The Nerderer
Alas, this weekend I became engrossed by the “OJ Simpson trial for nerds”: the ongoing trial of Hans Reiser (the famous Linux file system developer and supposedly-brilliant high-school accelerant) for the murder of his ex-wife Nina. What makes the case interesting is that Reiser’s defense largely consists of the claim that he was too nerdy and Aspbergerish, too lacking in basic social skills, to realize that doing innocuous things in the weeks following Nina’s disappearance like
- removing the passenger seat of his car, soaking the floorboards, and hiding the car several miles from his house,
- not returning calls from numerous friends and family members searching for his ex-wife (except to tell one that he needed to talk to his lawyer),
- hiding his hard disks, and
- telling his mother (in a wiretapped phone conversation) why he was happy his ex-wife went missing
might lead non-nerds to suspect he was guilty.
Like the “Twinkie defense,” Reiser’s “nerd defense” is an invitation to parody. But my feeling is that in this case, even the “nerd” characterization of Reiser itself is open to question. For one thing, Reiser has a blackbelt in judo and appears to have been obsessed with cultivating physical aggressiveness, both in himself and in his eight-year-old son. For another, it seems the reason he was able to attract Nina in the first place was his swaggering confidence. So while portraying Reiser as a nerdy nebbish might be convenient both for journalists and for Reiser’s defense team, he seems to me to be much closer to an aggressive narcissist.
(Of course that doesn’t imply he’s guilty. But I have to say that, thus far in the trial, Reiser and his defense lawyer have done an excellent job of convincing me that he is. Certainly the defense theory — that in an elaborate frame-up of Hans, Nina suddenly abandoned her two children, friends, and new job, left her car by the side of the road with the groceries to rot in the back, and went into hiding in an unspecified former Soviet state with a fake identity and passport — is difficult for a sane person to accept. And unfortunately for Hans, the fact that Nina was far from a perfect specimen of humanity — sleeping with Hans’s best friend, embezzling his company’s money, and divorcing him as soon as she got her US citizenship — only adds to the prima facie likelihood that her body is currently rotting somewhere in the Sierra Nevadas.)
On the other hand, Reiser was certainly wise to hide his hard disks rather than relying on disk encryption. For this week a team of nine researchers at Princeton and elsewhere — including my friends Alex Halderman and Nadia Heninger — released a paper showing how to take a DRAM chip out of one computer, put it into another computer, and read its contents even though the chip had no power in the interim. (One hint: use canned-air spray dusters as a cheap alternative to liquid nitrogen for “cryopreserving” the chip.) The story made it to the Science Times, although they failed to mention most of the authors by name.
But, you ask, how else have I been procrastinating this weekend? Ah. Peter Woit links to a remarkable set of oral histories from people who were involved with the Princeton math department in the 1930’s. Read Alonzo Church (he of the Church-Turing Thesis) list his graduate students and forget to mention Alan Turing, and Nathan Jacobson talk about the disgusting food that Mrs. Einstein would bring to department receptions. In the midst of possibly the greatest concentration of intellect the world has ever seen or will see, and on the eve of perhaps the greatest calamity the world has ever seen, what is it that people worried about? The oak paneling in Fine Hall, and other trivialities completely different from the sorts of things we academics would worry about today.
Oh right: at the behest of you, my loyal readers, I’m now more than halfway through Vernor Vinge’s A Fire Upon The Deep, an entertaining novel that depicts a far future with malevolent AI beings, faster-than-light travel, and (possibly the nerdiest science-fiction premise of all time) Usenet newsgroups spanning the galaxy, whose flamewars play a major role in the rise and fall of civilizations. Vinge’s estimate of how much longer Usenet would stay relevant was off by a factor of only about 10,000.
Follow
Comment #1 February 25th, 2008 at 12:28 am
I fail to see why studying martial arts implies that one is “obsessed with cultivating physical aggressiveness”.
Comment #2 February 25th, 2008 at 12:39 am
Chris, it doesn’t demonstrate any such obsession by itself, but it’s not just that. (In general, this case strikes me as having the property that any one piece of evidence is completely innocuous; only together do they paint a clear picture.) See for example this Wired article, including the following passages:
On December 22, 2004, the dispute intensifies. Reiser arrives to pick up the kids at the house Nina is renting, and, according to Nina, he shoves her to the ground. The next day, she files a request for a restraining order against Reiser — quickly granted — and reports that he threatened to “make me hurt for the rest of my life.”
