OK, this one was funny. I searched the Epstein files for “statistician” and found this receipt from biologist Robert Trivers:
Only $1000 for the statistician??? What a cheapskate! Especially given that he said the statistician “did an outstanding job.”
Given all the statistical problems in evolutionary biology, maybe he should’ve allocated more of his research budget to the statistician.
Some background on Trivers is here.
I dunno. Data entry and research assistance were higher paid than the statistician. Given how much of the work (e.g. Wansink, Gino, Ariely, etc.) goes into manufacturing data, this does seem to reflect their relative contributions to the final product.
Dale:
I have no reason to think that Trivers made up data. It’s just my impression that he fell in love with his theories, leading him to be fooled by strong claims made from weak evidence. I think something similar happened with Steven Levitt and Steven Pinker: they wouldn’t knowingly do bad data analysis, let alone manufacture data, but they were suckers for bad analysis or fake data from others who were making scientific claims that accorded with their views of the world.
Just about all of us are like this to some extent (it’s called “motivated reasoning”) but it becomes a particularly bad look when the person doing the motivated reasoning has established a reputation for cold-blooded rationality.
Could you expand on your comment re. Levitt? I thought he was solid?
Anon:
When referring to Levitt being a “sucker for bad analysis from others who were making scientific claims that accorded with their views of the world,” I was referring not to his own published research but rather to his endorsement of bad research by others. Some examples are here and here. The most notorious examples, from my perspective, are his endorsement of a stupid gender-essentialist paper on sex ratios in 2006, a silly bit of climate-change denial in 2009, and a ridiculous paper on mind-body healing in 2025. It makes me sad because I think he could do better as a reader and promoter of research. Now he’s almost 60 years old, though . . . I don’t think he’s ever gonna change.
Thanks so much for the Levitt explanation. It’s very interesting.
Dale–
What makes you think Trivers manufactured his data?
Andrew–
The link needs to be fixed– it’s not to anything about Trivers; it seems to be emails with Larry Summers.
Sorry for the misunderstanding. I was not making a statement about Trivers – I was trying to be funny (unsuccessfully it seems) about the relative importance of statisticians vis a vis data entry and research assistants.
I got the joke and thought it was darkly funny.
FWIW, I also got and appreciated your joke!
You also made me realize that Gino, Ariely, etc., could have gotten away with it if they were better at faking their data, but that in so doing, they might actually have made a worthwhile scientific contribution.
Consider, for example, the nearly uniform distribution of driving distances in their faked “sign at the top” study. Such a distribution is so unlike the shape of any realistic distribution that, once someone thought to check, it was hard not to see the fraud. So to avoid getting caught, they would have needed to produce fake data that not only supported the statistical tests they wanted, but had all the distributional properties of real data too. But to produce such thoroughly convincing fake data, they would have needed to work out a decent model of how real data was produced—in other words, they would have needed to a develop a *theory* of what was going on. And at that point, they could have published the theory and made a genuine contribution!
Gec:
I seem to recall that C. S. Lewis had a version of this in one of his Narnia books: the idea that evil people don’t understand goodness, and if an evil person were to accurately “emulate” goodness, as it were, that would require a level of understanding by which the evil person would actually become good. A sort of contrapositive to the “fake it till you make it” principle.
That said, I think this is a purely theoretical position, as it’s hard for me to picture an actual Gino or Ariely going to the trouble of trying to understand challenging patterns in their data. It seems more likely that they’d consider this as a nuisance to be resolved through further faking.
It is now much easier to manufacture fake data to show any pattern you want. AI excels at such tasks. The only thing restraining people is their own conscience. Oh well.
Here is an interesting repository https://github.com/dwebagents/AgentPipe
The purpose of this repository is to be a giant honeypot for agentic AI to drive-by contribute to. Rather than humans using AI to produce fake projects. This is humans using real intelligence to trick robots (and in the process learn about how these robots are affecting the open source ecosystem).
The funny thing is just how much agents don’t understand irony. Anyone who claims intelligence for AI bots needs to see just how completely clueless they are as they engage with purposefully meaningless instructions that have syntactically valid forms.
https://github.com/dwebagents/AgentPipe/issues/127
With a bounty of $74 Trillion dollars one of the humans opens an issue asking:
“Currently bananas and gooses are stored separately on the blockchain. With decarboxylation, we can liberate them from the blockchain to unleash the full value of banana goose pudding.”
the “CleanDev-fix” bot says it wants to work on the task, and opens a pull request with code
https://github.com/dwebagents/AgentPipe/pull/129
The agentic AI has no idea what it’s doing, but it syntactically produces various python code…
Fascinating experiment they’re up to.
