Hacker News new | past | comments | ask | show | jobs | submit login
Who You Are (nytimes.com)
187 points by four on Oct 21, 2011 | hide | past | web | favorite | 34 comments

Kahneman and Tversky are legends in psychology, but they didn't change the fundamental way we view ourselves. The notion of humans as super rational utility maximisers was entirely an economic model, and it was that they attacked. It was well known (for over twenty years before their seminal 1974 science paper) that humans were poor at probability and utility judgments.

What they did do, was force the economic profession to face up (somewhat) to these issues, and their contribution to loss aversion and prospect theory more generally, is a significant advance.

That being said, their dual process models are about as predictive as those of Freud (which is to say, not at all). its currently a really active phase of research, so I suppose I can thank them for making it easier for me to get funding.

They also did not invent priming, though they made heavy use of it. Likewise framing effects were well known before them, going back at least to Asch 1951 study of conformity in judgements of line length.

To summate, Kahneman is an amazing scientist, but this reporter does not appear to know much about what he is talking about.

>humans as super rational utility maximisers

Is not an economic model, it is a straw-man of the economic "model" which states that en masse and over time humans behave rationally where their interests are involved. (I put model above in quotes because even that is really only an aspect of how economics models human behavior.)

As many "critics" of economics have pointed out, and as Surowiecki reported in The Wisdom of Crowds, the mass often make better decisions than an individual because, among other reasons, their errors tend to cancel out.

That is a common theme with David Brooks. I'm not trying to be ad-hominem, just noticing a pattern.

What are the best books available on the subject of predicting human behavior, if you don't mind me asking? Advanced or technical are fine, scientific journals would be over my comprehension, I'm afraid.

One great book on behavioral economics that's based on a lot of research and experimental data is http://www.amazon.com/Judgment-Managerial-Decision-Making-Ba...

Came highly recommended from a Stanford professor and is one of the most "meaty" books on the topic.

Animal Spirits is probably a lighter read (it's still on my to-read list), but I've been impressed when hearing the author, Robert Shiller, speak:


Discussing animal spirits in various videos on youtube: http://www.youtube.com/results?search=Search&resnum=0...

David Brooks has a really terrible track record of horribly misinterpreting social science research and drawing completely unfounded conclusions from it. He's been taken down by academics many times over it, most memorably (for me) here http://languagelog.ldc.upenn.edu/nll/?p=478

"Most of our own thinking is below awareness."

Indeed!. Minsky once said that consciousness is the brain's debug trace.

Have been reading a very interesting book on related subject -

On Being Certain - Robert A. Burton

This article reminded me of my undergrad econ classes. I understand that it was just undergrad and we were learning a basic tool set. But it seriously scares me when I remember how my classmates and I (some of whom are on Wall St) took class after class that drilled supply/demand graphs premised on rational, utility maximizing populations into our heads. Whether the theories in this article are oversimplified or not, they do provide an important counterweight against anyone who thinks that they can reliably predict people's decisionmaking. I hope econ textbooks are evolving to reflect the growing marriage between econ and neuropsych.

I agree with the basic tenants of this article. Yes it was true, that k&t were moving the model into the 'economic sphere'. But I think you cannot overstate the importance of this. Behavioral finance/economics coming back into cog. psych. and cog. neuro. is absolutely earth-shattering. the money dictated the research and now that research is FINALLY being applied back into where it belongs. I can't wait for these ideas, and those of choice designers/specialists/researchers to make it into 'machine learning' and 'weak ai'. If ever there was a subject that was full of shit from the beginning with regards to how people actually think, and needs to be re-architected from the ground up. Also, someone needs to start listening to other Princeton Researchers like Eldar Shafir on these topics as well.

It actually has made it into machine learning. See the works of Gigerenzer on heuristic decision making, where he shows that simple heuristics outperform complex statistical models unless the amount of data is really large. It blew my mind when I saw it, and a good paper to start with is here: http://citeseerx.ist.psu.edu/viewdoc/download?doi=

That being said, almost anything Gigerenzer has written in the last five years is extremely relevant to this topic.

You mean tenets intead of tenants

Anyone else feel that the article was exceedingly lacking in subtance, especially for an NYTimes.com article?

I thought that the article was exceedingly lacking in substance, typically for an nytimes.com article.