Reiser is amused by the implication that he is violent. “In reality I am just a computer gamer, and when someone says I have been demented as a result of [videogame] combat the laughter comes easily,” he writes. He believes mental health professionals scorn people who “teach the culture of manhood to little boys, with all of its inherent opposition to wallowing in wimpiness.”
Reiser delves into this “culture of manhood” in a 32-page filing he submits to the court after Nina accuses him of hurting her. In it, he explains the difference between appropriate and inappropriate violence. Grand Theft Auto, for instance, demonstrates inappropriate violence because players can get away with killing innocent people. “Many other computer games heavily penalize shooting the wrong person, and I prefer those,” Reiser says.
Comment #3 February 25th, 2008 at 12:56 am
Twinkie defense? That’s quite legit. Reiser’s defense has more in common with the Chewbacca defense than anything else.
Comment #4 February 25th, 2008 at 2:56 am
What I would like to know is whether Reiser insisted on a trial out of chutzpah, or whether prosecutors did not seriously negotiate a plea so that they could make a spectacle of him. It sounds like the trial shouldn’t have happened, and that the jury certainly shouldn’t let him off.
I admit that there are two other depressing possibilities. One is that my basic sense of probability and common sense is terrible, and that I am as far fooled by these cursory details as I sadly was by cold fusion for about two weeks. The other is that when criminal prosecutors are up against expensive defense lawyers, it is like playing chess with a missing rook; and that the prosecutors only have an easy time of it against public defenders. That second possibility is suggested by the trials of OJ Simpson, Robert Blake, and Phil Spector. (Maybe it is more like a missing rook if the defendant is a celebrity.)
Comment #5 February 25th, 2008 at 4:13 am
I don’t believe for one second that Reiser did a pre-planned murder, since he wouldn’t have done such a stupid job. More likely, if he had coldly concocted a plan that involved Nina becoming dead, he’d have had the Russian mob take care of the whole thing while he was 1000’s of miles away. So if he’s guillty at all, it could only be because he and Nina got into a fight spontaneously, and he killed her in a fit of rage, then tried to cover it up afterwards. I do remember following the story in the press from the beginning, and being unimpressed at that time with how bad the stories made him look, since I figured the info was coming almost entirely from the police and prosecutors. These days we’re hearing from the defense and it’s not so convincing either, of course.
His car was a Honda CRX which has lots of cargo space in back (a Civic sized car with only 2 seats, replacing the entire rear seat area with cargo space), so he would not have had to remove the passenger seat to transport a dead body.
My theory about the missing car seat and the pile of cash was that he expected the cops to be looking for him, so he planned to make himself scarce by going on a road trip, living off the cash and sleeping in the car to avoid checking into motels.
Comment #6 February 25th, 2008 at 5:09 am
Although, as interesting as the disk encryption result is it still requires that the target computer be on when you get to it.
(Software) disk encryption isn’t really supposed to work against an attacker who has physical access to your system when keys are in memory.
Comment #7 February 25th, 2008 at 5:48 am
Those Princeton histories contain some real gems. I like this:
Comment #8 February 25th, 2008 at 7:23 am
asdf: In general, I think that the physical evidence is weak, and that the case rests on the “what the @*^# else could’ve happened to her?” evidence and the “look at how murderer-y he’s been acting” evidence. In particular, I didn’t understand the missing car seat either, and given its late removal, I think your explanation is the most likely one (the other possible explanation being he thought the seat was somehow contaminated with material from the body). His removal of $10,000 from the ATM’s certainly supports the idea that he was planning an escape. Your explanation has a harder time with the wet floorboards though.
I also don’t know if it was premeditated: it’s entirely possible that the kids being in the basement and the mother being at Burning Man simply created the conditions for him to fly into a fit of rage. One can only say it was post-meditated.
Comment #9 February 25th, 2008 at 8:17 am
Robin: The difficulty with that game is, how do you tell when an academic is “working”? (Admittedly, it was probably easier before computers.)
