As I nominate this, Holden Karnofsky recently wrote about “Minimal Trust Investigations” (124 upvotes), similar to Epistemic Spot Checks. This post is an example of such a minimal trust investigation.
The reason why I am nominating this post is that
It seems to me that Guzey was right on several object-level points
The EA community failed both Guzey and itself in a variety of ways, but chiefly by not rewarding good criticism that bites.
That said, as other commenters point out, the post could perhaps use a re-write. Perhaps this decade review would be a good time.
This post is an example of such a minimal trust investigation.
It’s an example of a minimal trust investigation done badly.
In order to get smarter from minimal trust investigations, you need to (a) have a sense of proportion about how to update based on your conclusions, and (b) engage with counterarguments productively. Guzey has demonstrated a wild lack of (a), and while I respect his willingness to engage at length with these counterarguments on this Forum and elsewhere, the apparent lack of any updating (and his continued fixation on the leak screwup years later) speaks pretty badly.
To be clear, I do think this post provided some value, and that versions of this post quite similar to the one that actually exists would have provided much more value. But Guzey’s actual behaviour here is not something we should emulate in the community, beyond the very basic idea of epistemic spot checks on EA books (which I support).
The EA community failed both Guzey and itself in a variety of ways
CEA’s screwup with the essay draft was pretty bad (I’ve said before I think it was sufficiently bad that it should be on their mistakes page). But I was actually quite proud of the way the rest of the community (at least on the Forum) responded to this. Lots of people responded thoughtfully and at length, did the requisite background reading, and acknowledged the basic validity of some of his points. The fact that people didn’t agree with his wildly-out-of-proportion conclusions doesn’t mean they failed him.
Ahhh, but it is not clear to me that this is that disproportionate. In particular, I think this is a problem of EA people having more positive priors about MacAskill. Guzey then starts with more neutral priors, and then correctly updates downwards with his review, and then even more downwards when a promise of confidentiality was breached.
With regards to the contents of the book, I think the size of the downward updates exhibited in the essay dramatically exceeds the actual badness of what was found. Identifying errors is only the first step in an exercise like this – you then have to accurately update based on what those errors tell you. I think e.g. David Roodman’s discussion of this here is a much better example of the kind of work we want to see more of on the Forum.
With regards to the confidentiality screw-up, sure, it’s rational to update downwards in some general sense, but given that the actual consequences were so minor and that the alternative hypothesis (that it was just a mistake) is so plausible, I don’t respect Guzey’s presentation of this incident in his more recent writings (e.g. here).
Do you believe that the following representation of the incident is unfair?
Yes, at present I do.
I haven’t yet seen evidence to support the strong claims you are making about Julia Wise’s knowledge and intentions at various stages in this process. If your depiction of events is true (i.e. Wise both knowingly concealed the leak from you after realising what had happened, and explicitly lied about it somewhere) that seems very bad, but I haven’t seen evidence for that. Her own explanation of what happened seems quite plausible to me.
(Conversely, we do have evidence that MacAskill read your draft, and realised it was confidential, but didn’t tell you he’d seen it. That does seem bad to me, but much less bad than the leak itself – and Will has apologised for it pretty thoroughly.)
Your initial response to Julia’s apology seemed quite reasonable, so I was surprised to see you revert so strongly in your LessWrong comment a few months back. What new evidence did you get that hardened your views here so much?
And that since “the actual consequences were so minor and that the alternative hypothesis (that it was just a mistake) is so plausible” this doesn’t really matter?
It matters – it was a serious error and breach of Wise’s duty of confidentiality, and she has acknowledged it as such (it is now listed on CEA’s mistakes page). But I do think it is important to point out that, other than having your expectation of confidentiality breached per se, nothing bad happened to you.
One reason I think this is important is because it makes the strong “conspiracy” interpretation of these events much less plausible. You present these events as though the intent of these actions was to in some way undermine or discredit your criticisms (you’ve used the word “sabotage”) in order to protect MacAskill’s reputation. But nobody did this, and it’s not clear to me what they plausibly could have done – so what’s the motive?