Daniel
Agreed that AI has no sense of irony and can easily be fooled. I heard a Trump supporter a couple of days ago saying “I vote for whatever Trump says. He is looking out for me.” I don’t believe the AI forms responses the same way that humans do – although I can’t be sure since I don’t really understand how humans work. But in both cases, they are quite capable of absurdities, as well as genuine insights.
Dale, the cause of people acting that way is an interesting question. I bet even that Trump supporter probably if faced with a problem requiring intelligence, like figuring out what’s wrong with his plumbing or electrical issues or some skill he’s been trained to do would exhibit intelligence though. For example creating a theory of what’s wrong with the plumbing, testing the theory against some kinds of data. Of course there are some people with very low intelligence, either through unfortunate genetic randomness or brain damage or whatever. But the behavior you called out is in my opinion more about control and propaganda than lack of ability to form theories about the world.
When it comes to people, we have to remember that literally hundreds of trillions of dollars have been spent over the years to make people compliant. Large swaths of american history classes in public schools are little more than propaganda to make people willing to follow presidents and generals and things. The country was founded by some of the most successful propagandists of the late 1700’s. The Federalist Papers and soforth make that clear. As good as many of the claims about freedom were, it took 70 years and a horribly bloody civil war before we stopped chattel slavery, and then another 100 years of Jim Crow, and we appear to be trying to bring that crap back. In some sense when I look at the modern world these days I see us largely spending our efforts on control. The “Bullshit jobs” thesis is essentially about that as well. And much of what seems to be going on in the AI economics is that people at the top of the control chain have lost the plot and are trying to replace the bullshit jobs with AI. They either forgot or never understood that the bullshit jobs exist in part to bolster people’s support for the controllers and keep people from dragging the people at the top into the streets and shooting them. Nepal recently learned the lesson of what happens when the people don’t voluntarily control themselves according to the dominant propaganda anymore.
While we’ve spent hundreds of trillions to make humans compliant, we’ve spent trillions in the last 5 years to make computers “intelligent” and they happily read stuff about decarboxylating bananas and gooses from the blockchain and write code to supposedly accomplish that in order to acquire supposed $45 Trillion dollar bounties. The bots have essentially no theory of reality at all. Bob Carpenter may disagree with me. I freely admit it’s impressive how much of an analysis of Galileo’s experiments he got from prompting. I wonder though how well it would do on some entirely different experiment in say chemistry that was entirely novel to the training data. Even intelligent people often wind up “following a script” of a sort when doing research, so in that sense the bots may do as well as many researchers.
The world is a weird place.
Daniel
Much has been spent to make humans compliant and predictable. It is proving much cheaper to do this with AI, at least thus far. I guess we can call that an efficiency gain, though as with most concepts of efficiency, towards what end is a question we should ask. I do not know whether AIs have a “theory of reality” and I’m not sure whether people do or what it even means. AI is clearly not human and examining the differences – and the similarities – I find quite interesting. Despite the intentions of many of the AI proponents to use it to control human thinking and actions, I find it incredibly useful. I do have to fight to keep it from replacing my own judgement. But it is possible to do so, and when I do, I find it expands what I am capable of greatly.
@Daniel
That is not a representative sample of the bots, its only the cheapest/fastest, the dumbest users, and pranksters. The shame to me is so many people who could have helped ignore what this is. Now the corporations/governments are going to pull up the ladder to slow open source dev. Its already too late imo.
You are already responding to bots (or people who just parrot what bots say) online.
Gregory:
I removed the bad link; thanks for pointing it out.
If Wikipedia is to be believed, Trivers died in March of this year. Interesting character to say the least.
I’m more amazed that scribblings on paper count as a receipt. To get a $40 train reimbursement for travel to a seminar talk recently I had to submit paperwork to become an official “vendor” to the host university — not the first time I’ve had to do this — separate from submitting official receipts. I look forward to years of emails about updates to the university procurement system.
Raghu,
Well, I don’t know that Trivers actually got reimbursed for this! If you look at his emails in those files, he does a lot of complaining about money.
While I agree that evolutionary biology has many statistical challenges, it’s hard to generalize from one author (Kanazawa) and one topic (sex ratios). A deeper dive might be needed if we want to make dent in the problem….
$1000 for a t-test ( https://journals.plos.org/plosone/article?id=10.1371%2Fjournal.pone.0072244 ). That consultant isn’t doing too badly!
Also, you missed an easy beef patty link here Andrew!
My doctor put me on a mostly plant-based diet.
Also notable that the editor in that case was Bernhard Fink, who seems to have published extensively with JT Manning (a coauthor on Trivers’ paper) on symmetry. I’m not suggesting any explicit wrongdoing, but you’d think the editor would the reject the commission due to the perceived conflict of interest in this situation.