Comment #10 February 25th, 2008 at 9:13 am
And as if through magic, or possibly through the frantic typing of a web admin being caught exposing non-public material to the world at large, the Princeton archives go missing. Sudden 404.
Comment #11 February 25th, 2008 at 10:45 am
Hi Scott,
I think this might interest you, and maybe could give origin to a new blog post on the subject…
An article giving reasons why people shouldn’t follow science careers here: http://wuphys.wustl.edu/~katz/scientist.html has spawned interesting discussion here: http://news.ycombinator.com/item?id=122106
What bothers me is that the conclusion of many very intelligent people basically agrees with the content of the article… Maybe you could be interested in giving your own impressions on the subject, based on the fact that you are a very successful member of the scientific world and could probably give us a different account on the way the academic world works?
Best regards and sorry if this is not the place for asking…
Hernan
Comment #12 February 25th, 2008 at 11:44 am
hhm: That’s an old piece; I read it years ago. Katz’s rant is correct that opportunities in academic science (at least in the US) are much scarcer than they were during the Cold War; I think government shortsightedness deserves a huge part of the blame for that. On the other hand, countless would-be grad students have already followed the invisible hand and taken Katz’s advice, and are doing quite well in Wall Street, Silicon Valley, etc. So the ones going to grad school are mostly the ones willing to assume the (by now well-known) risks: if they weren’t, they wouldn’t be there.
My fundamental disagreement with Katz is that I think PhD work is increasingly excellent preparation for industry careers. Of course, in some cases (e.g. a quantum computing PhD going to work for an Internet startup), it’s hard to argue that the PhD provides anything beyond general skills like analytical thinking, teamwork, project completion, etc., and that those skills couldn’t just as well be obtained elsewhere. But even in those cases, I think a PhD certainly won’t hurt your chances in industry these days (notwithstanding Phil Greenspun’s PhD expunging service), and it therefore gives many people an opportunity to spend six years doing something they enjoy (and also advancing human knowledge) before switching to something that’s actually rewarded by the economy. (One corollary is that, if you’re not enjoying grad school, then you shouldn’t be there. But this is just an instance of a general rule: don’t choose a career option that causes you years of suffering in the hope that the suffering will end later; it probably won’t.) Also, if there used to be a stigma attached to leaving grad school for industry (either before or after receiving one’s degree), I think that’s basically vanished, and that many PhD programs even see training students for industry as a fundamental part of their mission.
The above comments apply most directly to CS PhD programs, which are what I’m most familiar with, but I think they also apply to physics and other sciences. (As for humanities PhD’s … dude, you’re on your own. 🙂 )
Comment #13 February 25th, 2008 at 12:49 pm
I actually enjoyed Vinge’s A Deepness in the Sky even more, and I highly recommend adding it to your reading list.
Comment #14 February 25th, 2008 at 1:29 pm
Thanks a lot Scott for your kind reply. I was very interested in your opinion about it and I really appreciate it.
Best regards,
Hernan
Comment #15 February 25th, 2008 at 2:24 pm
Scott: So the ones going to grad school are mostly the ones willing to assume the (by now well-known) risks
plus, unfortunately, those who won’t be able to get any decent job in industry. That’s why the grad students can’t do programming or any other meaningful work that requires at least some creativity. Some even don’t know how to multiply matrices… Of course, I’m not talking about top schools, but in the majority of U.S. universities it’s the reality.
Comment #16 February 25th, 2008 at 7:12 pm
asdf: In general, I think that the physical evidence is weak, and that the case rests on the “what the @*^# else could’ve happened to her?”
In my view, it would be a big improvement if courts did more to relate “reasonable doubt” to probability. The concept has stayed qualitative, and by implication, irrational. There are trials where the jury takes any doubts to be reasonable, even if the odds are 10^-10. And there are trials where the jury is so outraged that all doubt is rejected as unreasonable, even if the doubt expresses an 80% chance that the defendant is innocent.
Let’s take 1% as a threshold of reasonable doubt. That may seem glib, but I’ve heard that 1% positive error would be much better than the actual performance of criminal trials. (In another sense it is ungenerous, because in trials as in experiments, you want an extra statistical margin as partial protection against bad analysis.) It doesn’t take many of the facts presented in this case to get to 99% likelihood. To begin with, very few murdered women with restraining orders against former partners are killed by anyone other than the obvious suspect. If you just take the official record, divorce plus restraining order plus vanished woman, then the odds are already probably over 90% that she either ran away or he killed her. Any other incriminating evidence comes on top of that Bayesian prior.
