reasonable theory. Or not. But this experiment doesn't suggest anything other than "if we make cooperation an asset in our experiment, then cooperation will work better in our experiment".

Begging the Question

[Anonymous Coward]

So if we stipulate in our environment that smarter brains are more likely to reproduce, then the smarter brains reproduce just like would happen if human brains evolved to be smarter as a competitive advantage, so human brains evolved as a competitive advantage? They've stacked the dice to make evolution happen in their artificial world, so why should we make the inference that the world's dice are stacked in just the same way?

Re:Begging the Question

[Score Whore]

Yes, their paper is a tautology. Shorter paper: "We created a simulation of our rules. Then the simulation proved our rules."

Re:Human intelligence ?

[Anonymous Coward]

Please reference previous comment. This clearly is designed to model intelligence in Tennessee. Now if we could just fast forward 50,000 generations...

Can't be true ...

[richieb]

I've seen "Idiocracy". This can't be true.

Re:Human intelligence ?

[SuricouRaven]

Keep going. C.Elegans has more neurons than that.

You can go look at them - someone has been nice enough to digitise the entire nervous system, down to every last synapse. It's browseable at http://wormweb.org/neuralnet#c=BAG&m=1

For the singulatity fans: Yes, this is almost the first full brain upload. It isn't quite, as it doesn't store synapse response data and the brain-map is actually a composite from multible individuals, but give it a couple more decades and one of the little worms may go down in history as the first naturally-occuring intelligence (If you can call it that) to make the transition to digital immortality.

Cave Johnson here...

[nman64]

Fact: The key to any successful cooperative test is trust, and as our data clearly shows, humans cannot be trusted. The solution: robots! Then, fire the guys who made those robots, and build better robots. Then, run those robots through a regimen of trust exercises, creating a foundation of mutual respect, reinforced by the simulated bonds of artificial friendship. Inspiring stuff. And finally, we put that trust to the test. Bam! Robots gave us six extra seconds of cooperation. Good job, robots. Cave Johnson. We're done here.

https://www.youtube.com/watch?v=AZMSAzZ76EU

Re:Human intelligence ?

[Anonymous Coward]

You mean 200 generations, from 6000BC.

Re:Socialism ve Capitalism?

[dryriver]

The free market is a "vehicle for freedom and self-determination" only for people who get the "top jobs" in an economy and "call the shots" from the top of the pyramid. That would be 1 - 5 percent of the population, depending on the country you live in. For almost everyone else, unfettered ("unregulated") free-market capitalism is a system of abuse and serfdom that lasts a lifetime. Do you like Software/Content DRM? It was dreamed up by supposed "free-market capitalism". As was almost everything in the modern world that kills/maims/abuses/denigrates/disadvantages the common man. The hard-left and hard-right will never enter a synthesis, because the hard-right is so morally corrupt, exploitative, predatory and irresponsible that nothing good can come from cooperating with the hard-right. Period.

Re:A bit of a deceptive title

[geekoid]

Every time a break through in AI is made, it gets taken out of AI.

I suspect that if I created an AI with a cat level intelligence, people would say the cat isn't intelligent.

Internet search was one considered an AI problem, the moment it was solved it was taken away from AI.
Many games have AI, but the moment they where created they where some how technical and not AI.

10 years ago what Siri does would have been considered AI.
Too many people put a mystical belief on top of intelligence. When you remove the mystical element people don't consider it intelligence.

Re:Human intelligence ?

[zlives]

please account for inbreeding...

The Problem with this Study

[CanHasDIY]

... is the same problem pretty much every study has -- it's based on the concept of ceteris paribus, which does not exist in reality.

From a purely academic standpoint, however, it is a neat experiment.

Re:Not sure what to call this

[kurzweilfreak]

Yes it is, if you understand what the hell you're talking about.

So *someone* creates the neurons

Where the neurons "come from" is irrelevant, whether they sprang forth from the magic of the intelligent designer or were created through abiogenesis. What is relevant is that they reproduce with a mechanism for heredity.

creates connections

Connections could be generated randomly and you will end up with the same result as long as you have heredity, mutation, and selection.

establishes rules of selection

Where the rules of selection come from is irrelevant, whether it's "natural" selection (environment) or "artificial" selection (we choose the rules, like for dog breeding). What matters is that there is selection choosing from a randomly generated pool of variety.

creates mutations for random change

What part of "random mutation" do you not get? Random mutations happen in DNA. This produces random variety that is then culled through selection. Same thing here.

and that is somehow a model for completely random unguided uncreated evolution???

Yes, because you have exactly the same mechanisms: heredity, random mutation, selection. Evolution isn't random; mutations are random.

It's amazing how ridiculous something seems when you think you understand it but you really don't, doesn't it?