addressalign-toparrow-leftarrow-rightbackbellblockcalendarcameraccwcheckchevron-downchevron-leftchevron-rightchevron-small-downchevron-small-leftchevron-small-rightchevron-small-upchevron-upcircle-with-checkcircle-with-crosscircle-with-pluscontroller-playcrossdots-three-verticaleditemptyheartexporteye-with-lineeyefacebookfolderfullheartglobe--smallglobegmailgooglegroupshelp-with-circleimageimagesinstagramFill 1launch-new-window--smalllight-bulblinklocation-pinm-swarmSearchmailmessagesminusmoremuplabelShape 3 + Rectangle 1ShapeoutlookpersonJoin Group on CardStartprice-ribbonprintShapeShapeShapeShapeImported LayersImported LayersImported Layersshieldstartickettrashtriangle-downtriangle-uptwitteruserwarningyahoo

Cognitive Science Reading & Discussion Group Message Board Cognitive Science Reading & Discussion Group Discussion Forum › Interesting thoughts of Peter Norvig

Interesting thoughts of Peter Norvig

Dana R.
user 2673220
Group Organizer
Berkeley, CA
Post #: 53
Michael Schwarz sent a link to an article On Chomsky and the Two Cultures of Statistical Learning. It's very interesting. After reading it, I can't help feeling that both Norvig and Chomsky take views that are too extreme. On the one hand, it is ridiculous to not look at empirical evidence, but on the other, understanding the why and why not is one of the main reasons to collect that empirical evidence in the first place. Causal inference seems to be what is missing from Norvig's approach, and empirical data seems to be what is missing from Chomsky's. Collecting huge amounts of data on the probabilities of word combinations is certainly interesting and useful, but having some sort of causal model of why words are used in certain contexts is precisely what we should be trying to figure out with those huge amounts of data. You can use statistics to infer causation, figure out the why. While brute statistical probabilities of word combinations alone might be useful for many things, inferring the intentions of the speaker/writer are likely to help imbue meaning/grounding to the symbols. Why did the communicator say what they did? Did your interpretation jive with those intentions? Also, understanding the intentions of the person being spoken to plays a causal role in the word choices of the communicator. If Google really wants their search results to be spot on, it would probably help if it could ask the user questions about their motives for inputting the search terms that they did.
Joshua A.
jabiv
San Francisco, CA
Post #: 28
Great synopsis, Dana. I look forward to reading Two Cultures. You raise many of the questions I've had for years. I liken this dichotomy to the example of GPS: Newtonian mechanics is sufficient to pinpoint a location to a certain level of precision, say 1 square kilometer for arguments sake. Statistic on how many people enter each square meter within that square kilometer can be useful in saying whether there is *likely* to be someone in a given square meter. But in the moment, it does not tell if there is actually someone in a given square meter. This requires relativity theory, which is regularly employed by the GPS we regularly use, not statistics.

This does not mean that statistical correlations aren't useful in rendering smooth trends of sufficiently large populations. But the limits of this tool must be tempered by a better understanding of the causal non-linear internal intent of the single agent.
Powered by mvnForum

People in this
Meetup are also in:

Sign up

Meetup members, Log in

By clicking "Sign up" or "Sign up using Facebook", you confirm that you accept our Terms of Service & Privacy Policy