Monday, April 28, 2014

The outsized influence of Harvard computer science

Please forgive a bit of bragging about the program of which I have been a part for fifty years, forty of them on the faculty. I started at Harvard as a freshman in the fall of 1964 and became Assistant Professor of Computer Science in the fall of 1974.

Until very recently, the process of building up computer science has been painfully slow. Around 1978, while I was till junior faculty and was more reserved than I now am, I raised with my colleagues that idea that we might have an undergraduate major in computer science. "Computer Science?" one of the engineering faculty snorted indignantly. "Why would we have a major in computer science? We've never had a major in automotive science!" (Actually, we call these "concentrations" rather than "majors," just to be different.)

I did not press the idea again for a few years, and the first undergraduate CS degrees were awarded in 1982 or 1983.

Those days of indifferent commitment are a thing of the past, happily. Now half the college takes our introductory CS course and next month we'll be awarding about 100 CS degrees to graduating seniors (out of a class of about 1670).

Of course, even though the faculty size never broke into the twenties until the past few years, Harvard  College has been producing lots of important computer scientists for decades. Some of them, like Butler Lampson, had been math majors. Some of them, such as Bill Gates and Mark Zuckerberg, didn't bother sticking around to collect their degrees. Some, like Marvin Minsky, are so associated with other great institutions that most people forget they were ever at Harvard. But we had many of them when they were young.

I knew that, and yet I was stunned to see the results of a study of top-50 computer science departments. Among the questions asked was, where had the faculty been undergraduates?

More top-50-CS-department faculty had MIT undergraduate degrees than degrees from any other institution, by a large margin. But Harvard came in second with 67, more than twice the number who had been Stanford undergraduates. If you divide these numbers by number of departmental faculty, another table listed on the site, the comparison becomes even more dramatic (Harvard is near the bottom of that list, and our number is higher now than it has ever been in the past).

Probably Harvard's number of undergraduates-turned-faculty-in-top-50-CS-departments should be counted as 68, since Sheila Greibach graduated from Radcliffe College in the days when it still awarded its own degrees, even though all instruction was done coeducationally in the regular Harvard classes. On the other hand, the data are a bit rough--a fair number of faculty on the roster don't have any undergraduate degree listed. (It's a Google Doc. You can't sort it in place, but you can make a copy and sort it by any column.)

Details. Whatever the exact counts may be, the data scream to me the importance of recruiting top talent at the undergraduate level, people who know how to take advantage of opportunities and how to look into the future rather than simply learning what they are taught.

The dramatic reversal when you ask instead where faculty earned their PhDs--there are more than three times as many Stanford PhDs as Harvard PhDs on the roster-- is a dramatic indicator of the years of lost opportunity at Harvard while Stanford CS was flourishing.

Those days are over. We have a strong faculty group now, we are still growing, and we're planning our move to a new campus in Allston. And the students are better than ever.

Thursday, April 24, 2014

Book Club online chat

Sorry for the late notice, but at 10:20am today (April 24) Ellen Condliffe Lagemann and I will be chatting about our book "What is College For?" on the site of the Chronicle of Higher Ed Book Club.

Added later: I understand that the video will be posted later on that site.

Friday, April 11, 2014

One way to handle cheating

They Yankees beat the Red Sox last night. They cheated. The Red Sox TV crew noticed that the Yankees pitcher, Michael Pineda, had something on the heel of his hand. Pine tar probably. Sweat and dirt, he says. Dunno, says his manager.

In the fifth inning, it disappeared. Guess it must have gotten enough cooler as the evening wore on that he stopped sweating. Still, Mr. Pineda, I'd have your endocrine system checked out. Sweating like that isn't normal.

This has turned into a great moral debate in the media. My esteemed fellow blogger Richard Bradley, who has on other occasions taken cheating allegations very seriously, has a no-big-deal attitude this time.

He's right for a change.

Whether Pineda was disqualified or not would not have changed the outcome of the game. The Red Sox are not hitting.

Some Red Sox players even suggested they were glad Pineda was cheating, because that meant he would be less likely to lose control of one of his fast balls and bean somebody. Pretty generous, given that it also meant that his slider was more effective.

Like a lot of cheating, this is something for management to sort out. Farrell did not want to show up Girardi, not so much because "everybody does it" (they don't, to that degree), but because he knew that Girardi was on notice to fix the egregious cheating, and there was no reason to embarrass him or Pineda, or to cost Pineda a fine in the short run or a scarred reputation in the long run.

I'd say Farrell acted like a grown up, everybody on both teams learned a lesson, and nobody got hurt. Good.



Saturday, April 5, 2014

Honor, Freedom, and Honors

The Faculty of Arts and Sciences last Tuesday discussed the proposal to adopt an honor code. The item was formally moved, so it could be voted at the next meeting.

Several comments were offered to the effect that there would be something wrong with asking students to write the formulaic honor statement repeatedly. No decision has been taken as to how frequently the affirmation would have to be made, and that seems to me a problem with the legislation. As the Crimson accurately quotes me as saying, there is a big difference between affirming one's commitment to academic integrity once at matriculation, or once a year, and reciting the commitment like a ritual prayer on each item of submitted academic work. (CS 20 students have to answer "check in questions" on the reading the night before every class. There are just three or four questions and they are multiple choice, click-click-click and you are done. Would they be expected to type out the honor pledge in addition?)

