ian@mva.cs.liv.ac.uk (03/11/90)
In article <1407@oravax.UUCP>, ian@oravax.UUCP (Ian Sutherland) writes: > In article <2752@castle.ed.ac.uk> aighb@castle.ed.ac.uk (Geoff Ballinger) writes: >> >> An intelligent system is aware of both itself and it's >>environment (and the relationship between the two) and can respond >>accordingly. >> > > Comment 1: what do you mean by "aware", particularly when applied to > the system itself? > > Comment 2: what does it mean to respond "accordingly"? > Exactly! The original statement shifts the search for a definition of `intelligent' to a search for definitions of `aware' and `respond accordingly'. Of course, if these two terms can be more easily pinned down than `intelligence', we're getting somewhere (if we want to define intelligence). I'm not sure that they can! To dig out an old example, I'm pretty sure that your definition of intelligence could be applied to a thermostat! I feel that any definition of intelligence should have a context. In some circumstances it may be useful to ascribe intelligence to a thermostat (for example when telling a very small child how a room stays the same temperature), even though normally it would not. Ian ---
g2g@mentor.cc.purdue.edu (Ranjan Muttiah) (03/16/90)
In article <5177.25fa2fb6@mva.cs.liv.ac.uk> ian@mva.cs.liv.ac.uk writes: >I feel that any definition of intelligence should have a context. In some >circumstances it may be useful to ascribe intelligence to a thermostat (for >example when telling a very small child how a room stays the same >temperature), even though normally it would not. If mathematics be the most precise of sciences and its desiderata of numbers and sets are left undefined, then how in a more imprecise field such as AI can we define intelligence ? Answer: You don't !
kp@uts.amdahl.com (Ken Presting) (03/17/90)
In article <8533@mentor.cc.purdue.edu> g2g@mentor.cc.purdue.edu (Ranjan Muttiah) writes: > >If mathematics be the most precise of sciences and its desiderata of >numbers and sets are left undefined, then how in a more imprecise >field such as AI can we define intelligence ? > >Answer: You don't ! The axioms of a set theory or a number theory can serve as *either* a a definition of the entities under discussion, or as assumptions regarding the properties and relations of those entities. If the axioms are treated as definitions, then it is uncertain whether any entities actually exist which satisfy the definitions. It is impossible for an axiomatic definition to distinguish between isomorphic models, so the "proper" reference of "set" is indeterminate. If the axioms are treated as assumptions, then the truth of the axioms is (of course) unprovable. Consistency, utility, representational power, et. al. may be investigated, but such work bears at best indirectly on truth. We can't have both certainty of reference and certainty of truth, but by no means does mathematics provide an excuse for vagueness. Ken Presting
utility@quiche.cs.mcgill.ca (Ronald BODKIN) (03/17/90)
In article <8533@mentor.cc.purdue.edu> g2g@mentor.cc.purdue.edu (Ranjan Muttiah) writes: >If mathematics be the most precise of sciences and its desiderata of >numbers and sets are left undefined, then how in a more imprecise >field such as AI can we define intelligence ? > >Answer: You don't ! Maybe intelligence is an "axiomatic" concept, but even still we should endeavour to name properties of it (numbers are defined in math, sets have properties). AI is more "imprecise" only in that it is more concerned about its abstractions bearing on reality (not so directly though, as AI is concerned about intelligence in the widest sense). Ron
ian@mva.cs.liv.ac.uk (03/17/90)
In article <8533@mentor.cc.purdue.edu>, g2g@mentor.cc.purdue.edu (Ranjan Muttiah) writes: > > If mathematics be the most precise of sciences and its desiderata of > numbers and sets are left undefined, then how in a more imprecise > field such as AI can we define intelligence ? > > Answer: You don't ! > We can define intelligence any way we like, the question should be whether or not this is a useful task. The ultimate aim of AI is, presumeably, to produce an artificial intelligence. In order to know what we are shooting at, we need to pin our quarry down a little tighter. One way of doing this is to seek a rigorous definition of intelligence, another is to state a task that a machine must be able to do to satify is that it exhibits intelligence. It is for this reason that people seek to define intelligence. Sometimes a rigid definition of a well-known concept is necessary (is a ninety quid chess machine intelligent? To the person in the street, it is!). Of course, most research is not interested in the ultimate aim of AI, but rather strives for some goal in one part of AI. For example, I am interested in explanation systems and do not care whether or not they are intelligent (or even whether they appear to be intelligent). I can however indulge in similar arguments about what constitutes explanation :-). By the way, I thought that numbers and sets were defined by starting from the empty set and building sets from there. Ian ---