My name appears 41 times in the most recent release of the Epstein files. It shows up in flight logs, email threads and one guest list for a dinner in 2013. Since the release, three organizations have removed me from their advisory boards and my department chair has asked me to "take some time." I think it's worth explaining why every one of those visits was correct. For the avoidance of doubt, I can neither confirm nor deny whether any event that may or may not be alleged to have occurred did or did not occur, nor whether I was or was not present for any such purported non-event, to the extent that any responsive facts exist.
The Stakes
On longtermist assumptions, the future could hold on the order of 10^35 lives. Nick Bostrom estimated that reducing existential risk by even one billionth of one percentage point is worth more than a hundred million present lives. A dollar that goes to existential-risk research is not in the same moral category as a dollar spent anywhere else.
Between 2011 and 2015, Jeffrey Epstein was one of very few people willing to fund speculative research into AI risk and long-run futures. Mainstream foundations considered the field unserious. Government agencies wouldn't touch it. Epstein would.
The Funding
| Year | Amount | Use |
|---|---|---|
| 2011 | $250,000 | Two postdocs, decision theory |
| 2012 | $400,000 | Workshop series on superintelligence |
| 2013 | $600,000 | Early alignment research program |
| 2014 | $350,000 | Travel grants for junior researchers |
| Total | $1.6 million |
At the time, that was a meaningful fraction of all money going to the field. Two researchers supported by those grants now hold senior positions at frontier labs. On a conservative estimate that the program reduced existential risk by one part in ten million, the expected value runs to 10^28 future lives. No reputational cost to one professor enters the same calculation.
The Island
The funding came with a condition: attendance. Epstein wanted researchers at his dinners and on his island. The weekends at Little St. James were how the money flowed. I went eleven times. Each visit was a fundraising meeting that happened to involve a boat.
People now ask what I saw there. My answer is that I adopted a strict attention policy. On each visit my job was to secure funding and discuss research priorities. Information outside that scope couldn't improve any decision I faced and could only interfere with the ones I did face. The value of information can be negative, and a responsible agent declines information with negative value. I didn't ask about his legal history, his staff, or the other guests, and I spent my evenings in the library, which was excellent.
Funging
Critics say I should have refused the money. Suppose I had. Epstein's philanthropic budget wouldn't have shrunk. It would have gone to his other interests, which by every account were far worse than decision theory. Every dollar I took was a dollar diverted from his other interests toward the long-term future. Refusing would have made the world worse and left my hands clean. Clean hands are a luxury good, and the cost is paid by the people whose futures you declined to protect.
This is the same principle behind earning to give. The source of a dollar has no bearing on what that dollar can do. We don't ask where a bednet's funding came from before we hang it.
The Response
My former colleague Dr. Hannah Reyes wrote to me last week: "You flew to a pedophile's island eleven times and you're talking about bednets." Her message contains no argument. It lists facts and names a feeling. My analysis includes all the same facts and weights them against 10^28 lives. That's why we reach different conclusions, and why hers is the one that shouldn't be trusted.
The advisory boards that removed me have made a longtermist error of their own. My network includes people who fund this field, and whoever replaces me will not have the same access. Their decision reduces expected funding for existential-risk work in exchange for a cleaner press release.
Update: A reporter has asked whether I'll return the money. The money was spent in 2015, on research. I've invited him to calculate its value in future lives and have offered to walk him through the table.
The funging argument fails in the usual place. Epstein didn't fund research out of a fixed philanthropic budget that had to go somewhere. He used funding to buy access and legitimacy, and the money only bought those things if respectable people took it and came to dinner. Taking the grants didn't divert his money from worse uses. It sold him the thing he was paying for.
Whatever his motive, the research happened. The counterfactual where I decline doesn't contain the postdocs.
Negative value of information requires that the information can't change any decision you face. What was happening on the island bore directly on the one decision you faced every visit, which was whether to keep taking the money. That's about as clear a case of positive value of information as exists.
Did your attention policy include not reading the court records, or was that a separate policy?
The legal disclaimer in the first paragraph says you can neither confirm nor deny whether you were present for anything. The fourth paragraph says you went eleven times.
The disclaimer covers events. Attendance is a matter of record.
The 10^28 comes from multiplying 10^35 by a reduction in risk of one in ten million that the post never defends. That guess carries the entire result. Make it one in 10^20 and you're at 10^15, and that's still a guess.
I called it a conservative estimate, and I'd defend it as such. Two of the grantees are now senior at frontier labs.
Even granting the expected value, the longtermist writing you're leaning on takes side constraints seriously in this kind of case. Most of the people who wrote it would say there are things you don't do for funding at any expected value, and accepting hospitality from this particular person, repeatedly, after 2008, would be on their list.