What chess champion Gary Kasparov can teach us
Recently, Gary Kasparov wrote an essay about humans and computers playing chess, under the guise of a book review. Andrew McAfee today published an essay on Kasparov's ideas, with a specific focus on one observation by Kasparov.
Kasparov noted that recent matches have shown that weak human chess players with computers can beat a chess supercomputer, and, in addition, a chess grandmaster with a computer but a weak organization of the human-computer collaboration. In Kasparov's words,
Weak human + machine + better process was superior to a strong computer alone and, more remarkably, superior to a strong human + machine + inferior process.
McAfee starts from this and says that Kasparov may have stumbled upon a better model of business processes. From my point of view, I see Kasparov's insight as one example of the great benefit to be gotten if we can only adapt mechanism design theory to capture the fuzziness of humans and the precision of computers, acting in tandem, better. (I think there are many examples to urge us to change mechanism design towards more human-compatible decision-making models, on which I plan to blog more.)
I am making no grand claim that I know how we can approach this goal. I am simply noting that it seems a very worthy goal, one that I would rather see research in mechanism design aim for. Instead, the current thrust of the mainstream mechanism design research seems to be to get more and more refined mathematical results based on the assumption that the actors in the mechanisms studied, whether human or computer agents, behave with the precision of computers. I am aware of some work that attempts to introduce errors in the decision-making of agents in mechanism design theory, such as work by Kfir Eliaz, but I would certainly love it if more of the very clever mechanism theorists attacked the fuzziness problem head on.
Let us not leave the topic of a better business process to Harvard Business Review articles only. Some Econometrica articles on it, please.
Lecture Notes on Dominant Strategy Implementation
Rather later than I was hoping, I am posting my lecture note on dominant strategy implementation. The notes are for my graduate microeconomics course at Temple University, and the book referenced is my book A Toolbox for Economic Design, co-authored with Karen A. Campbell, Emina I. Cardamone, Scott Deacle, and Lisa A. Delgado. The book covers this topic in chapter 2.
The pundits’ dilemma
Mark Liberman says this in the Language Log today, among other good points:
Overall, the promotion of interesting stories in preference to accurate ones is always in the immediate economic self-interest of the promoter. It's interesting stories, not accurate ones, that pump up ratings for Beck and Limbaugh. But it's also interesting stories that bring readers to The Huffington Post and to Maureen Dowd's column, and it's interesting stories that sell copies of Freakonomics and Super Freakonomics. In this respect, Levitt and Dubner are exactly like Beck and Limbaugh.
We might call this the Pundit's Dilemma — a game, like the Prisoner's Dilemma, in which the player's best move always seems to be to take the low road, and in which the aggregate welfare of the community always seems fated to fall. And this isn't just a game for pundits. Scientists face similar choices every day, in deciding whether to over-sell their results, or for that matter to manufacture results for optimal appeal.
In the end, scientists usually over-interpret only a little, and rarely cheat, because the penalties for being caught are extreme. As a result, in an iterated version of the game, it's generally better to play it fairly straight. Pundits (and regular journalists) also play an iterated version of this game — but empirical observation suggests that the penalties for many forms of bad behavior are too small and uncertain to have much effect. Certainly, the reputational effects of mere sensationalism and exaggeration seem to be negligible.
Mark Thoma says, among other things, this, in the post that brought Liberman's post to my attention:
I'm not sure I know the answer to that, but I suspect it has something to do with increased competition among media companies for eyeballs and ears combined with an agency problem that causes information organizations to maximize something other than the output of credible information (maximizing profit may not be the same as maximizing the output of factual, useful information).
Though this type of behavior was always present in the media, it seems to have gotten much worse with the proliferation of cable channels and other media as information technology developed beyond the old fashioned antennas on roofs receiving analog signals. I don't want to go back to the days where we had an oligopolistic structure for the provision of news (especially on network TV), competitive markets are much better, but there seems to be a divergence between what is optimal for the firm and what is socially optimal due to the agency problem.
Some people have argued that there are big externalities to good and bad reporting, and therefore that "some kind of tax credit scheme for non-entertainment news reporting might enhance societal efficiency and welfare." That might help to change incentives, but I'm not sure it solves the fundamental agency problem. There must be reputation effects that matter to the firm, some way of making the firms pay a cost for bad pundit behavior. But that is up to the public at large, people must reward good behavior and penalize bad, it is not something the government can control. I suppose we could try something like British libel laws to partially address this, but looking at the UK press does not convince me that this solves the problem.
So I don't know what the answer is.
I would not want to jump in and say that I know what the answer is. However, it is clear that there is a mechanism design question here. The economist's knee-jerk reaction to this would be "if the consumers of information are more interested in being entertained than informed, then it is efficient to provide them entertainment as long as the marginal cost of entertaining each one of them meets her/his marginal willingness to pay". As Thoma notes, it is noted that reporting has external effects. These would seem to push us in the direction of amending the rule for social optimality and looking for ways to align pundits' incentives to what efficiency would require.