I observed that it seemed to me unlikely that a defiant soul like Emerson would have ritually and uncomplainingly written out the honor pledge for four years. Wouldn't he and some of Harvard's other eminent nonconformists at some point have refused? What sanction would Harvard want to have visited on such young cranks, destined for greatness, for the perfectly logical sin of refusing to affirm their own honesty? I am trying to imagine Harvard's great logicians, Quine and Putnam and Sacks et al., explaining to their logic classes that yes, "I am being honest" is a self-refential, semantically challenged proposition, but it didn't matter, they had to keep saying something like that anyway, because the Faculty had voted to require it of them.

Part of the rationale for the code is that we have a cheating problem, and schools with honor codes have less cheating. A literature review was offered in support of the latter thesis, but upon reading it, I am not convinced that it actually supports any such conclusion. (Alas, the literature review, like the draft honor code and implementing legislation, is confidential by the protocol of Faculty meetings, not to be shared with those who are not members of the Faculty of Arts and Sciences.)

Even if it were true that reciting the pledge made people less likely to cheat, I wouldn't favor it. "Why not, if it works?" Because there are values more important than effectiveness. It might also work to lower the room temperature or to paint the walls pink, but academic integrity is not a matter of behavioral manipulation by tried and true methods of psychological trickery. If it were, we'd just hire an ad agency. And getting people to be honest at Harvard is not the objective anyway; we want to graduate honest people. When Harvard students become corporate executives or political leaders or professionals, there will be no one there to make them recite the pledge that reminded them to be honest while they were students. Integrity has to come from some deeper place in their beings.

A requirement that students affirm the honor pledge, repeatedly, feels to me like an infringement of students' right to free speech, which includes not only the right to say what they wish but also not to say what they don't wish.  Rather than training students to recite formulas, we should be educating them toward exactly the opposite frame of mind, to resist and challenge attempts to require them to say things.

At the end of the day, I think the honor pledge is an attempt to find the solution to a problem under the lamppost that is easiest illuminated.

Students cheat for a variety of reasons. But one important reason students cheat, especially in catastrophic cases like Harvard's Gov 1310 or the Dartmouth CS course a few years ago, is that they feel they are being cheated. Students tend to work hard when they think the course is making an effort to teach well and to make fair demands on students; they cut corners when they think the demands on them are unfair, or they reason that lazy professors should expect little work in return. This is a version of Harvard sociologist Chris Winship's Low-Low Contract between faculty and students: "Faculty pretend to teach, students pretend to study, and as long as parents and others paying the bills are oblivious, everyone is happy." A major cheating scandal disrupts the oblivious compromise, but making students recite the honor pledge without changing faculty behavior isn't a solution to the real problem.

Here is what I think is really going on, aside from those egregious examples of catastrophic failure. The Faculty, as a corporate body, honors students for basically only one thing: Grade Point Average. For honors a student requires a departmental recommendation in addition, especially for High and Highest Honors, but a large component of the basis for those recommendations is itself based on GPA. Of course, even to graduate without honors a student needs grades that are above a very low minimum. So grades are the coin of the realm, whether you are struggling to meet the minimum standard or vying to graduate summa.

We do nothing to control the currency. There has been no Faculty-wide discussion of grades in years. Instructors used to get information about overall grade distributions, but such information is no longer routinely distributed.

The only person who ever asks about grades is Professor Mansfield, and he always couches his question as one about grade inflation, sometimes, in my opinion, incorrectly.

The result of the lack of effort at conversation, much less standardization, is that the grading practices of individual faculty members have drifted apart. The only people who realize that this is happening seem to be the students who go from one course to another and are surprised, at the end of the term, to discover that quite different standards have been applied. Sometimes students make life-altering decisions as a result of their very limited view of Harvard grading practices. Economics Professor Claudia Goldin recently discovered quite stunning differences between the way men and women respond to getting B's in introductory economics. If Econ gives a B for the level of work for which English or Computer Science gives an A–, the variability is important, regardless of the compression. (I have no idea whether that might be true, but we did decide a year ago to offer a SAT-UNSAT track through our introductory course.)

I have thought about things that my own Computer Science colleagues might do without a College-wide reform movement. We could certainly talk to each other about our grading practices, and we have started to do that. A more radical step would be to stop basing our honors recommendations so much on GPA. We might, for example, collectively examine and discuss students' programs, and reward, quite subjectively, the programs that are ambitious and daring as well as meeting some more relaxed standard of achievement. This would have various problems -- there could be inconsistencies and biases, and students would still have to meet certain College-wide grading standards to actually receive honors, whatever the local faculty group might recommend. It might be unworkable and it might be unfair, but the alternative is not without its own problems. Carrying out GPA calculations to five decimal places to check against a numerical threshold, when the input data are so noisy, uncalibrated, and unreliable, ought to offend the sensibilities of any proper scholar.

Students, to the extent they understand that they are playing a game (one course from each of 8 Gen Ed areas, so many for the concentration, grading standards all over the place, and your goal is to exceed GPA x in the concentration and GPA y overall, while relying on assumptions and gossip about the grading practices in individual courses), are likely to cut corners when they decide the rules of the game are meaningless and the objective is arbitrary but valuable. One way of cutting corners is to take the easiest courses. Another is to take the courses you already know the best, and from which you will therefore learn the least, because the only metric for which Harvard will reward you is the grade you receive. And another is to skim the edge of honest behavior and hope you won't get caught, especially in a course where the professor doesn't seem to care much about what students are doing.

I wish we could go after the real problem rather than the symptoms. I don't look kindly on cheating and I wish there were less of it (though I don't really have the sense that there is a lot of it in my courses). But to say that we need to respond with an honor code to our cheating problem, rather than taking up some of the larger educational issues that lie behind it, is a mistake. It feels like the patient has cancer and we are treating the acne, because acne medicine is what we have available to us.