notes-cog-ai-reasoning-ambiguousReasoning

todo: move most of the related links from inheritanceNetworks to here

Provenance for contradiction

" E VERY HUMAN harbors mutually inconsistent beliefs: an intelligent person may be committed to the scientific method, and yet have a strong attach- ment to some superstitious or ritual practices. A person may have a strong belief in the sanctity of all human life, yet also believe that capital punishment is sometimes justified....If we observe inconsistencies we do not crash—we chuckle!

Dependency decorations on data that record the justifications for the data give us a powerful tool for organizing computations. Every piece of data (or proce- dure) came from somewhere. Either it entered the computation as a premise that can be labeled with its external provenance, or it was created by combining other data. We can add methods to our primitive operations which, when processing or combining data that is decorated with justifications, can decorate the results with appropriate justifications. For example, the simplest kind of justification is just a set of those premises that contributed to the new data. A procedure such as addi- tion can decorate a sum with a justification that is just the union of the premises of the justifications of the addends that were supplied. Such simple justifications can be carried without more than a constant factor overhead in space, but they can be invaluable in the attribution of credit or blame for outcomes of computations—in knowing what assumptions matter for a particular consequence. By decorating data with dependencies a system can manage and usefully com- pute with multiple, possibly inconsistent world views. ... Dependencies allow a system to separate the potentially contradictory consequences of different assumptions, and make useful progress by exercising controlled in- credulity.

If a contradiction is discovered, a process can determine the particular no-good set of inconsistent premises. The system can then “chuckle”, realizing that no computations supported by any superset of those premises can be believed; compu- tations can proceed in worldviews that do not include the nogood set. This chuck- ling process, dependency-directed backtracking , [Stallman and Sussman, 1977, Lieberherr, 1977, Zabih et al., 1987], can be used to optimize a complex search process, allowing a search to make the best use of its mistakes "

-- http://dspace.mit.edu/handle/1721.1/49525, Chapter 4, page 51 (pdf page 52)

misc ideas:

• conceive of the mind as holding multiple possibly contradictory motivations, and sets of beliefs, and actions/decisions taken by voting across the different "subminds"
• to decide whether to state that you agree with some belief, calculate the similarity between your belief and that one, and say 'i agree' if it's high enough, even if you technically disagree with some 'details' in the query
• words and sentences have a set of potential meanings rather than a single meaningws
• make provisional implications, eg if you are not sure if something is true but you think it is probably true over some threshold, then provisionally consider it true
• leave off qualifiers depending on context. Eg in a context in which the listener assumes that you aren't an expert about what you are talking about, you don't have to preface each statement with "well i don't know much about this so this is just my guess:"