NYTimes makes some mistakes and publishes some junk, but they also have great content. It's been uneven. What's better?

I couldn't get my NYTimes this weekend, so being a Bay Area resident I got the SF Chronicle. The writing there is for grade school kids. Not trying to be funny, but it was sad.

(Don't get me started on the Economist...)

It's an editorial column, not a news article.

It definitely makes a lot of sense. Just try speaking with someone about politics or religion. Even if you conclusively prove that the other persons views are not based in fact or reason they will refuse to acknowledge you are right and then usually get mad and stop talking to you. Humans are most certainly not 'rational beings'. Our thinking is constantly biased by our formative experiences and our environment.

Humans are most certainly not 'rational beings'. Our thinking is constantly biased by our formative experiences and our environment.

It goes far deeper than that, too. Our raw pattern matching sensitivity is cranked to the max at a very low level, and this hypersensitivity to perceived order ricochets throughout the entire system of data processing that our brain engages in.

We see patterns everywhere, whether they're real or not, and we have trouble unseeing them even once we know for a fact that the data is random, or that the pattern fails. Statistically speaking, we're a freaking mess, we're constantly pulled towards the wrong answers, we never have good estimates about how reliable our inferences are, it's just an all around bad scene.

And yet the combination of all of these seriously flawed pattern inferences leads to a creature that, all said and done, makes pretty damn useful predictions about a lot of things, even if the details of how those predictions get made are all wrong. This is surprising, since typically in statistics when we use algorithms that are too optimistic or sensitive we end up with pure garbage. If I had to guess, humans end up implementing something like the reverse of a typical boosting algorithm, in that we take a bunch of too-strong pattern recognizing subunits, and then put them together into something that pits them against each other to become more robust against mis-prediction, but I don't have any data to back up that assumption, or any clear idea how it might work - which is, I guess, a perfect example of exactly this kind of mental stupidity that we're so commonly driven by.

"If I had to guess, humans end up implementing something like the reverse of a typical boosting algorithm, in that we take a bunch of too-strong pattern recognizing subunits, and then put them together into something that pits them against each other to become more robust against mis-prediction"

"Ensemble methods" seems to be what you're talking about. ( http://en.wikipedia.org/wiki/Ensemble_learning )

The application of many models put together to produce one signal to accurately predict the future.

I believe the Netflix challenge was won using ensemble methods acting in concert.

"Our final solution (RMSE=0.8712) consists of blending 107 individual results. Since many of these results are close variants, we first describe the main approaches behind them. "

( PDF paper: http://citeseerx.ist.psu.edu/viewdoc/download?doi= )

QIM, a large hedge fund that works futures, also uses the same model.

"In more direct language, Woodriff uses a statistical technique called the ensemble method, which is a way of mining data to produce something akin to the wisdom of crowds. A bundle of computer models, each searching for patterns in different ways, are linked together to produce a consensus statistical prediction—a sort of prediction by algorithmic committee. Scientists use the method to help predict ozone levels, for example. Woodriff uses it to help predict where futures markets are headed over a 24-hour period. His predictions are derived from four basic bits of historical pricing information: the open, close, high and low of specific markets.

Rishi Narang, whose Telesis Capital is a longtime investor in QIM, says other fund managers use similar methods and techniques. "The core idea is not so magical," Narang says. "It is how he puts it together. Getting the program correct is very challenging."

( http://www.absolutereturn-alpha.com/Article/2361672/QIMs-Jaf... )

Boosting, which he mentioned, is an ensemble method so I assume the parent is familiar with them.

Ensemble methods incorporate multiple weak classifiers and work to make them stronger. I think the parent was thinking of the reverse of this, although that idea seems pretty alien to me.

Yes, I'm familiar with ensemble methods, I use them a lot for classification. But those are not really what I'm thinking about (I'm still groping towards concrete ideas here, so forgive me if the following is a bit vague). Perhaps my saying "the reverse of boosting" is not really an accurate way to put this, in retrospect, so let me clarify.

Ensemble methods typically take several distinct (either by method or training) weak learners and combine the predictions to get one strong hybrid by smoothing, averaging, or otherwise combining the results. They are still vulnerable to overtraining, though, and they're not very good at generalizing from small amounts of data because the individual weak learners don't learn from each other or from context.

