Intelligent Design

[jtls1986]

Is anyone familiar with this concept...I was introduced to this recently and I find it to be quite convincing...

Considering that Darwin himself stated that his theory of evolution would completely break down ***IF*** a biological entity was capable of developing complex systems without taking slow steps of slowly evolving similar structures that would eventually lead to the complex systems...

After observing a bacterium, and focusing on a single structure, the flagellum...scientists revealed a very complex biological machine....involving structures similar to a human machine that would run wheels or something like that...

Anyway, the scientists declared that such a complex system could not have been capable of evolving from organisms that originated from a "proto-earth", since the proteins and enzymes must connect in a particular fashion...and cannot connect differently...if the enzymes connect incorrectly....the enzymes will fall apart...and the protein itself would not have been produced...

These enzymes have thousands...if not billions of information that tell the enzyme to connect to a specific enzyme....and after connecting...the enzymes will roll up in a certain fashion to finally produce the protein..

How could primitive cells that originated from amino acids suddenly form such a complex chain of information that would form enzymes...and finally proteins that would together....form a complex bacterial flagellum?

[QED]

You are all talking as if AI researchers were unaware of GAs. They simply do not produce AI.

In what way are you not simply arguing from incredulity here?

You need INTELLIGENCE to think of the GA in the first place and to design an engine round it. Nobody I have ever talked to feels that a GA is capable of producing AI on its own.

Do you have any other evidence or is it just a feeling? You see, after browsing the various pages of your families website for a an hour or so, I can't help thinking that you have started out with a particular hypothesis in mind and have then gone about looking for some data to fit it.

Irreducible complexity is often presented as refuting evolution, yet as you are familiar with genetic programming methods, you will know that no such claim is sustainable. But it seems to be important for you that AI is irreducibly complex, so I can't help wondering if your argument is an attempt to switch the focus away from a lost cause.

[Ian Parker]

Do you have any other evidence or is it just a feeling? You see, after browsing the various pages of your families website for a an hour or so, I can't help thinking that you have started out with a particular hypothesis in mind and have then gone about looking for some data to fit it.

No it is is not just a feeling. I think you should take a look at the literature. With the Web it is easy enough to get at. Just enter "Genetic Algoritms" in Google. The pattern that emerges is that GAs work well (if not optimally) in a differentiable manifold. They have rediscovered negative feedback and designed a QM flip flop. They cannot play Go or write C++. Analogue electronics in fact bears quite a striking resemblence to the fossil record.

[QED]

The pattern that emerges is that GAs work well (if not optimally) in a differentiable manifold. They have rediscovered negative feedback and designed a QM flip flop. They cannot play Go or write C++. Analogue electronics in fact bears quite a striking resemblence to the fossil record.

OK, I'll start searching for mathematical proofs that show GA's can't play go or write C++. (I hope I can find some -- then I won't feel so bad about struggling with C++ )

But I'm intruiged by your statement that analogue electronics bears a striking resemblance to the fossil record. I used to work on analogue computers in the 60's but I don't recall any similarities to natural history. What do you mean by this?

[Ian Parker]

Basically in terms of the dimensional complexity. What I am doing is examining what kind of problem can and connot be solved by GAs. Plus rate of convergence.

[Curious]

As you seem to be discussing AI I would like to ask why you believe that algorithms cannot produce AI, if that is your belief. The whole evolutionary process does not have to keep to the same algorithm throughout. During the course of an organisms evolution, the instructions(eg. dna) may change as may the expression of the instruction. We might as well ask for an algorithm to create a human mind from scratch as whatever AI we create could be proved to be artificial and only seemingly intelligent. AI, like evolution, is much easier to understand when you realise that it is unnecessary to restrict it to a single unchanging impetus. The human does not evolve into a human due to the ability of the human algorithm to create a human being. It is the environment that leads the organism to whatever path is most survivable. This is why the use of AI as an argument to explain the difficulties of evolution is inadequate as it starts from the premise that the algorithm is internal and unchanging(except for its own self modification) whereas evolution is driven by the external , everchanging environment.

[otseng]

We have a thread on genetic algorithms/genetic programming here.

We also have a thread on abiogenesis here. Not much was posted in that thread so it would be interesting to have that thread revived.

[Ian Parker]

I do not say that AI is impossible. I know there are some people that do but I am not one of them. What I am saying is that AI is not evolvable by the rules of Darwinism. If we do succeed in producing an AI program it will almost certainly be irreducible since most of the algorithms we are familiar with, like Householder, Gaussian Elimination and FFT are irreducible and do not occur in Nature. Most of Engineering is irreducible too. Insects like bees wasps and ants have a social structure which is completely hard wired genetically. They have no evolved wheels and carts. Perturbation Theory, unlike Householder, is reducible which is why it has a biological pedigree. As I have said before Householder is a far better algorithm than Perturbation Theory. Operating systems too are irreducible and will be a vital part of any AI system. It is thus impossible for AI to be reducible (at least in the first instance) although this is a slightly pedantic point

With Engineering and Mathematics happily producing so much that is irreducible, and failing to produce AI it is hard to reconcile Intelligence and reducibility. We know in fact that we need an evolving inference engine to produce AI, this in fact makes it pointless (from the professional AI view point) to attempt to evolve a boot. Investigations of GAs show that they will do many marvellous things but not develop AI.