On the other hand, there are probably more women who run away than women who are murdered, even among those who have obtained restraining orders against ex-husbands. But most of those women who run away show up again.
As for the car seat, it takes a lot of work to remove a car seat and it is incriminating. It could be that he killed her in the car seat and she bled all over it. Also, as for the bizarre lack of advance planning, I would not be surprised if he had had repressed thoughts of murder for years, and then finally lost his temper and acted on them.
Comment #17 February 25th, 2008 at 7:43 pm
Greg, I wanted to respond to this question from earlier:
What I would like to know is whether Reiser insisted on a trial out of chutzpah, or whether prosecutors did not seriously negotiate a plea so that they could make a spectacle of him.
While I would put the probability of his being guilty at about 98% (with the remaining 2% filled with a grab-bag of strange possibilities like Sean Sturgeon killing her), I’d put the probability of his being convicted at only about 80%. As you said, and as we saw with OJ and countless other examples, jurors interpret the concept of “reasonable doubt” in wildly divergent and often completely unreasonable ways. So insisting on a trial seems rational on Reiser’s part. On the other hand, I can’t describe other aspects of his post-murder behavior as rational, if by “rational” we mean “aimed at not getting convicted.”
Comment #18 February 25th, 2008 at 7:50 pm
Greg:
Would you then condone someone’s conviction based entirely on the prior? (that is, NO evidence is presented that relates to any event that occurred at the time of or after the crime) I agree with the rest of your post, but this does not sit well with me.
Comment #19 February 25th, 2008 at 8:35 pm
While I would put the probability of his being guilty at about 98%
I guess that I can concede 2%, simply because we’re outsiders, so we don’t know what we might not have been told about the case. On a similar note:
Would you then condone someone’s conviction based entirely on the prior? (that is, NO evidence is presented that relates to any event that occurred at the time of or after the crime)
Yes and no. The line between prior and new information is somewhat arbitrary — the distinction concerns the order in which you learn things and not the age of the evidence itself. But let’s say that the police know that a woman took a contract out on her husband, and then later he disappeared. If they can really argue that the odds are truly overwhelming, then I don’t see the problem with convicting her. Note though that if you are to convict based on probabilities, then you have to be willing to reopen a case in the face of new evidence.
There is another, larger caveat. Some of the rules of criminal trials are game-theoretic rather than just empirical. They are designed not only to steer the jury to the truth, but also to rob the establishment of incentives to manipulate the evidence. You can have evidence that is entirely reliable or conclusive, but still inadmissible because it is a bad precedent for other trials in the future. Maybe a conviction based only on evidence before the crime is too big of a loophole.
Comment #20 February 25th, 2008 at 9:52 pm
I think AnOnEmOus makes a compelling argument for why no legal system based on Bayesianism could ever work. Why not just enter the defendant’s race, religion, age, clothing, hairstyle, criminal record, etc. into a Bayesian formula, and convict if Pr[guilty|evidence]>1-ε? Several answers immediately present themselves:
The conclusion seems inescapable: for epistemic, game-theoretic, and moral reasons, the law can’t aim for Bayesian rationality, except in certain special cases (like DNA tests) where reasonable people can more-or-less agree on the sample space and prior. The law also needs the more nebulous concept of “fairness”. And that seems to me like an unavoidable reason why any legal system will screw up a constant fraction of the time, and reach verdicts wildly at odds with what any idiot can see is the truth. (Of course, there are also many avoidable reasons for this phenomenon.)
Comment #21 February 25th, 2008 at 10:49 pm
I once saw some article by a lawyer actually interpret the saying “better ten guilty men go free than one innocent be punished” as meaning that reasonable doubt meant 90% likelihood of guilt. Most trials end in convictions because (the theory goes) responsible prosecutors don’t indict and try people unless the prior likelihood of their guilt is quite high, so the ten guilty men who go free do so because they aren’t even put on trial.
