(fetzer)lnai

Logic for the New AI
by Bruce MacLennan

I. Introduction
A. The New AI, p 163
In general, much of the lack of progress in contemporary pattern recognition
can be attributed to the attempt to classify by context-free features.

p 164
Heidegger has shown that much of human behavior exhibits a ready-to-hand
understanding of our world that is not easily expressed in propositional form.
"When we use a piece of equipment, we actualize a bodily skill (which cannot be
represented in the mind) in the context of a socially organized nexus of
equipment, purposes and human roles (which cannot be represented as a set of
facts).[p 99]" [...] Our use of this knowledge in unconscious in that we do not
think in terms of propositional rules that are then applied to the situation
at-hand.

p 166
With our present techniques, for the computer to know *how*, it is necessary to
know *that*. ['immediate' vs 'mediate', 'virtual' vs 'switch ... case']

Can there be such a thing as a non-verbal logic?

B. Why a New Logic?
1. The three roles of logic
- Empirical logic: descriptive laws
- Mathematical logic: formal laws
- Normative logic: prescriptive laws.

2. Conventional logic inadequate for the new AI
3. Potential value of a new logic
II. Requirements for a New Logic
A. Indefinite Classification, p 169
- the phenomena of nature are continuous
- perception is subject to noise and error

B. Context Sensitivity
C. Logical Holism
D. Intentionality
E. Mixed-Mode Reasoning, p 173
[V]erbalization permits longer chains of inference by preventing the
accumulation of error.

III. Preliminary Development of New Logic
A. Approach
B. Discrete Time
1. Definitions
2. State transition function
3. Activation of a single trace, p 179
[Oops: why can we take 'gamma_1 near 1' and 'gamma_i inferior to epsilon'?]

4. Activation of a cluster of traces
5. Family resemblances, p 183
[Error: either rho or 1/rho larger than 1; thus result nonsensical...]