Blargh Blog

Wednesday, April 06, 2005

Blink II

When I summarized Malcolm Gladwell's new hit book, Blink, back in mid-January, I promised to add some of my own thoughts on the strengths and limitations of the book and what we can take away from it. But I didn't say when. So you're getting my thoughts now.

Unless you've either read my earlier post, the book itself, or some other detailed summary of it, then large parts of this post probably won't make much sense.

Also, if you want to read more about Blink, either before or after (or, I suppose, instead of) reading my posts, then your options include the Brooks and Posner pieces and the Gladwell-Surowiecki discussion that I mentioned last time, plus these links:

- Gladwell's website, with excerpts from Blink
- Gladwell talks sports in a January interview with Jeff Merron of ESPN
- David Adesnik reviews Blink at Oxblog (in progress)


How The Mind Works

I enjoyed reading Blink, and I think that most people would too, but there are lots of enjoyable books out there. Why read Blink?

One thing that we can take away from Blink is a better understanding of how the mind works. Much of what it does is "below the surface." Many of our judgments are fast, intuitive, and associative, and we can't know how it is that we make them. We are often unable to tell directly when our judgments are really responding properly to the relevant facts and when they are not serving our purposes. People are good at coming up with reasons to explain what they think and do, but, even though we think that we are just explaining ourselves through introspection, the stories that we create about ourselves do not always correspond to our actual mental processes.

Now, you can get this lesson without reading a couple hundred pages. You could just stop by Mixing Memory or look at pretty much any other piece of writing on cognitive psychology. One advantage of Blink, though, is that its richness in examples can make this lesson start to seem intuitive to you, rather than being merely an interesting and perhaps counterintuitive fact that you remember. Since applying this lesson to your life is not a rigorous deductive process (a claim I'll defend below), this intuitive understanding of it might help you make this lesson part of your life, rather than just something that you have learned. (Of course, I have no actual evidence for Blink helping to tune people's intuitions, and there is no sense in trying to generalize from my own case, in part because I had probably already studied more cognitive science than the typical reader.)


Yay Intuitions!

Is there anything more substantial and less vague that we can take away from Blink? One thing that you can take away from the book (or the subtitle, "The Power of Thinking Without Thinking"), is that Gladwell is very positive on intuitive thinking. He thinks that people are often too deliberative. You see the same thing in his ESPN interview, where he says "I think that the worst thing about the Super Bowl is the two-week layoff. I think teams get over-coached in the second week." (Gladwell is perhaps overly positive on snap judgments in this case - methinks his intuitions are mis-tuned.)

So what's the evidence in favor of smart intuitions? One problem with judging this claim from the examples that Gladwell selects, and with the generalization-from-a-few-examples style of argumentation in general, is that it can be highly sensitive to which examples are selected. But let's completely ignore that problem.

Looking at Gladwell's examples, we get a very mixed story. Some cases (like with the kouros and the double faults) involve experts using their intuitions well, some involve experts succeeding with a blend of intuition and deliberation (like the Millennium Challenge), and some involve expert judgments that fail (like the doctors' judgments about heart attacks). Some judgment failures are due to a readily identifiable bias (like the judgments of female trombonists or tall CEOs), while others (like the heart attack judgments) are not. In some cases a computerized algorithm proved superior to human judgment (like the heart attack judgments and the marriage-stability judgments). In some cases ordinary people were able to make good intuitive judgments from a little bit of information (like college students' judgments about conscientiousness, emotional stability, and openness to experience), but in many cases ordinary people's intuitions were overwhelmed by irrelevant information (like when Gladwell tried to judge relationship problems).

So we can't draw any simple conclusions from these examples - nothing like "intuitions good, deliberation bad." One rough generalization that we can make is that expertise usually helps. Experts tend to have better intuitions than ordinary people. (This often isn't clear in Gladwell's book, but it's very clear in his ESPN interview: "The point of thin-slicing -- the art of making accurate predictions from very "thin-slices" of experience, is that it's something that only experts can do.") Good intuitions are not just natural instincts that automatically get things right. They are built up and improved over time.

This tells us something about learning: it's not just a matter of gaining explicit knowledge that you can give a verbal account of, it also involves training your intuitions. (Some people already knew this fact about learning, but Blink may help you understand it intuitively.)


Expertise

What is expertise? Do good intuitions simply come from spending lots of time with the subject matter, or is it necessary to have practice making similar decisions? Here again, we return to the "on the one hand, on the other hand" analysis. In some cases people make good judgments without having practice with similar decisions (like with Braden predicting double faults), but in some cases people with only familiarity do not make better judgments than naive people (like college students who assess their friends' personalities). Indeed, familiarity with a person actually seemed to hurt students' personality-assessment abilities (with respect to conscientiousness, emotional stability, and openness to experience). Further, in some cases, even people who do have practice making similar judgments do not seem to develop good intuitions (as with diagnosing heart attacks).

One risk of familiarity is that people develop theories which may be wrong. In the cases of doctors diagnosing heart attacks, doctors had a theory about what factors predicted heart attacks. In their theory, lifestyle factors (like smoking) were an important predictor of who was having a heart attack at that moment. They were wrong, and in all of their experience diagnosing heart attacks their judgments never became calibrated to reality. Apparently theories can make you resilient to the kind of learning that trains your intuitions.

The orchestra experts who thought that female trombonists weren't as good as the men were also under the sway of an incorrect theory, and they were influenced in a particularly subtle way. When they knew the sex of the trombonist, males actually sounded better to them. A more general bias against women probably also contributed to their judgment errors. Bias can also exist with a supporting theory. People seem biased to favor tall people over short people, as evidenced by the $789/inch salary advantage that height brings you, even though there does not seem to be much of an accompanying theory to that bias.