Re the wet floorboard: according to Reiser’s son’s description at the trial, Reiser carried a garbage bag out of the house that looked big and heavy enough to contain Nina. So my explanation of the floorboard (presuming innocence, which is maybe farfetched at this point) is that the car was full of the usual fast-food effluvia of cars owned by slobs, so he hosed it down in order to be able to sleep in it.
Comment #22 February 25th, 2008 at 11:20 pm
The conclusion seems inescapable: for epistemic, game-theoretic, and moral reasons, the law can’t aim for Bayesian rationality
Now wait a minute. It is true that criminal justice can’t just be the Bayesian calculation for various game-theoretic reasons. As I see it, those are the moral reasons, so you’re double-counting. Meanwhile the epistemic reasons seem like thin ice to me. We are not talking about the philosophical Bayesianism of assigning a prior distribution on questions like whether life exists on Mars. Criminology is a science, and science always invokes restricted Bayesianism in the context of repeatable experiments. I will grant that there is also a non-Bayesian part of science, to the extent that situations lack precedent.
The other wait-a-minute is that even though I agree that trials shouldn’t just be probability calculations, justice certainly should be informed by them. Current practice is to reject any mathematical discussion of “reasonable doubt” as a slippery road to unfair Bayesianism. Now, this rejection may be necessary if the jury or judge is very bad at probability. However, I argue that the results are much less fair than if the system could address the math.
After all, experimental scientists know all about the epistemic, moral, and game-theoretic issues in reporting scientific findings. The question of whether a hypothesis is “confirmed” is a lot like the question or reasonable doubt. But scientists take these concerns as a mandate to do more statistical calculations, not to reject statistics as dangerous.
Comment #23 February 25th, 2008 at 11:29 pm
On the other hand, I can’t describe other aspects of his post-murder behavior as rational, if by “rational” we mean “aimed at not getting convicted.”
Except that if he’s guilty, then it’s tricky to put yourself in his shoes. You would be faced with a lot of bad choices. If you buy crime books after you commit murder, then it looks bad. But if you don’t buy crime books, then how would you figure out how to get rid of the evidence? Etc.
Comment #24 February 25th, 2008 at 11:59 pm
As I see it, those are the moral reasons, so you’re double-counting.
To me, the race example seems moral but not game-theoretic, whereas the hairdo example seems game-theoretic but not as directly moral. The difference is that changing your hairdo is a possible game play while changing your race is not.
Criminology is a science, and science always invokes restricted Bayesianism in the context of repeatable experiments. I will grant that there is also a non-Bayesian part of science, to the extent that situations lack precedent.
To me, Bayesianism is to rationality more-or-less as special relativity is to general relativity: it’s an approximation valid only in flat, well-behaved patches of hypothesis space, those for which there’s a sensible prior. Extending this analogy, I would say that interesting science (almost by definition) is science that ventures outside the Bayesian patches of hypothesis space, and likewise for interesting trials (those that attract wide attention, set precedents, go to appeals court, etc).
One corollary of this argument, which I didn’t anticipate but am forced to, is that Bayesianism could be used much more widely in “routine” trials.
though I agree that trials shouldn’t just be probability calculations, justice certainly should be informed by them.
Here I’m in total agreement. As one famous example, had the OJ jury understood that
Pr[husband murders wife|husband abuses wife]≠Pr[husband is the murderer|husband abuses wife AND wife is murdered],
the verdict might have been different.
Comment #25 February 26th, 2008 at 12:00 am
Dutch nurse Lucia de Berk is an example of the possibilities afforded by Bayesian prosecution. She became a suspect as a result of being on duty during the deaths of a suspiciously large number of patients. The prosecution presented a p-value of 1 in 342 million against this being a coincidence. (Report from Nature on doubts about the case.)
Comment #26 February 26th, 2008 at 1:10 am
Greg, having thought about it more, I now agree with you that the game-theoretic arguments against a Bayesian justice system seem like a subcase of the moral arguments (though not vice versa).
To see why, let’s first consider correlates of innocence that are easy for anyone to adopt: say, not having a villainy mustache. If it became widely known that the courts were preferentially convicting mustachioed men, then rational villains would simply respond by shaving their mustaches, thereby negating the law’s advantage — but in the meantime, the right of innocent men to grow mustaches would arguably have been compromised for no good long-term reason.
