The Development of Computer Intelligence
The knowledge valuable to the development of computer intelligence
includes topics, decisions, qualities, strategies, and methods.
This knowledge comes in two parts,
which are existing knowledge and questions that need answers.
The answers to these questions will allow intelligence computer programs
to start to function reasonably effectively.
- algorithmic compression
- a strategy for algorithmic compression
that uses the set of operations in a model of computation
- algorithmic optimization
- high compression on small subsets
- definitions subsume axioms
- reverse refutation
- open-ended questions and broad mandates
- theorem generation and discovery
- simplicity for theorem inclusion
- languages natural to computers
- intelligence needs all languages
- the definition of intelligence
- ultimate goals, bridging goals, and scaffolding
- unplannability and unpredictability
- the optimal rate of self-improvement
- goal searching with algorithmic compression
- searches with
low branching factors, strong guidance, and strong pruning
- multiple narrow specific incomplete heuristics
- multiple-level higher-order searches for heuristics
- explanations through higher order logic
- searches that use meaning
- deductions that use strategies
- the generation of multiple goal functions
that are dynamic and hierarchically dependent
- Which search spaces, sets of transformation operators,
and languages should the program use?
- What parts of the program
need to be deterministic, random, algorithmic, recursive, or finite
instead of being undecidable or non-terminating?
- Where should the search process start?
Should it start from the ultimately high or ultimately low goal?
- How should the search process start?
Should it start with recursion
and work back to the simplest case that it cannot decompose?
- What type of information does the search process need to record?
- What is the relationship between
the performance data that the search process records
and the higher order goals?
- How should the program represent strategies?
- What is the relationship
between the local and higher order searches?
- How should the program
distribute effort between the local and nonlocal searches?
- How does the success of local searches
depend on their set of transformation operators and their strategies?
Is it random, linear, exponential,
or a dependency that is similar to similar searches?
- When should local searches terminate?
- To what extent does the search process
need to satisfy the higher order goals?
Copyright © 2004, Carl Sommerfeldt.
The author grants everyone permission to
copy and distribute verbatim copies of this document.