STAR@LAVALVM1.BITNET (Spencer Star) (05/06/88)
This discussion about free will doesn't seem to want to go away, so
let me ask a question or two. Let's agree that a deterministic world
is one in which a given state of the world at time 0 is sufficient to
determine the world state at time T, where T is after 0. An indeterministic
world is one in which that proposition is false. These are propositions
about the nature of the world not about what we know about the world. In
many situations we do not have enough knowledge about the world to be able
to predict without error state T from the available information about
state 0.
Free will seems to me to mean that regardless of state 0 the agent
can choose which one of the possible states it will be in at time T. A
necessary precondition for free will is that the world be indeterministic.
This does not, however, seem to be a sufficient condition since radioactive
decay is indeterministic but the particles do not have free will.
Free will should certainly be more than just our inability to
predict an outcome, since that is consistent with limited knowledge in
a deterministic world. And it must be more than indeterminism.
My questions:
Given these definitions, (1) What is free will for a machine?
(2) Please provide a test that will determine
if a machine has free will. The test should
be quantitative, repeatable, and unabiguous.
Perhaps McCarthy could summarize how he would answer those questions
based on his article.
--Spencer Starbwk@MITRE-BEDFORD.ARPA (Barry W. Kort) (05/06/88)
Spencer Star asks:
(1) What is free will for a machine?
(2) Please provide a test that will determine
if a machine has free will. The test should
be quantitative, repeatable, and unambiguous.
I suggest the following implementation of Free Will, which I believe
would engender behavior indistinguishable from a sentient being
with Free Will.
1) Imbue the machine with a Value System. This will enable the machine
to rank by preference or utility the desirability of the anticipated
outcomes of pursuing alternative courses of action.
2) Provide a random choice mechanism for selecting among equal-valued
alternatives.
3) Allow the Value System to learn from experience.
4) Seed the Value System with a) the desire to survive and b) the desire
to construct accurate maps of the state-of-affairs of the world
and accurate models for predicting future states-of-affairs from a
given state as a function of possible actions open to the machine.
--Barry Kort