What sharing the draft with MacAskill did enable was a prepared response – but that’s normal in EA and generally considered good practice when posting public criticism. Said norm is likely a big part of the reason this screw-up happened.
I’m commenting because you are really good in every sense, also your comment is upvoted and together this is a sign that I am wrong. I want to understand more.
Also, the consequent discussion would do as you suggest in giving attention to Guzey’s ideas (although in my comment I don’t find much content in them).
This is technically true but seems to be a nitpick.
What is going on is that MacAskill is probably pointing out that the focus on expenses and not theory of change/effectiveness is a massive hurdle that contributes to the culture of scarcity and “winning two games”. This undermines effectiveness of charities.
I guess that the truth is Charity Navigator has little ability or interest in examining the uses of overhead or understanding theory of change of charities.
It seems that Guzey objects to MacAskill using 0.1% as an exaggeration, and seems to point out Charity Navigator is “ok” with 25%. This isn’t that substantive (and I’m skeptical this is the complete truth of Charity Navigator’s position).
2. MacAskill cites different sets of evidence to support deworming compared to other interventions, deworming fails some metrics :
The truth is that leading scientists often point to different sets of evidence when comparing to different interventions (and there can be bitter disputes between two respected scientists). What probably happened is that MacAskill believed this was a good intervention and cited the current evidence at the time.
To be clear, it’s sometimes the truth and correct to cite different sets of evidence when comparing different interventions.
Even if this was wrong, and this occurred more than once, it’s not clear this is a defect in science or epistemics.
There’s a lot of talk around GiveWell and I’m not an expert, but sometimes I hear some people say GiveWell is conservative. Maybe it’s because of the intense politics that it has to deal with while carrying the flag of rigor. If this is true, maybe focusing on episodes like this is counterproductive.
Also, there seems to be some misfires in criticism and other things that are ungenerous or wrong with the content in Guzey’s essay. It’s impractical to list them all.
Leaking
The vast majority of Guzey’s comment is focused on this episode where his book was “leaked”. It seems this leak really happened. Also, it seems pretty plausible it was an accident, even reading the quotes that Guzey cites to suggests something nefarious is going on.
Guzey suggests this was really bad and hints at retaliation, insinuating how critical or harmful this was to him many times “What happened around the publication of the essay, however, completely dwarfs the issues in the book itself”). But doesn’t describe any issues besides the leaking of the document itself (which seems like it would go to MacAskill soon enough anyways).
Even rounding down all of the other things that are negative signals to me, this fixation on this episode after these years seems like a strong sign to me, and most people I know would be a strong signal to me, and would be a strong signal to most people I know, of the low value of the ideas from this person.
Another way of looking at this is that sometimes, there can be a large supply of people with critical ideas and many of these turn out to be wrong, and vexatious, really. There would be no work done if we didn’t use these heuristics before engaging with their ideas.
For me, I think a crux is that I suspect Guzey faced no retaliation, and that really undermines his apparent fixation. My opinion would be changed if he faced actual repercussions because of the leak or because of ideas in his book.
Even rounding down all of the other things that are negative signals to me, this fixation on this episode after these years seems like a strong sign to me, and most people I know, of the low value of the ideas from this person.
This part sounds deeply wrong to me, but that’s probably because I’ve read more of Guzey’s work besides this one piece.
I occasionally encounter people who would have been happy to burn their youth and spend their efforts on projects that would good bets from an EA perspective. When they don’t get to for one reason or another (maybe they were being too status-seeking too soon, maybe they would have been a great chief of staff or a co-founder together with someone who was epistemically stronger, maybe they didn’t have the right support networks, etc.), it strikes me as regrettable.
I think that some of Guzey’s later work is valuable, and in particular his New Science organization looks like a good bet to take. I think he got some funding from the Survival and Flourishing Fund, which is EA adjacent.
Below is my reply to this comment and your other one.
I’m not sure this is valuable or wise for me to write all this, but it seems better to communicate.
I was sincere about when I said I didn’t understand and wanted to learn why you rated Guzey’s criticism highly. I think I learned a lot more now.
