RE: PHIL: Is it ethical to create special purpose sentients?

Billy Brown (
Tue, 2 Mar 1999 08:09:53 -0600

Glen Finney wrote:
> We now have a computer that has knowledge of itself and the world around
> it, able to understand and communicate with humans; able to pass a Turing
> about as much as any human. CES even has a sense of right and wrong, but
> is still devoted to doing Cardiology alone, and then only when it is
> to CES. And all the profits are going to the creator of CES, the retired
> human Cardiologist. Now, imagine we eventually cure all disease, perhaps
> trade in our old bodies for robotic ones. No need for a CES. The CES,
> although aware of this, doesn't care. CES is now obsolete, with no other
> goals, and no resources even if CES did have the motivation to change. So
> is simply turned off. Was CES ever "truly" conscious? At its height, its
> patients might have sworn CES was. Did CES deserve any of the share of
> earnings? Should CES even be created? What do you all think of this
> hypothetical situation and variations of it?

An expert system program could not grow into a person by gaining more knowledge. It really can't even make progress towards that goal. To understand why not, we need to look at what is happening inside that program:

An expert system is really just a fancy database search engine. A simple one takes a set of data you provide, searches its database for the best fit, and spits out its results. A more complex version uses an iterative version of the same process: use the initial data to get a list of possible diagnosis, then gather additional data to weed out possibilities until a conclusion can be reached. A really advanced model could use the same kind of mechanism to talk you through a complex procedure, check your goal against an ethics database, etc.

Now, at no point does the expert system actually understand what it is doing. It just manipulates a bunch of abstract tokens in a pre-programmed fashion. In fact, I would argue that there is nothing in there capable of understanding anything at all. The program will be able to apply rules of logic to its search for an answer, but that is a very rudimentary form of thinking. The program also has no self-awareness, no sense of identity, no volition, and (probably) very little ability to learn or remember.

On a side note, it probably isn't possible to make a program like this that could pass the Turing test, make accurate moral evaluations, or cope with anything else outside of a very narrow specialty. There is at least one big project that has been trying to make an expert system do these things for over a decade now, and they aren't having much luck. The world is just too complex for this kind of knowledge representation to deal with. What you need is a program with more of an ability to reason, understand, and learn on its own.

Billy Brown, MCSE+I