Investigating Evolution (that is to say the evolution of everything) seems to leave us with the impressions.

1) That evolution follows a minimum information path. If we believe in a minimum information path, we will not automatically gravitate to God for explanations. We will proceed in a proper professional scientific way. If we do finally decide on Divine Intervention it will have been done after an exhaustive search. I think that statement should remove most of the misgivings of the Scientific Community.
2) Minimum information is not the same as zero information. Intelligence cannot have arisen with zero information.

Intelligence is where most of the information in evolution resides. Other things like speciation and the fossil record are readily explicable.

Quantum mechanics is a red herring although an interesting one. There is the question of how the QM algorithms themselves evolved. I think QM should be rejected for a number of reasons.

1) Some snakes have pits on their skin which enable them to see in total darkness. Their pits are in fact pyros. If their brains were in any way QM the pits would contain QM computers which would have a Hamiltonian with energy levels. The QM computer would act as a refrigerator giving a cold system. This would in fact be the consequence of running it. This would provide a quantum well system which is the basis of high sensitivity military thermal imagers. If could get a QM computer we might be able to eliminate the Sterling engines which are so much a part of thermal imaging.
2) Householder after all! A QM system has eigenvalues and eigenvectors. Certainly if our brains were QM they would not work in the way in which the constructors of Neural Networks have supposed in the past.

If as it would seem that minimal information input (consistent with gynothropism [biologists always talk about lines of daughters, when discussing Evolution or anything else]). This sort of set up would also fit another possible theory, that of infinite gynothropism.
There will be Nexp(-S) worlds with intelligent life. If N is infinite then the problem is solved and we will then have a minimum information system. Somehow I think most people would believe in Intelligent Design rather than infinite gynothropism.

[Curious]

I do not say that AI is impossible. I know there are some people that do but I am not one of them. What I am saying is that AI is not evolvable by the rules of Darwinism. If we do succeed in producing an AI program it will almost certainly be irreducible since most of the algorithms we are familiar with, like Householder, Gaussian Elimination and FFT are irreducible and do not occur in Nature.

If we are to discuss AI it is necessary to first discuss pseudo intelligence and to draw a distinction between what appears to be intelligent and what is actually intelligent. This is not an easy matter.
I don't see how the program needs to be irreducible. As any programmer knows, it is possible, and far easier, to produce the same functionality within a program by using ten times as much code. Pseudo intelligence(PI) is far harder to achieve with small amounts of code than with vast amounts of largely superfluous instruction. Admittedly, after the initial breakthrough, it is likely that the necessary inheritance of the required instruction set would eventually streamline the program significantly. Failure to pass on superfluous code would in no way inhibit the population through each generation.
Back to the question of P.I., this is not too difficult to achieve by the use of comparitive algorithms. Using such an algorithm, a computer can "learn" behaviour and use this to choose a course of action to an event that it has not been programmed to respond to. This can be done by defining entities, actions and relationships as types. The type is dependent upon the characteristics. As an example:
A car of type "vehicle" travels towards the computer.
Assume here that through several generations only the computers that moved from the cars path survived and so passed on this trait.
A lorry of type "vehicle" travels towards the computer.
The computer moves out of the way as the comparison between lorry and car shows it to be of type vehicle and the correct response is of type avoidance.
This is incredibly simplistic but it shows that it is possible for a computer to show pseudo intelligence. Increase the database and number of types and it is soon possible to show very acceptable levels of pseudo intelligence. I am unsure where you would draw the line between pseudo and real as we ourselves could be described as pseudo intelligent.
Even humans use this type of comparative assessment as evident in schools and by our frequent use of analogy. Complex problems are always broken into smaller, more easily manageable pieces. Even the most complex decisions are fomulated depending on the entity, action,relationship principle.

...Somehow I think most people would believe in Intelligent Design rather than infinite gynothropism.

Assuming either one is correct. This is a little like saying if we dispove evolution(according to Darwin) we prove creation. It is quite possible that both ideas are incorrect.

[Ian Parker]

If we are to discuss AI it is necessary to first discuss pseudo intelligence and to draw a distinction between what appears to be intelligent and what is actually intelligent. This is not an easy matter.

It is indeed not an easy matter. On one definition you could say that anthing which does anything, for example Gaussian Elimination is intelligent. A computer will solve simultaneous equations better than we can. We do not however class this as being artificial intelligence.

The program you are advocating is precisely the program of CYC and Lenat. The starting point of CYC is completely irreducible, I don't think anyone could really deny this. I think that perhaps we could make a start by mandating some evolution at the level of an inference ingine. An inference engine or expert system is not intelligent unless it can produce new entries acting autonomously.

This leaves us with an intreguing thought. Suppose we use a GA not to produce a boot but to modify an inference engine. This would perhaps be the start of intelligence. The thought is interesting because we are indeed using GAs but to develop the engine. In human evolution it is envisaged that a boot evolved.

[Curious]

The program you are advocating is precisely the program of CYC and Lenat.

Then might I suggest, at the risk of sounding arrogant, that they are on the right path. If they have encountered problems with this method then perhaps they should modify their algorithms to incorporate pseudo logic.