Familiarity and practice making judgments do not seem to be sufficient for overcoming biases and incorrect theories. It is possible that people who get a more direct feedback on the quality of their decisions will be able to calibrate better. However, the feedback that orchestra experts got from actually listening to the trombonists was not sufficient (because that experience itself was apparently corrupted), and the feedback that doctors got from their experience with people who they diagnosed as having a heart attack or not was not sufficient (perhaps because it wasn't direct enough).


Information

One of Gladwell's more useful ideas is the way he uses "information." Central to good decision-making are 1) identifying the most relevant pieces of information and 2) not being overwhelmed or led astray by irrelevant pieces of information. This insight leads directly to two ways of improving your decision-making:

- attend to the useful information and ignore the rest (like with the heart attack algorithm)

- don't even let yourself know about the irrelevant information (like the auditions behind a screen)

Although the first of these strategies seems obvious and the second often seems unnecessary, keeping them in mind and trying to apply them can help you make better decisions. There is a temptation to think that more information is always better, or at least that it cannot hurt, but you have to break free from that. You are not immune to being thrown off by irrelevant information, and it can help to remind yourself to focus on the most important pieces of information.

The biggest problem with these strategies is that it's not easy to distinguish the relevant information from the irrelevant information. As we saw from the doctors' theories about heart attacks, even years of experience doesn't necessary give you this ability.

Learning successfully does involve gaining this ability to identify what information is relevant and what is not. It is also important to come up with a way of organizing the relevant information so that you can use it. These kinds of learning can be purely intuitive and inexplicable (like Braden's intuition for double faults), or they can be catalyzed and accelerated by a (correct) theory. People like John Gottman (who can predict impending marital problems) and professional food tasters (who can distinguish jams on very specific dimensions) learn with the aid of a theory, and they develop both well-calibrated intuitions and an ability to explain these intuitions.


The Robots are Coming

Two of Gladwell's most remarkable examples, the heart attack algorithm and the divorce prediction, show the power of computers. If it's possible to come up with explicit criteria for when a judgment is correct , and if it's possible to come up with a thorough way of explicitly identifying most of the potentially relevant data that might be used in a judgment, then there's a good chance that a computer can come up with an algorithm that's better than any human at making predictions. Medical data and the outcome of heart attack or no heart attack were sufficient for the heart attack algorithm. For Gottman's relationship research, the outcome is obvious (divorce), but the data are every statement and facial expression in a conversation, and they have to be recorded on camera, observed by people, and coded into the computer. Computers are still at work (as far as I know) trying to break down the biomechanical data of a player's serving motion to match Braden's ability to predict double faults.

So why didn't Gladwell call his book Algorithm: The Power of Letting Computers do your Thinking for you ? Sure, it might not have sold as well, and he would have had to shift the balance of his examples, but don't computers really come out the winner in Blink, ahead of both intuitions and deliberation?

One answer is "yes." There is a lot of neglected work out there on purely statistical methods of decision making, and the consistent finding is that algorithms outperform human experts. Computers have an under-utilized ability to figure out what information is relevant and to organize it to make good predictions. I'll bet that there is an interesting book waiting to be written, not just about Deep Blue and Billy Beane, but about all of the improvements that we could make in important domains like healthcare (and perhaps the penal system), if we just trusted the predictive powers of algorithms that had proven their predictive ability.

A second answer is "yes, but..." Algorithms are good at picking out what data to use and how to use it, but in many cases we don't want them making decisions for us in real life. Their use is more as a learning tool, to help us come up with the right theory so that we can train our intuitions, explain our intuitions, and make better decisions. The doctors at Cook County hospital now use a simple heart attack algorithm based on the most important factors identified by a more complicated statistical analysis (although they probably could just put their patients' data in a computer and let it crank out a decision). Gottman developed an algorithm based on rigorously coded data, and he's used it to train himself to identify relationship problems by picking out the most important signs of trouble, like facial expressions that indicate contempt. Since he doesn't have to rely on all of the cameras and carefully coded data, his talent could be very useful in real life. Further, even in cases where computers can outpredict humans, there are ethical issues to consider. Computers that are good at predicting have limited usefulness.

A third answer is "no." Computers are a useful tool in some cases, but a lot of situations, especially important real-life situations, are too ill-defined, complicated, time-pressured, and unique for computer decision-making to supersede human decision-making. In the Millennium Challenge, for instance, the high-tech computer information system proved too unwieldy for use in the field, and so the intuitive expert Van Riper managed to hold his own against the amazing power of the United States military. Intuition also triumphed over careful scientific analysis in identifying the kouros as a forgery. In another situation that Gladwell describes, a firefighter had a feeling that something was wrong while in the midst of a blaze that wasn't responding to the water. He led his men out of the building, just moments before the floor that they had been standing on collapsed. These sorts of acts of intuitive virtuoso weren't a result of training by computers and they couldn't be performed by computers.

So do computers really come out the winner, ahead of both intuitions and deliberation? The best short answer I can give you is "sort of." I prefer the three paragraph answer that I just gave you, though.


Concluding Remarks

You can get a lot out of Blink, especially if you try to question it and extend it. If you've made it through both of my long blog posts, though, does that mean that diminishing marginal returns will be setting in if you go on to read the book? Or, does it mean that you'll be better prepared to process what you're reading? If you do read Blink after reading my summary and reactions, maybe you could let me know.


Related:
Blink (previous)
Algorithm (next)

1 Comments:

At November 29, 2007 5:17 AM, Blogger Jeff Ginger (FBP) said...

Just wanted to say this post was helpful to me in searching for a few reminders related to the book, so thanks!

 

Post a Comment

<< Home