CE 00875-3 Character AI

almondpitterpatterIA et Robotique

23 févr. 2014 (il y a 3 années et 5 mois)

67 vue(s)

CE 00875
-
3 Character AI


Diane Bishton
-

K229

(d.k.bishton@staffs.ac.uk)

these slides at http://www.blackboard.staffs.ac.uk





Free Will




Introduction




Despite the fact that we find ourselves influencing & being
influenced by other people on a daily basis, as human beings we
have the notion of ‘Free Will’.


In this lecture we’ll look at how Free Will may be defined & will
describe it in terms of potentially useful properties.


The question for you will be,


“Is there such a thing as Free Will, or not ?”

Whatever your answer there will be implications, not the least of
which will be how you might incorporate such a concept in an AI.


Evolution of Free Will




“ Free Will…(may) have been selected for over the course of
evolution…we can have both Free Will and a mechanism to
implement it”


Edmonds, B. in Davis D. (2005) “Visions of Mind. Architectures
for Cognition and Affect” pp108
-
9


The suggestion is that Free Will brings advantages to our species,
and that some process has occurred through millennia which has
changed the structure of our brain


the mechanism.


But hold on…

If there’s a mechanism, then because mechanisms are predictable,
how can there be free will ?


The bulk of this lecture is based on the above chapter.

Modelling Free Will



All modelling occurs in a context


models have scope (the set of
circumstances in which they work) and hence limitations.


The scope has been determined by the contexts in which the
model was first developed & tried out e.g.


Flowcharts
(actually earlier as Therbligs in work redesign)

&
batch processing systems

State transition diagrams & real time systems


Sometimes, of course, we try to extend/alter the scope & try out
the model in new contexts (in the case of Free Will, in social
contexts). May or may not ‘fit’. The less a model fits, the more it
is taken to be under the influence of variable external factors.
Eventually, we may need a new model, a new theory.

The Functional Description or
‘Properties’ of Free Will





According to Edmonds, Free Will has the following properties
that have given us a selective advantage in nature i.e. in the world
in which we find ourselves:


Exterior Unpredictability (context: what a competitor sees)


Interior Rationality (context: ones internal cognition)


Social Accountability (context : translating internal to external)


These
properties

are interdependent (& about different contexts,
as indicated above.)

Exterior Unpredictability (EU)




Are co
-
operation & competition mutually exclusive ?

(The seminars next week will explore aspects of these).

Typically as humans we find that in any situation there are elements
of both.


EU is about leaving a competitor unable to predict what you’ll
do next.

Meet new challenges in unexpected ways.
Sometimes, the rational
thing to do is something that
appears

to be random, and rationality
is a condition of membership of many social groups.

Of course, rationality depends on the expectations of that group in
terms of norms & rules established (incentives / sanctions) by that
group


Ministry of Funny Walks. Form, Storm, Norm, Perform
in team building.
Your examples ?

Interior Rationality (IR)






IR is about justifying your actions to yourself.


You have goals which you wish to achieve.

Where we are aware of these goals, then we might make plans to
get closer to them.

Sometimes, we may achieve a goal ‘by accident’, but then can
rationalise that achievement where we

either:

need to understand it ourselves,

or:

where we need to ‘make up a story’ ready to rationalise it

to someone else (see SA next).

The rationalisation shows how we’re being consistent in trying to
achieve the goal. Self preservation. Motives focus on oneself.

Whatever the cost


bravery, or stupidity ?
Your examples ?


Social Accountability (SA)



SA is all about passing on your rationalisation to others.


The rationalisation has to be understood by others.

It’s a public account of your decision
-
making process that shows
there was a structure/plan in relation to something which
appeared apparently irrational.


Examples:

Awarding banks £130,000,000,000 (Feb 2009)

Awarding directors of ‘failing’ banks huge bonuses

Going into a battle / fight when obviously outnumbered

(Is much of Politics about SA ?)

Your examples ?


Inter
-
relation of the Properties





Because humanity is a social species, SA must affect any action
that may or may not be taken.


How big the role SA has in decision
-
making depends on the
social group that would have to be accounted to, and how
‘distant’ it is from the decision maker.


The less one is seen (or sees themselves) as belonging to a
social group, the more free will the individual has.


EA could be seen as pure free will if there were not IR & SA to
affect what could otherwise be seen as a completely random
process.


IR is based on self preservation, but this is ‘kept under control’
by ER & SA.

A possible mechanism for Free Will ?




It is proposed to implement free will as a developmental process.

And that this process should be evolutionary in approach.

The result should be the ability to direct a mature brain to the
advantage of the individual, while maintaining the
individual’s place in the community, for example by using

Genetic Programming:

1)
A creative technique that often produces unexpected solutions

2)
Recombination maintains maximum variety in a population
-

preserving
sub trees, but with different combinations

3)
In the solution space there are different solutions that result in the same
behaviour in a specific training context, but which arbitrarily change in
other contexts

4)
Solutions match the goal (the goal is implicit in the selection mechanism)


Does Free Will really exist ?





Edmonds suggests in his version of Free Will that it does exist,
but that it is then
controlled

by a mechanism which essentially
incorporates the possibility of generating random actions which
are ultimately kept in check by our internal cognition and external
forces. Is this how you would imagine Free Will to operate ?

Thinking about the earlier material, also add to the mix if you
wish (there’s that free will again):


The work of Prof Susan Greenfield
-

brain state comes first)


Ideas from Behaviourist (as against Cognitivist and Constructivist
theories of learning)


Theological (Religious) ideas


without a God, there is chaos

Definitions of Free Will




“the ability or discretion to choose; free choice”


“the power of making free choices that are unconstrained by
external circumstances or by an agency such as fate or divine
will”

www.thefreedictionary.com


“the power of making free choices unconstrained by external
agencies”

www.websters
-
online
-
dictionary.org


Postscript





If there’s time:


Did you do the ‘experiment’ from last time ?

( Experiment (p.m.): look at the staff photographs in the Octagon.
Put the names of the face
-
holder under ‘higher’, ‘lower’ title.)

If so, let’s see what you got.