found drama

get oblique

not a surprise

by Rob Friesel

Via Boing^2: “Surprise! Computer scientists model the exclamation point” – – /sigh again?

As A will no doubt attest, this is nothing new. There have been biological and mathematical models of “surprisingness” for quite a while. (Decades, if I’m not mistaken.) Does science have to take place at UC Berkley USC to be important or something? Or is it just news “all over again” because it finally made it over there from Cornell and Princeton and Brown etc.?

And (again, as I’m sure A will attest) the thing that causes the most /shudder here is the eye movement stuff. “Eye movements” (also called “saccades”) are all the raging vogue among neuroscientists right now. Which I think is funny because that fails to challenge the assumption that eye movements are (in and of themselves) the most salient piece of biological feedback to the environment.

UPDATE: After a brief email exchange w/ David Pescovitz @ BoingBoing, I took his advice and fired off the following to the USC contact listed in at the beginning of the article:

Subject: re: “Surprise!”
Date: November 29, 2005 9:39:00 PM EST
To: mankin {at} usc {period} edu

Eric–

re: http://www.eurekalert.org/pub_releases/2005-11/uosc-scs112805.php

These are very interesting findings mentioned in the above press release. I was curious where I might find out more about the research that they’ve cited. Specifically, I’m wondering in what ways their work has paralleled or considered that of Rescorla & Wagner (1972) and Pearce & Hall (1980). Their assertion that surprise “lacked a formal definition” seems to fly in the face of over 30 years of learning theory.

Let’s see where this goes…

About Rob Friesel

Software engineer by day, science fiction writer by night. Author of The PhantomJS Cookbook and a short story in Please Do Not Remove. View all posts by Rob Friesel →

2 Responses to not a surprise

Amy says:

Yes, the science of surprise is decades old, but most people are too ignorant to admit that learning theory can add anything to these discussions—see below for what I know about surprise.

The Rescorla-Wagner model (1972) can be expressed formally as:
ΔV = αβ (λ-ΣV)
where α = salience of the CS and is a fixed value, β = salience of the US, λ = processing of the US when it’s unpredicted (i.e. processing on the first trial), and ΣV = the sum of the associative strength of all CSs on a given trial.

And of course, Pearce-Hall (1980): αn= │λn-1 – ΣVn-1│
The associative strength of the CS changes according to the following rule:
ΔV = αSλ
where α reflects CS associability, the term αS reflects processing directed to the CS (S= how the salience, or physical properties, of the CS affect learning), and λ reflects processing of the US.

Leave a Reply

Your email address will not be published. Required fields are marked *

*

*