Artificial Life and Intelligence

The game of "Life"...

My mom and I were discussing Watson on Jeopardy!

She is in her eighties and expressed some fear with regard to Watson; her thoughts on talking computers shaped forever by the HAL 9000 in 2001 A Space Odyssey.

What's interesting is how people associate intelligence with something like Watson - not the intelligence it took to create Watson but instead the fact that Watson appears intelligent even though its a simple machine. 

Since 1950 the measure of whether something like a computer is intelligent has been the Turing Test.  Basically this is a test where a human judge and a "subject" (a computer) communicate remotely to each other (so the human cannot tell simply by external evidence whether the subject is a machine or not.  The human gets to ask questions and the subject responds.  If the human judge cannot reliable determine if the subject is a machine the machine is said to have passed the test.

Something like Jeopardy! is not a Turing Test for several reasons.  One is that the questions are simply "right or wrong" questions.  So, if I had a book of that compiled all questions every asked on Jeopardy! I could always find the correct answer - effectively Jeopardy! itself is the database of all such questions and answers.

This database of questions does not represent intelligence - merely simple look ups.  Certainly Watson has to string together information to "find" answers - but the answers have to be in his database in the first place for him to find them.

Then there is the "domain" of Watson's knowledge.  This is basically a measurement of how big the "realm of knowledge" is that Watson can work from.  IBM said that this was comprised of 200 million pages of documents: dictionaries, encyclopedias, and so forth.  This realm is actually very small because its limited to reasoning about a fixed set of information (fixed in the sense that it does not change for the "life" of Watson's play on Jeopardy!) and it does not address current information (news, weather, and so on), knowledge about social things, information about doing things, and much, much more.  Watson also cannot "see" or "hear" or "touch" nor does it have a body with which to relate feelings.

All this means Watson cannot know about what it feels like to do anything, cannot be asked about whether he likes the weather today or weather (he/she/it) loves (he/she/its) significant other.

Watson has no internet connection - which means Watson cannot know about things that are changing or that are wrong (for example seeing the face of Elvis on Mars).  So Watson only deals with correct information - correct in the sense that it does not have to decide which of two conflicting facts is correct (this may happen inside Watson at a small scale but not in the sense that someone or something is actively trying to deceive another, e.g., a Bernie Madoff).

Watson is also "deterministic" - that is it is going to come up with the same answer today or ten years from now given the same database.  So he is not "learning from his mistakes."

Don't get me wrong - Watson is certainly an impressive technical achievement - be he is certainly no HAL 9000.

There are many other impressive technical feats of computer engineering on par with Watson.  For example, Google's self driving cars, the "Deep Blue" chess computer, and many others.  But they are just that - engineering.

Humanity certainly has the technology to attempt to create something like a HAL 9000 or a SkyNet today.  A few hundreds of millions of available computers all networked together across the face of the earth could be linked with AI-type software to demonstrate intelligence.  But if each computer equals a single neuron that network provides only about as much intellectual horse power as an octopus (see this link).

The problem and challenge with all this is simple: Intelligence exists for a purpose.  You cannot have software acting intelligently without that software having some reason to be intelligent, i.e., survival.

The first thing I would do is create an artificial environment in the vast see of computers - a place where something "alive" could be represented, could move, learn, perceive and act, could carry the equivalent of genes, and could have a purpose.  This concept already exists to a degree for projects like SETI, for networked games (Second Life, World of Warcraft, etc.)

I would then construct some type of artificial "life" designed to live in that environment - think of it as a computerized player in one of the games I mentioned. 

It would not be hard to have some sort of "life" living in this type of world - what it might be or might look like would be hard to say, but I don't see why humanity isn't working on this... Well, they are or rather were, actually (see this), but most of the links don't exist so perhaps the artificial life they were working on is dead???