Various projects exist today aiming at building a true Artificial
Intelligence. Sometimes these researchers use the term AGI, Artificial
General Intelligence, to distinguish their projects from mainstream AI
which tends to focus on specific tasks. A conference on such projects
will be held next year, agi-08.org.
Suppose one of these projects achieves one of the milestone goals of
such efforts; their AI becomes able to educate itself by reading books
and reference material, rather than having to have facts put in by
the developers. Perhaps it requires some help with this, and various
questions and ambiguities need to be answered by humans, but still this is
a huge advancement as the AI can now in principle learn almost any field.
Keep in mind that this AI is far from passing the Turing test; it is able
to absorb and digest material and then answer questions or perhaps even
engage in a dialog about it. But its complexity is, we will suppose,
substantially less than the human brain.
Now at some point the AI reads about the philosophy of mind, and the
question is put to it: are you conscious?
How might an AI program go about answering a question like this?
What kind of reasoning would be applicable? In principle, how would
you expect a well-designed AI to decide if it is conscious? And then,
how or why is the reasoning different if a human rather than an AI is
answering them?
Clearly the AI has to start with the definition. It needs to know what
consciousness is, what the word means, in order to decide if it applies.
Unfortunately such definitions usually amount to either a list of
synonyms for consciousness, or use the common human biological heritage
as a reference. From the Wikipedia: "Consciousness is a quality of the
mind generally regarded to comprise qualities such as subjectivity,
self-awareness, sentience, sapience, and the ability to perceive the
relationship between oneself and one's environment." Here we have four
synonyms and one relational description which would arguably apply to
any computer system that has environmental sensors, unless "perceive"
is also merely another synonym for conscious perception.
It looks to me like AIs, even ones much more sophisticated than I am
describing here, are going to have a hard time deciding whether they
are conscious in the human sense. Since humans seem essentially unable
to describe consciousness in any reasonable operational terms, there
doesn't seem any acceptable way for an AI to decide whether the word
applies to itself.
And given this failure, it calls into question the ease with which
humans assert that they are conscious. How do we really know that
we are conscious? For example, how do we know that what we call
consciousness is what everyone else calls consciousness? I am worried
that many people believe they are conscious simply because as children,
they were told they were conscious. They were told that consciousness
is the difference between being awake and being asleep, and assume on
that basis that when they are awake they are conscious. Then all those
other synonyms are treated the same way.
Yet most humans would not admit to any doubt that they are conscious.
For such a slippery and seemingly undefinable concept, it seems odd
that people are so sure of it. Why, then, can't an AI achieve a similar
degree of certainty? Do you think a properly programmed AI would ever
say, yes, I am conscious, because I have subjectivity, self-awareness,
sentience, sapience, etc., and I know this because it is just inherent in
my artificial brain? Presumably we could program the AI to say this,
and to believe it (in whatever sense that word applies), but is it
something an AI could logically conclude?
Hal
--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups "Everything List" group.
To post to this group, send email to everything-list.domain.name.hidden
To unsubscribe from this group, send email to everything-list-unsubscribe.domain.name.hidden
For more options, visit this group at
http://groups.google.com/group/everything-list?hl=en
-~----------~----~----~----~------~----~------~--~---
Received on Sat Jun 02 2007 - 17:26:29 PDT