It had to happen eventually.

So anyone who knows me knows that I have a sort of reputation for posting sub-par and rather eccentric code. I'm fine with this. I like making AutoCAD video games and useless but entertaining programs, even if the techniques are bad and no one cares. I think, though, that this post should top it.

So, as many of us should know, the LISP programming language was originally designed for work in artificial intelligence. I can hear many of you facepalming in anticipation of what's coming. For a while now, I've been wondering just how such a thing would come about. Sure, I've ran into "artificial intelligence" programs, but so far all of them are nothing more than processing and regurgitating data. Hell, by that definition, the Google search engine is A.I.; it records your search queries and what hits you select, processes them to find commonalities, and adjusts the results for your next search query to suit your personal statistics. The problem with these programs is that they do exactly what they were designed to do in and only in the limits they were designed within.

I decided that the concept of artificial intelligence was too much fun to pass up, so I thought about it for a while, and hit upon the idea that (as I'm sure many who came long before me have long since figured out, tried, and discarded) that complex actions are learned from simple actions. That got me thinking back to Psychology 101 where I learned about Pavlov and his dogs. Classical conditioning! Consider an empty mind that does nothing but random actions, but records the most minute detail about those actions, their effects, and the impact on the surrounding environment. That's a lot of data to process, but once that data is processed, a system of judgment can come along to decide which actions are "best". What do I mean by "best"? Most suitable for survival. Survival is an instinctive characteristic which is pre-programmed into most, if not all, of us. From there I had the groundwork for my first attempt: declare a base set of instinctive needs for survival, then create an interface by which an automated process performs random actions which are then judged and analyzed, as well as an automated environment that is analyzed as well. Over time, this program should be able to learn certain things by that simple facet of classical conditioning: do, analyze, revise, do something else.

I made some good headway on it and got some pretty funky code down, but ultimately the project began to taper off. The more I worked at it, the more I began to wonder if I was on the right track. The question of "what are the limitations to this approach?" kept popping into my mind, and I was discouraged at the fact that I simply couldn't answer it. The fact that I had no answer meant I did not know enough about the process. Rather than attempt to refine my knowledge, I opted to look at a different process: data processing in the human brain. If I could figure that out, surely that would give me my answer. For, after all, a learning program has a limit on the amount of knowledge it can process and contain, just like a human brain. This approach was just what I needed: the realization that I was dead wrong. The human brain does not process data. Frankly I don't know how the hell it works, but data in the brain is not stored as a variable like it is in code. It cannot be accessed by printing a variable to the REPL. I briefly played with the idea of trying to write a series of functions to mimic the process of the human brain; where data (electrical stimuli) is passed to a function (dendrite-neuron) which processes the stimuli then passes it to a function or series of functions (axon-terminal-synapse). The sheer number of functions and how to interconnect them drove me batty. I decided to take it a step further.

A cell is the most primitive form of what we call life, but parts of the cell are superfluous. There are things a cell doesn't need to be "alive". Where's the cut-off? How much can you remove before it's no longer alive? I dug down and down, and finally came upon the RNA World hypothesis. This theory states that life began as an RNA chain which, upon encountering a particular enzyme, split itself (with some mutation), and the process repeated. This, it turns out, was duplicated in a laboratory, so the process is sound. This is an entirely CHEMICAL process and none of its constituents are considered to be alive. Something like this, I could work with.

So the work began. I'm still in the developing stages, but I've made some interesting progress. So far it's just a thought experiment, and before I plunge into the code, I wanted to see what everyone else thought.

My current plan stands thus: Compare the actions and interactions of atoms and molecules, and distinguish a parallel in the world of computational processing. Then, set up an environment where the chemical reaction parallels can take place and develop.

So far, I've established that:

Protons, Neutrons, Electrons = Valid ASCII characters in the code language set

Stable atoms = Code commands recognized by the interpreter as valid

Stable molecules = Functions that are able to be successfully processed by the interpreter

Further, I've deduced that the process of a chemical interaction should have its equivalence in a number of functions being processed and potentially combined by the interpreter, yielding a new function or function set.

Ideally, this would continue brute-force style until... something happens.

What do you think, gents?

With the amount of work involved and no obvious benefit, waste of time IMO

With the amount of work involved and no obvious benefit, waste of time IMO

Sadly, I have to agree.

http://www.watsonlisp.com/

Wow. Shot down. Oh well, guess it was my fault for expecting differently.

Wow. Shot down. Oh well, guess it was my fault for expecting differently.

Wow. Shot down. Oh well, guess it was my fault for expecting differently.

I like the idea - artificial intelligence is awesome - but I can see you investing a lot of time into this without much return.

It beats listening to politicians arguing about how to clean up the mess in the gulf. I say go for it dude.

First, I want to move this thread (since it really is not Lisp or for that matter CAD-related) but I don't know where to yet.

Second, AI is very interesting and just a bit scary. I can't say that I understood all that you wrote about what it is that you are aiming for, but research is never wrong. And AI are coming more and more and more, so if that is something you are passionate for, go for it!

I have to wonder though if AI can ever be as intuative as humans. Computers are still somewhere down the line programmed by a human, and even if you can input a million different scenarios in a computer and the appropriate respons - there is always a new scenario that is not in the program. And if an AI really could react "spontanously" then...well then its just scary.

Why Tiger, that sounds like a senario straight out of the movie Terminator 3: Rise of the Machines. Should we expect to see a company called Cyberdyne in the near future too?

Why Tiger, that sounds like a senario straight out of the movie Terminator 3: Rise of the Machines. Should we expect to see a company called Cyberdyne in the near future too?

As far as I know in Terminator 1 the War started in 1999 so I guess not :wink:

Sounds like a very complex project. I think it would be fun to do even if it has no useful application.

I just had thought go through my head...

What is the secret to life?

(Hitchhikers guide to the Galaxy-- *They ask the super computer*)

Supercomputer: Come back in umpteen million years...

They do. Then they ask again and the super computer responds "seven".

As far as I know in Terminator 1 the War started in 1999 so I guess not :wink:

I've always thought the term 'Artificial Inteligence' was an oxymoron. But that is coming from a relative moron.

Computers will never be as 'smart' as humans. They will never develop the ability to ignore a deadline and go outside to be chased around the yard by the kids and dog with the garden hose.

i assume you've played with cellular automaton software popularised by dennett?

sounds similar to your thoughts.

i say go for it. but i have, like, 3 posts, so what do i know?!

the super computer responds "seven".

sorry to be pedantic:

they ask "what is the answer to life, the universe, and everything?"

DeepThought (the supercomputer) responds "forty-two"

i assume you've played with cellular automaton software popularised by dennett?

sounds similar to your thoughts.

 

i say go for it. but i have, like, 3 posts, so what do i know?!

Aw, come on. I am not a Postist. With 3 posts, at least you sit level on uneven ground.

Dennett? not even close. There's nothing mechanical about conciousness.:wink:

Nobody can yet even attempt to design a program that can randomly forget things it doesn't need to know, and some it does need, or to jump to conclusions based on gut instinct. Wait, that sounds a lot like Windows.

Watson6.0 by http://www.watsonlisp.com/ is NOT a waste of time.

It saves time by finding detail drawings by text and attribute content.