Towards Implementing Free Will - Bruce Edmonds

2 Conceptions of free-will

It is inevitable that in any implementation process one will move from an idealised to a realised conception of what one implements. Thus here I am not so interested with artificial or idealised conceptions of free-will, determinism, randomness etc. but with more practical concerns. For if a certain conception of free-will makes no practical difference then it is irrelevant to a discussion about implementation (and quite possibly to everything else as well). For if it is impossible to tell whether an entity has a certain property and that entity can do all the things without that property as with it, how can it be relevant in practice?

From this practical perspective, free-will is something that a normal adult human has but an newly fertilised human embryo hasn't. It means that an agent is free to choose according to its will, that is to say that sometimes it is its deliberations on how to achieve its goals that determine its actions and not just its circumstances (including past circumstances).course, many aspects of traditional philosophical analyses of free-will are relevant if one avoids the pitfalls of extreme idealisation. For example the points listed below come from philosophy, but are formulated with practical concerns in mind:

The process of deliberation leading to a choice of action has to be free in the sense that it is not constrained to a particular "script" - this means that there is also some choice in that deliberation, as well as choice in how to make that choice, and choice in how to make the choice in how to make that choice etc.;
In some circumstances, if others with whom the entity is competing are able to effectively predict its actions they may well exploit this in order to constrain its choice to its detriment - thus it can be important that actions are not predictable by others;
In order for an entity's will to be effective it has to be able to perform some processing that tends to result in actions that (as far as it can tell) furthers its goals - in particular it needs to be able to consider the likely consequences of different possible strategies and choose amongst them with a view to furthering its goals;
It must be possible that sometimes it might have taken a different action to those actually taken - that is, given indistinguishable circumstances, it would not simply repeat past decisions (even if it did not recall them).
In order to have an entity's decisions allowed by a society of peers it is often necessary that it is able to give an account of its reasons for actions that impinge upon that society, reasons that would be deemed acceptably rational - for those that are not reliably rational can pose a danger to the rest and hence may be prevented from taking certain actions.

These are the criteria I will take to guide my thoughts about implementation rather than abstract issues of theoretical determination and the like. They seem to capture the more important aspects of free-will - the aspects of free-will that make it worth having [3]. This is a similar approach to that of Aaron Sloman's [13], except that it focuses more upon a single issue: how can we develop an agent whose decisions are determined by its deliberations and not completely constrained by its circumstances. He is right to point out that an entity's decisions can be constrained in different ways and is dependent upon the capabilities and structure of the entity. However the multiplicity of factors does not dissolve the central issue which is concrete and testable; for any entity placed in the same circumstances one can measure the extent to which entity acts in the same way and (with humans) collect indirect evidence (by interview) to see the extent to which the actions correlated with the prior deliberations.

Towards Implementing Free Will - Bruce Edmonds - 16 MAR 0

[Next] [Previous] [Top] [Contents]

Generated with CERN WebMaker