October 25th, 2008


Minds and Datamines

Shower thought: the principles behind database normalisation and those of "deep processing" in human memory are curiously analogous - one could make a strong case that the less normalised a database is, the better the kind of representation it is for being powerfully/usefully remembered by a person. There's an invertedness in that "direction of goodness" though - databases that are far from normal are harder to work with - redundancy becomes a management problem in a database because updates need to think about multiple locations, constraints can only be used in the context of transactions when redundancy is involved, and beyond the runtime concerns, the person writing the query has to think harder to avoid problems (an aside: for those who haven't read much about database philosophy, the 「closed world assumption」 versus 「open world assumption」 is an interesting topic). This isn't so much of a problem for most types of memory in the brain, although if we see the brain as powerfully-non-database-normal, how do updates happen? Perhaps they don't - if database people want to come to some understanding of human memory through this metaphor, they could say that most types are database tables without primary keys, the strongest memories are the least normal (extending normalisation a bit so it could apply to tuples), and imagine a very bizarre query (query optimisation is an interesting analogy-node) manager.