But if the majority of the audience want to be entertained and not informed, shouldn't we economists, as children of the Enlightenment, bow to the consumers', our multitudinous Kings', desires? To take the idea that bad reporting carries negative externalities seriously, one has to take seriously the possibility that people express preferences for the wrong things, things that will in the long term, collectively conspire to harm them. Is this only because of the word "collectively" and so only a question of externalities, one step removed? I think that there is more "irrationality" to consumers than that. We need to come to grips, as we consider mechanism design, with "irrational consumers". The misnamed "behavioral economics" (all economics is behavioral) field has some valuable ideas here. It seems to me economic theorists of the mechanism-design bent, should adopt these ideas and do their formalizing magic with them to reach some results. After all, no lesser theorist than Leonid Hurwicz made a foray into "irrational" agents all the way back in the 1980s.
Remark: I always place "irrational" and "rational" within quotation marks. Given what I know of game theory, including Binmore's work on the application of Goedel's Theorem on games played by automata, and games such as the Prisoners' Dilemma and the Centipede, I feel I have no way of even pretending that I know what "rational behavior" really ought to mean in the case of individuals interacting in a game. Worse, in the context of consumer not knowing "what's good for them", we have an additional level of "irrationality" which seems to resolve to time inconsistency in the behavior of a single person. This post being long enough, I have to leave further development on my thoughts on these points to another post.
Paul Romer on Elinor Ostrom’s Nobel prize
My old teacher Paul Romer has a fantastic post on Ostrom's prize award. A long quote follows, but I do strongly recommend the whole thing:
Most economists think that they are building cranes that suspend important theoretical structures from a base that is firmly grounded in first principles. In fact, they almost always invoke a skyhook, some unexplained result without which the entire structure collapses. Elinor Ostrom won the Nobel Prize in Economics because she works from the ground up, building a crane that can support the full range of economic behavior.
When I started studying economics in graduate school, the standard operating procedure was to introduce both technology and rules as skyhooks. If we assumed a particular set of rules and technologies, as though they descended from the sky, then we economists could describe what people would do. Sometimes we compared different sets of rules that a “social planner” might impose but we never said anything about how actual rules were adopted. Crucially, we never even bothered to check that people would actually follow the rules we imposed.
...
Economists who have become addicted to skyhooks, who think that they are doing deep theory but are really just assuming their conclusions, find it hard to even understand what it would mean to make the rules that humans follow the object of scientific inquiry. If we fail to explore rules in greater depth, economists will have little to say about the most pressing issues facing humans today – how to improve the quality of bad rules that cause needless waste, harm, and suffering.
Glaeser on the “Nobel” prize in economics
Ed Glaeser has a nice essay explaining the work of the awardees of this year's prize in the New York Times.
Levitt on the “Nobel” prize in economics 2009
Here is Steven Levitt admitting that he should have known about Ostrom. Nice to see an admission that over-narrowness is not very productive in science.
Swedish National Bank’s Prize in memory of Alfred Nobel, 2009
Congratulations to Elinor Ostrom and Oliver Williamson for the honor they received. I wish I had time to discuss this properly right now, but class prep is calling. Meanwhile, here is a post by Alex Tabarrok on Wiliamson and one on Ostrom. I should also mention that my book, A Toolbox for Economic Design, discusses Ostrom's work.
Dan Pink on Motivation
Here is a TED talk about the importance of intrinsic motivation for business. Thoroughly enjoyable, and it gives a good challenge to mechanism design theorists and economists in general. (Via www.economistsdoitwithmodels.com).
Excellent undergraduate micro course
Jeff Ely at Northwestern University has an excellent page with notes on his undergraduate microeconomics course. (The whole blog in which this appears, Cheap Talk, is very interesting, and you can find a link to it in my blogroll on the right.) As the fall semester approaches, I find myself quite drawn to this, as it is a good background reference for the graduate micro class I teach, as well as a nice plan for a microeconomics course that would engage undergraduate students without math overload and without the slavish devotion to markets that too many microeconomics courses exhibit.
How do we reform university teaching?
The procrastination is over. This site is designed and almost all its pages are pathetically blank as of now, with only place-holding text to keep the blank page fear away. But it is time to start one the blog, and I can hope to flesh out the rest of the site, as well as enrich the blog with frequent posts and get a regular blog-posting rhythm, before the Fall semester is upon me.
I thought I would start the blog with a discussion of a provocative suggestion for the reform of university teaching by Philip Greenspun. It has to do with incentives and with a fine desire to improve undergraduate teaching. Greenspun's site, from which the link above comes, has more information on him and the course he teaches at MIT. I want to discuss here only his proposals for reforming teaching, and not his rather unquestioned taking of Clark's book A Farewell to Alms as offering the one and only explanation for economic growth.