_________________________
You said “This part sounds deeply wrong to me, but that’s probably because I’ve read more of Guzey’s work besides this one piece”:
Note that I made a writing error in the relevant paragraph. It is possible this changed the meaning of my comment and this rightfully offended you. When I said:
Even rounding down all of the other things that are negative signals to me, this fixation on this episode after these years seems like a strong sign to me, and most people I know, of the low value of the ideas from this person.
I meant:
Even rounding down all of the other things that are negative signals to me, this fixation on this episode after these years would be a strong signal to me, and would be a strong signal to most people I know, of the low value of the ideas from this person.
The first version could imply there is some “establishment” (yuck) and that these people share my negative opinion. This is incorrect, I had no other prior knowledge of opinions about Guzey and knew nothing about him.
_________________________
You said here:
I occasionally encounter people who would have been happy to burn their youth and spend their efforts on projects that would good bets from an EA perspective. When they don’t get to for one reason or another (maybe they were being too status-seeking too soon, maybe they would have been a great chief of staff or a co-founder together with someone who was epistemically stronger, maybe they didn’t have the right support networks, etc.), it strikes me as regrettable.
I think that some of Guzey’s later work is valuable, and in particular his New Science organization looks like a good bet to take. I think he got some funding from the Survival and Flourishing Fund, which is EA adjacent.
This seems wise and thoughtful. I didn’t know about this.
_________________________
You made another comment:
I find it surprising that your comment only provides one-sided considerations. As an intuition pump, consider reading this unrelated review, also by Guzey, and checking if you think it is also low quality.
I skimmed this, in the spirit of what you suggested. The truth is that I find reviews like this often on the internet, and I use reviews of this quality for a lot of beliefs.
If I didn’t know anything about Guzey, I would use his review to update in favor his ideas. But at the same time, I find many of Guzey’s choices in content and style different than people who successfully advance scientific arguments.
_________________________
I think that, combined with virtue and good judgement, being loyal to someone is good.
In non-EA contexts, I have tried to back up friends who need support in hostile situations.
In these situations, the truth is that I can become strategic. When being strategic, I try to minimize or even rewrite the content where they were wrong. This can lead to compromise and closure, but this needs coordination and maturity.
I hadn’t realised that your comment on LessWrong was your first public comment on the incident for 3 years. That is an update for me.
But also, I do find it quite strange to say nothing about the incident for years, then come back with a very long and personal (and to me, bitter-seeming) comment, deep in the middle of a lengthy and mostly-unrelated conversation about a completely different organisation.
Commenting on this post after it got nominated for review is, I agree, completely reasonable and expected. That said, your review isn’t exactly very reflective – it reads more as just another chance to rehash the same grievance in great detail. I’d expect a review of a post that generated so much in-depth discussion and argument to mention and incorporate some of that discussion and argument; yours gives the impression that the post was simply ignored, a lone voice in the wilderness. If 72 comments represents deafening silence, I don’t know what noise would look like.
I find it surprising that your comment only provides one-sided considerations. As an intuition pump, consider reading this unrelated review, also by Guzey, and checking if you think it is also low quality.
The low quality of Guzey’s arguments around Doing Good Better (and his unwillingness to update in the face of strong counterarguments) substantially reduced my credence in his (similarly strong) claims about Why We Sleep, and I was confused about why so many people I know put so much credence in the latter after the former.
I think Guzey is very honest in these discussions (and subsequently), and is trying to engage with pushback from the community, which is laudable.
But I don’t think he’s actually changed his views to nearly the degree I would expect a well-meaning rational actor to do so, and I don’t think his views about MacAskill being a bad actor are remotely in proportion to the evidence he’s presented.
For example, relating to your first link, he still makes a big deal of the “interpretation of GiveWell cost-effectiveness estimates” angle, even though everyone (even GiveWell!) think he’s off base here.
On the second link, he has removed most PlayPump material from the current version of his essay, which suggests he has genuinely updated there. So that’s good. That said, if I found out I was as wrong about something as he originally was about PlayPumps, I hope I’d be much more willing to believe that other people might make honest errors of similar magnitude without condemning them as bad actors.
