[Probcogsci] Next Week : Optimal Predictions in Everyday Cognition
Vicente Malave
vicente.malave at gmail.com
Thu Jul 2 22:25:44 PDT 2009
Next week we will be meeting (still 4pm, natcomp conference room), and
I will lead discussion.
Two weeks ago, Ben asked "Where do the priors come from"? -The best
way to answer this is to look at a situation where we can determine
the prior, and ask if people's decisions are consistent with that. The
main paper we will be focusing on is "Optimal Predictions in Everyday
Cognition" by Griffiths and Tenenbaum.
http://cocosci.berkeley.edu/tom/papers/predictions.pdf
The task is to predict how long an event will last, given 1 data point
(the current duration). In this case for most of the events (i.e.
baking a cake) they were able to get a good prior. Its a very short
and simple paper, and we will work through all the math in the
appendix.
This paper was inspired by
Implications of the Copernican principle for our future prospects, by
JR Gott III, published in Nature as a "Hypothesis"
http://www.nature.com/nature/journal/v363/n6427/abs/363315a0.html
which estimates a number of interesting things using a the same
argument, including how long the journal Nature will exist, the
possibilities of colonizing the galaxy, and if SETI will find
extraterrestrial life. Its pretty interesting, but quite a tangent,
maybe we won't discuss this.
Finally, when gathering these I found a recent paper by Mike Mozer,
Hal Pashler, and Hadjar Homaei (Cogsci Journal) directly contradicting
the optimal predictions paper. Their argument seems to be that
Griffiths and Tenenbaum have fit the aggregate prediction across
subjects. If you find yourself screaming out against the target
article, perhaps your criticisms will be contained here.
http://www.cs.colorado.edu/~mozer/papers/reprints/MozerPashlerHomaei2008.pdf
So, for next week, read the "Optimal Predictions" paper, and come
prepared to discuss. Other readings are supplementary.
--
Vicente Malave
More information about the Probcogsci
mailing list