Now let’s consider correlates of innocence that are hard to adopt: say, not having a criminal record, being respected by one’s neighbors, or being middle-class. Here we need to distinguish between those attributes that are “morally relevant” and those that are not. For the morally relevant attributes (e.g., not having a criminal record), the courts can and should enter them into their Bayesian calculations — since even at equilibrium, criminals will find these attributes hard to fake, and thus they’ll continue to provide useful information. The morally irrelevant attributes, like being middle-class, are also hard to fake, and thus also provide the courts with useful information even at equilibrium. But those we’ve already decided they shouldn’t use for moral reasons, having nothing to do with game theory.
Comment #27 February 26th, 2008 at 1:41 am
Dutch nurse Lucia de Berk is an example of the possibilities afforded by Bayesian prosecution.
But Bayes is not at fault if the problem is bad statistics. For one thing, both the frequentists and the Bayesians believe in p-values, yet the concern in this case is that the p-value is wrong. Qualitatively similar statistical evidence played a role in the Harold Shipman case, and that time they were right.
It would be amusing if anyone was such a dogmatic frequentist as to reason as follows: No matter how many seniors died on Shipman’s watch, that does not make it likely that Shipman killed them because there is no prior probability of his guilt.
Comment #28 February 26th, 2008 at 2:12 am
Greg, having thought about it more, I now agree with you that the game-theoretic arguments against a Bayesian justice system seem like a subcase of the moral arguments (though not vice versa).
This train of discussion is just semantics. What I meant is not the game played by the police and the defendants, but rather the one played by the police and the jury. The jury wants a fair outcome; the police want a conviction. (Indeed, it’s like Arthur-Merlin.) Whenever two parties have divergent interests, you can call that game theory. Among other things, you’re pondering the problem of race-based statistical arguments. The issue is exactly that racist crime control could be the police’s objective, but not yours.
Comment #29 February 27th, 2008 at 4:22 pm
Isn’t that all Valentine and Peter were doing in Ender’s Game?
Comment #30 March 1st, 2008 at 8:10 pm
Mr Reiser does not deserve to be categorised as “The Nerderer”. After all, his Mrs had an understanding with an admitted serial killer. Isn’t it more probable that a wight of that character would more likely have caused her disappearance? Or would this be more of a case of “Who Framed Sean Sturgeon”? Was Mr Reiser diabolically crafty enough to put Mr Sturgeon in the spotlight?
Comment #31 March 1st, 2008 at 8:42 pm
Graham: Sturgeon “admitted” to being a serial killer; the trouble is that there’s no evidence beyond his admission that he actually was one! Thus, I think the “default theory” is that he’s a severely mentally disturbed person who (for whatever reason) fabricates wild stories about himself. Still, his role in this case is the main reason I gave only a 98% chance of Reiser being guilty instead of (say) 99.8%.
Comment #32 March 3rd, 2008 at 1:20 am
Thus, I think the “default theory” is that he’s a severely mentally disturbed person who (for whatever reason) fabricates wild stories about himself.
Another theory is that he is not entirely crazy, and that he’s trying to repay a debt to his former friend.
In other news, I usually think of Wired Magazine as fatuous. (Some people call it Weird Magazine.) However, the article on Reiser is really very good.
Comment #33 March 3rd, 2008 at 1:41 am
Still, his role in this case is the main reason I gave only a 98% chance of Reiser being guilty instead of (say) 99.8%.
At least as the case has been reported, it’s hard to put any stock in what Sturgeon has said. I do not see a 2% chance that Sturgeon will be indicted for anyone’s murder. However, from my outside vantage point, I can imagine a 2% or so chance that the story has simply been grossly distorted in some way that I can’t anticipate. Hypothetically, Nina may have run away before. Or, hypothetically, the crucial passenger seat has been found. Or something else like that. I would want to see the trial transcript if my opinion actually mattered.
In a way, it’s not really a remark about this particular case. I remember reading about the fiasco with David Baltimore, his coauthor Thereza Imanishi-Kari, and his postdoc Margot O’Toole. I was convinced for a long time that Baltimore’s lab had probably seen scientific fraud. But when I eventually got access to a thorough report, I realized that I just hadn’t had the facts. Maybe I still don’t have the facts, who knows, but at the moment I have no reason to believe the original accusations.