As I nominate this, Holden Karnofsky recently wrote about “Minimal Trust Investigations” (124 upvotes), similar to Epistemic Spot Checks. This post is an example of such a minimal trust investigation.
The reason why I am nominating this post is that
It seems to me that Guzey was right on several object-level points
The EA community failed both Guzey and itself in a variety of ways, but chiefly by not rewarding good criticism that bites.
That said, as other commenters point out, the post could perhaps use a re-write. Perhaps this decade review would be a good time.
It’s an example of a minimal trust investigation done badly.
In order to get smarter from minimal trust investigations, you need to (a) have a sense of proportion about how to update based on your conclusions, and (b) engage with counterarguments productively. Guzey has demonstrated a wild lack of (a), and while I respect his willingness to engage at length with these counterarguments on this Forum and elsewhere, the apparent lack of any updating (and his continued fixation on the leak screwup years later) speaks pretty badly.
To be clear, I do think this post provided some value, and that versions of this post quite similar to the one that actually exists would have provided much more value. But Guzey’s actual behaviour here is not something we should emulate in the community, beyond the very basic idea of epistemic spot checks on EA books (which I support).
CEA’s screwup with the essay draft was pretty bad (I’ve said before I think it was sufficiently bad that it should be on their mistakes page). But I was actually quite proud of the way the rest of the community (at least on the Forum) responded to this. Lots of people responded thoughtfully and at length, did the requisite background reading, and acknowledged the basic validity of some of his points. The fact that people didn’t agree with his wildly-out-of-proportion conclusions doesn’t mean they failed him.
Ahhh, but it is not clear to me that this is that disproportionate. In particular, I think this is a problem of EA people having more positive priors about MacAskill. Guzey then starts with more neutral priors, and then correctly updates downwards with his review, and then even more downwards when a promise of confidentiality was breached.
Am I missing something here?
With regards to the contents of the book, I think the size of the downward updates exhibited in the essay dramatically exceeds the actual badness of what was found. Identifying errors is only the first step in an exercise like this – you then have to accurately update based on what those errors tell you. I think e.g. David Roodman’s discussion of this here is a much better example of the kind of work we want to see more of on the Forum.
With regards to the confidentiality screw-up, sure, it’s rational to update downwards in some general sense, but given that the actual consequences were so minor and that the alternative hypothesis (that it was just a mistake) is so plausible, I don’t respect Guzey’s presentation of this incident in his more recent writings (e.g. here).
(deleted)
Yes, at present I do.
I haven’t yet seen evidence to support the strong claims you are making about Julia Wise’s knowledge and intentions at various stages in this process. If your depiction of events is true (i.e. Wise both knowingly concealed the leak from you after realising what had happened, and explicitly lied about it somewhere) that seems very bad, but I haven’t seen evidence for that. Her own explanation of what happened seems quite plausible to me.
(Conversely, we do have evidence that MacAskill read your draft, and realised it was confidential, but didn’t tell you he’d seen it. That does seem bad to me, but much less bad than the leak itself – and Will has apologised for it pretty thoroughly.)
Your initial response to Julia’s apology seemed quite reasonable, so I was surprised to see you revert so strongly in your LessWrong comment a few months back. What new evidence did you get that hardened your views here so much?
It matters – it was a serious error and breach of Wise’s duty of confidentiality, and she has acknowledged it as such (it is now listed on CEA’s mistakes page). But I do think it is important to point out that, other than having your expectation of confidentiality breached per se, nothing bad happened to you.
One reason I think this is important is because it makes the strong “conspiracy” interpretation of these events much less plausible. You present these events as though the intent of these actions was to in some way undermine or discredit your criticisms (you’ve used the word “sabotage”) in order to protect MacAskill’s reputation. But nobody did this, and it’s not clear to me what they plausibly could have done – so what’s the motive?
What sharing the draft with MacAskill did enable was a prepared response – but that’s normal in EA and generally considered good practice when posting public criticism. Said norm is likely a big part of the reason this screw-up happened.
(deleted)
I don’t agree with this review at all.
