[FoRK] Human behavior is 93% predictable

J. Andrew Rogers andrew at ceruleansystems.com
Wed Mar 3 12:20:46 PST 2010


On Mar 2, 2010, at 4:08 PM, Jebadiah Moore wrote:
> In all these cases, the models get the 90% right but can't deal with the
> last 10%--which is often the really interesting part.  The model has to have
> abstractions somewhere, and the choice of abstraction will determine which
> (proverbial, since the number will vary) 10% is hard, whether the
> abstractions are chosen by a human or the model itself.  In the case of mass
> models of human behavior, the hard 10% is most likely going to be deviant
> behavior, which in the case of surveillance is also the interesting 10%.  Of
> course, the model can adapt, and maybe it can even do so well enough to beat
> the well-informed surveillee.


When we think about prediction, we only think about it in terms of patterns that *we* can discern.  If we can't discern it or conceptualize it, we assume it does not exist.

The incorrect intuition most people have about prediction assumes a very simple, stupid, static probability model of the type that everyone is already well-acquainted with, hence the intuition. Newer methods have a complex distribution of algorithm probabilities over very high-order models.  The distributed prediction error bounds that make up the "last 10%" do not have the implications usually assumed.

This is why you need a computer to defeat it.  Computers will see all manner of exploitable pattern that humans do not and those abstract patterns are very difficult for humans to reason about in the context of the overall model.  Attempting to increase your entropy and predictive error bounds without a computer's help is a fool's errand. Human brains are inadequate for the task.


It is increasingly subtle and transparent precisely because it can systematically operate on patterns at a level of abstraction that human cognitive ability is too limited to discern.  Interesting and spooky at the same time.




More information about the FoRK mailing list