Project 4
Start the paperwork now to get your instructional computing account to hold your WWW pages.
Do not wait until the last minute, as other people have to do things for you.
(1) Go to the sinc site in main library and ask for an ic account. Takes about 24 hours.
(2) Send email to webmaster@sinc.sunysb.edu to request a place to put your WWW pages.
(3) Check out http://www.sinc.sunysb.edu/IC/howtolynx.html
(4) Once this is set up, use ftp to transfer your files, eg. `ftp libws1.ic.sunysb.edu', login, and `put filename.htm'
Can Computers Think?
Thinking machines are one of the staples of Science Fiction. The question ``Can computers be made to think?'' is one of the most important in modern philosophy.
In some sense, the question is analogous to the question ``Can submarines swim?'' It hinges on the definition of what really thought is.
Does thinking require consciousness?
Do animals think?
Does a program capable of playing chess think?
What about thinking outside a fixed logical system?
The Turing Test
Alan Turing, a brilliant pioneer in computer science, proposed that a machine could be shown to think if it could fool an interrogator into thinking that it was human.
In the Turing test, a human communicates via a keyboard with some entity, person or program, and makes a decision about which one it is. If a machine can fool a person into thinking it is a person, then it must be intelligent.
The Turing test is a very hard challenge, and no serious attempt has been made to pass it. Any such program must have a large amount of knowledge, and understand what it means, in order to answer such questions as:
Are you a computer?
How does this poem strike you?
Describe your parents.
Eliza
There has been some success building conversational programs over restricted domains.
A famous program was Eliza, which pretended to be a psychoanalyst. It fooled many people into confiding in it, but this is because the role of the program is to lead the discussion or echo back the questions.
Eliza only seems intelligent, because it really knows very little. Once you start to play with it, you soon start to see its flaws, as evident from the transcript against Parry.
Game Playing Programs
Chess and other games represent one of the oldest attempts to build intelligent programs. They have been very successful - the best current program Deep Blue beat Gary Kasparov in a game, though not in a match.
Computer Chess programs work in a completely different way than chess masters. People consider relatively few moves at each turn, making a careful evaluation of each. Successful programs do a brute force search for as far as they can, and make a quick evaluation of all of them.
If computers could look deeply enough, they could always win, however, there are about possible games to consider! The fastest machines can only look 12-20 moves ahead.
Sometime soon, a computer is likely to become world chess champion, but not because it thinks.
Expert Systems
Expert systems are programs which attempt to encode a lot of knowledge about a particular topic in the form of rules, and then use these rules to make decisions.
A typical rule might be ``if X is a company, and Y manages a company, the Y works for X. Reasoning about such rules requires logical inferences.
Expert systems have had some success in medical diagnosis, but would you trust being diagnosed by a computer?
Agents, are software programs which make decisions for you based on rules like expert systems. There is considerable current work on producing agents for things like identifying the news articles you will be interesting it, ordering your email, or scheduling your appointments.
The State of the Art in AI
The Turing test is nowhere near being passed, although there is an annual contest for natural language programs.
Machine translation of natural languages is commonly employed as a first pass in certain applications.
Computer vision systems are getting better. CMU has developed an autonomous vehicle, capable of driving a car by itself slowly on special roads.
Speed recognition systems keep getting better, with larger vocabularies and more speaker independence.
Natural language parsing programs are getting better, but they still do not understand what they are doing.
Neural Networks
Because brains are real thinking machines, it makes sense to try to make computers which work the way brains do.
But how do brains work? Instead of one, powerful processor, they seem to be built from billions of neurons connected together. Each neuron can be modeled as a very simple processor.
Brains work quite differently from computers, although a lot is unknown about how knowledge is stored in the brain.
Research is continuing on building artificial neural networks. Although today's networks are small and crude, they have proven useful is certain pattern recognition applications.