Greenspun observes that universities are still operating on the model established in 1088 with the opening of the university of Bologna. Lecturing was an efficient way to reach many people at once, back then. Students had no IM/Facebook/Limewire/Hulu/Twitter to keep them distracted, no video game consoles at home, and the lecture hall was more comfortable than their living quarters. But since then we have seen the arrival of so many technologies that upset the logic of lecturing that we should step back and reconsider our inertia in changing how we teach students.
I skip over the entertaining skewering of the video of some lectures at Yale. Suffice it to say that they show that a big-name professor at a big-name university can use classroom time inefficiently. So let me discuss Greenspun's concrete proposals.
"Stop grading your own students." I may be woefully uninformed here, but I am surprised that it is not a mechanism design theorist who came up with this suggestion first, but Greenspun, an engineer. For many courses taught at universities this makes excellent sense. Professors should grade others' students' exams, anonymously. This eliminates with one stroke the ego-boo bias that makes professors unwilling to give F grades as often as necessary, because their teaching can't be so bad as to allow so many students to come out of their classes unenlightened. At the same time, student anonymity eliminates gender, racial and other biases in grading. I recall Steve Landsburg arguing that, on the basis of the famous Groves incentive schemes, that municipalities should have their populations pay each other's taxes, which is an even more radical proposal than this one.
The biggest problem I can see with this grading-swap approach across universities is that it would encourage a cookie-cutter approach that would over-homogenize the course content. For some fields (many?), this would stifle creativity and innovation. Suppose I want to teach microeconomics by emphasizing game theory and using simulations on the computer to illustrate big points. I would have less time to teach the standard stuff I would be expected to teach (never mind that perhaps we should not be teaching the same stuff as Alfred Marshall did in 1890). Meanwhile, the grader is convinced that microeconomics is completely the study of competitive markets and game theory is a waste of time. Who sets the grading standards? And what about grading in subjects such as music? I can see grading a video of a soprano's senior recital being graded remotely, but that will miss the way it sounded in the room, the way the soprano built (or failed to build) rapport with her audience, and, heaven forbid that the remote grader does not like her interpretation of the songs (maybe the grader never really liked Schubert, either, but she's doing a lot of Schubert songs that she loves).
Another problem would be the stratification necessary to avoid introducing name-brand bias in grading. You would not want a professor at Boondocks College grading MIT students, and vice versa, as it may lead to the MIT students getting overly low grades out of pure spite. But if you try to escape this bias by stratifying universities and allowing, say, only professors from top-20 schools to grade students at top-20 schools, then what do you do about the incentive to make the other school look bad by giving its students bad grades? Perhaps this proposal can best be adopted within an institution, but then staffing levels can severely limit the number of courses in which the grading-swap can be used.
Yet there is meat in the grading-swap and I would argue in favor of its serious consideration in higher education, with whatever refinements will be necessary to address the points I raise here and others that I surely have missed.
"Stop lecturing." OK, the human attention span is not much longer than 20 minutes at a stretch. More interaction among students in the classroom is essential. I cannot find anything to criticize here, except my own teaching practices, which I need to reform. There are good reasons when teaching something heavily technical to lecture at length, but such lecturing can always be interleaved with classroom activities to break monotony and reinforce points.
"Build open offices for students." The idea here is that students study in close proximity to other students so they can get help from them as well as motivation. All of the student work for the day should be accomplished on campus, there should be no homework. Greenspun discusses well the implication that the semester system would need to be abandoned for this idea to succeed. He does not discuss the cost of the idea, however: at many universities space is at a premium and changing classrooms and labs to accommodate this approach would not come cheap. It may well be that this idea is good enough to overcome the cost, but we should have an idea of the latter before we can tell.
"Modest Change: Provide detailed review of all work; grade students on their ability to assist other students." This sounds excellent, but again the cost in time for the faculty would be very high. It contributes to the feeling I get from reading Greenspun's suggestions that they would work best at universities that are devoted to good teaching rather than research. At the least, if a university does not properly compensate professors for this level of feedback (and does not monitor its quality somehow), not much good reviewing will be done. If the game is publish-or-perish, who will spend hours upon hours giving real, in-depth reviews of student work? To be fair, Greenspun anticipates the cost objection and proposes peer review of students and reviews by outside graders.
Greenspun's essay continues with suggestions to improve the teaching of engineering. They seem workable and effective, but they would have less wide applicabiilty than the suggestions I have chosen to discuss here. Nevertheless, Greenspun's essay is well worth a read. I may well be wrong in not seeing some excellent ways to adopt even his engineering-specific suggestions in the teaching of my field, economics, and if I think of such ways, I will post another discussion here. For now, though, this post has grown to gigantic proportions and it is time to call it a day. No, let me call it the first full-blown blog post in this site, instead. I promise many more to come (and, yes, to fill out these embarrassingly empty pages listed in the top navigation bar).