My theory is that we might be able to get rid of the ensemble and tolerate massive overtraining without detriment if instead of merely combining results, we took a recursive approach and let the classifier use its output as input at another level. My thought is that overtraining on some patterns could be mollified by the ability to recognize error due to overtraining as a pattern at a different depth of recursion.

This obviously would not be generally applicable to weak learners, it would only apply to a particular subset of learners, and that's where my thoughts get a lot muddier and speculative.

My really wild speculation: in the limit, if you set something like this up in the right way, you might be able to come up with an efficient approximation to Solomonoff induction as restricted to the subset of patterns that you're actually exposed to, rather than over the entire set of possible inputs. If I'm correct about that, it would enable staggeringly effective learning within a domain, as long as the domain itself displayed patterns that had some sort of underlying order.

But I don't have any codez to show, or really anything more than a hunch at this point, so don't take me too seriously. :)

Indeed. The closest I can think of to what he is saying is pareto coevolution

Makes me wonder if the singularity is going to be more about how AI handles random and non-random information than anything else.

>Our thinking is constantly biased by our formative experiences and our environment.

And current mood. I have no numbers on this but i am pretty sure that people with chronic pain tend to have more negative thoughts than the average person.

edit: space

There is also an article consisting of an excerpt from the Kahneman and Tversky book here:


(Empty) HN discussion of it: http://news.ycombinator.com/item?id=3141022

"We are players in a game we don’t understand."

We've had the opportunity to understand the game ever since Darwin published the Origin of Species. Yet, even Darwin himself struggled with the ramifications of what we truly are (and aren't) after his mind numbing discovery. The truth is far too devastating for the majority, and it is this fact that divides us. A brain can only be of three dispositions: one that understands reality, one that refuses to, and one that doesn't. A subset of the last is a brain which simplifies a complex, poorly understood reality into one that is far easier to grasp. This last one is where the majority find comfort.

Human motivation is frighteningly simple if looked at objectively, and it is this truth that we hide from ourselves at all costs to preserve our sanity.

I must admit that I didnt quite get what you said here :( Would you care to elaborate?

Yet another entry in the long list of pop psychology books. It seems like they all gear up on one or two navel-gazing insights that just about anybody can intuitively identify with (You have a slow, rational side and a fast, emotional side! Doesn't that explain everything?) and then, they try to run as far as they can with the implications of this overly dumbed-down hypothesis. Carefully cherry-picked statistics from the millions of social phenomena and psychological experiments taking place around the world are sprinkled into the narrative to keep you engaged. (side rant: all of which have their methodological details conveniently obscured to prevent your critical thinking from kicking in, and you are extremely lucky if the sample size is provided, much less any attempt at a p-value or other discussion of statistical significance. Nope, it's usually just "Amazingly enough, 89% of ...")

Example: the silly birdie vs. bogie data presented in this little article. Great, people want birdies more than they don't want bogies, and perhaps it ties back into some aspect of your central hypothesis. But how many other oversimplified statements about human nature could I "prove" with this example? Probably hundreds. Maybe it's a completely rational strategy on the part of the golfer, since their experience has taught them that the (emotional|physical|mental) effort required to sink a birdie putt is not as productive in the long-term as at least making par on every hole. That kind of alternative thinking doesn't matter though, so we simply move to the next experiment and supportive conclusion. Repeat ad infinitum, until we've fulfilled the length requirement for a novel.

No, I did not enjoy Freakonomics (can you tell?).

I agree with you but wonder if the article was like that from Daniel Kahneman himself or the journalist. Having dealt with the later I really understand how good at bending what you said they are.

Also, if Daniel was a pop scientist, would he be given nobel prize in his field? Yes, Obama comes to mind, but still.

I wouldn't try to make any comment on the overall value of his research, and I am sure his academic writing is much more rigorous. But I really think these "universal secrets of the mind" books, written for a general audience, go too far in clouding original and critical thought by presenting such a slick, skewed narrative. They're sold as guides to better thinking but wind up inducing the opposite. It feels dishonest.

This, and what karolist said, is terribly unfair if you haven't actually read Kahneman and Tversky's Prospect Theory, which is not a book but a paper. It's not very long; you should read it.

I think you are judging their work solely on Brooks' description of it.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact