Follow-up on Artificial Intelligence
Way-back machine: this post was made on 1 Dec 2007 on Spaces.Live.com
Comments were not migrated.
Again it has been many weeks since I blogged. I thought that I could reply to a question about my previous post within a week. Unfortunately, work has kept me busy every weekend, my technical blog and my involvement in forum discussions at forum.skeptic.za.org has taken all of my spare time.
Enough with the excuses, time to reply to the question.
Snoop asked (on 2007/11/11 22:51) :
“Please explain to me however at what point – to your mind – would the barrier between AI and artificial life be erased. The behaviour of the beforementioned eaters does to some extent simulate life to some extent.”
Okay, the distinction between them is very clear. You could get more complete descriptions from Wikipedia on Artificial Intelligence, Artificial Life and Artificial Consciousness, but the question asks for my opinion. Having not yet read the articles myself, I risk contradicting some widely accepted definitions. Anyway, my definitions are as follows:
- Artificial Intelligence is about developing machines or programs which have a goal to achieve and achieve them utilising one or more of a number of Artificial means of decision. Most of the time this science is the most hum-drum of the three (although to me it is fantastically exciting). Some of the following examples should help understand the scope of the science;
- Robotic motility systems such as the robots which assemble cars or walk up or down stairs.
- Vision systems such as a robot which watches a conveyor belt and removes apples which do not look good enough to be sold or the Traffic Department reading license plates from pictures or video in order to issue speeding fines.
- Exhaustively programmed systems which contain all rules and know all outcomes and can make fast decisions such as the autopilot on airliners.
- Probabilistically programmed systems which can rapidly calculate thousands of alternatives and select the one which probably leads to success such as Deep Blue beating Kaspirov at chess.
- Neural networks which compare new problems to old problems in order to choose the best course of action. You can interact with a neural network at www.20Q.net which is an online 20 questions game utilising neural networks. Play classic 20Q and start by thinking of something like “a house plant” or “a wedding ring” or “a spoon” or “love” and see if it can guess (it will try up to 25 guesses). Don’t lie … it will tell you at the end if you lied.
- Generic algorithms which I described in the comments of my previous post.
- Artificial Life is about simulating a process in life or simulating the actions of living things. Again, this is a hum-drum science involving scenarios of life and not concerned with whether the entities in the system are conscious. The example from my previous posting is not concerned with Artificial Life although there are some similarities. Some examples of wha A.L. is about are:
- Modelling the behaviour of virii when infecting a life form.
- Modelling the behaviour of lions when 30 are released into a 10 000 ha preserve where there is an estimated population of 2000 kudu. Then trying a scenario where 40 are released and comparing the outcomes.
- Artificial Consciousness is the attempt to create a consciousness via a programmed system.
Artificial Consciousness is the science that most people confuse with Artificial Intelligence. I don’t know much about this myself but some of the core principles are grounded in emergent behaviour or emergent consciousness. An Artificial Consciousness system designer does not model consciousness directly, it is likely to be impossible to design a system which can simulate every aspect of consciousness, instead a system designer starts with small designed components which have a simple function, when all the simple components interact new (unexpected) functionality emerges. This is close to the consciousness of animals and insects where we know that the simple interaction of nerve cells (which have very limited function) lead to a level of consciousness where reactions are automatic and a level of decision is possible the more that additional nerve cells are involved the greater the “level” of consciousness.
Michael Crichton’s book Prey explores an idea where scientists designed networked nanobots where nanobots had simple function but the interaction lead them creating a consciousness. In true Crichton style it leads to disaster (a la Jurassic Park).
So, how do you know that you have found consciousness? That is the most dificult question. The answer is in analysing consciousness in animals and what “pieces” are needed to have consciousness. Those, IMHO, are free pattern recognition and memory. Here free pattern recognition doesn’t refer to a designed pattern recognition (like looking for bad apples) but rather an emerging pattern recognition that was not designed in the system but the system has it. Also, memory doesn’t refer to the memory containing the individual components but rather that the state of those components retaining information.
So the short answer to the question is that appearances can be deceiving. If the system wasn’t designed for the purpose of immitation of life or consciousness then no matter what the results may look like, that it not what the demonstration is showing.