I’m commenting because you are really good in every sense, also your comment is upvoted and together this is a sign that I am wrong. I want to understand more.
Also, the consequent discussion would do as you suggest in giving attention to Guzey’s ideas (although in my comment I don’t find much content in them).
Here are comments on object level points in Guzey’s recent reply:
The book misrepresented Charity Navigator’s emphasis on reducing overhead per https://guzey.com/books/doing-good-better/#charity-navigator
This is technically true but seems to be a nitpick.
What is going on is that MacAskill is probably pointing out that the focus on expenses and not theory of change/effectiveness is a massive hurdle that contributes to the culture of scarcity and “winning two games”. This undermines effectiveness of charities.
I guess that the truth is Charity Navigator has little ability or interest in examining the uses of overhead or understanding theory of change of charities.
It seems that Guzey objects to MacAskill using 0.1% as an exaggeration, and seems to point out Charity Navigator is “ok” with 25%. This isn’t that substantive (and I’m skeptical this is the complete truth of Charity Navigator’s position).
2. MacAskill cites different sets of evidence to support deworming compared to other interventions, deworming fails some metrics :
https://guzey.com/books/doing-good-better/#educational-benefits-of-distributing-textbooks-and-deworming
The truth is that leading scientists often point to different sets of evidence when comparing to different interventions (and there can be bitter disputes between two respected scientists). What probably happened is that MacAskill believed this was a good intervention and cited the current evidence at the time.
To be clear, it’s sometimes the truth and correct to cite different sets of evidence when comparing different interventions.
Even if this was wrong, and this occurred more than once, it’s not clear this is a defect in science or epistemics.
There’s a lot of talk around GiveWell and I’m not an expert, but sometimes I hear some people say GiveWell is conservative. Maybe it’s because of the intense politics that it has to deal with while carrying the flag of rigor. If this is true, maybe focusing on episodes like this is counterproductive.
Also, there seems to be some misfires in criticism and other things that are ungenerous or wrong with the content in Guzey’s essay. It’s impractical to list them all.
Leaking
The vast majority of Guzey’s comment is focused on this episode where his book was “leaked”. It seems this leak really happened. Also, it seems pretty plausible it was an accident, even reading the quotes that Guzey cites to suggests something nefarious is going on.
Guzey suggests this was really bad and hints at retaliation, insinuating how critical or harmful this was to him many times “What happened around the publication of the essay, however, completely dwarfs the issues in the book itself”). But doesn’t describe any issues besides the leaking of the document itself (which seems like it would go to MacAskill soon enough anyways).
Even rounding down all of the other things that are negative signals to me, this fixation on this episode after these years
seems like a strong sign to me, and most people I knowwould be a strong signal to me, and would be a strong signal to most people I know, of the low value of the ideas from this person.Another way of looking at this is that sometimes, there can be a large supply of people with critical ideas and many of these turn out to be wrong, and vexatious, really. There would be no work done if we didn’t use these heuristics before engaging with their ideas.
For me, I think a crux is that I suspect Guzey faced no retaliation, and that really undermines his apparent fixation. My opinion would be changed if he faced actual repercussions because of the leak or because of ideas in his book.
This part sounds deeply wrong to me, but that’s probably because I’ve read more of Guzey’s work besides this one piece.
I occasionally encounter people who would have been happy to burn their youth and spend their efforts on projects that would good bets from an EA perspective. When they don’t get to for one reason or another (maybe they were being too status-seeking too soon, maybe they would have been a great chief of staff or a co-founder together with someone who was epistemically stronger, maybe they didn’t have the right support networks, etc.), it strikes me as regrettable.
I think that some of Guzey’s later work is valuable, and in particular his New Science organization looks like a good bet to take. I think he got some funding from the Survival and Flourishing Fund, which is EA adjacent.
Below is my reply to this comment and your other one.
I’m not sure this is valuable or wise for me to write all this, but it seems better to communicate.
I was sincere about when I said I didn’t understand and wanted to learn why you rated Guzey’s criticism highly. I think I learned a lot more now.
_________________________
You said “This part sounds deeply wrong to me, but that’s probably because I’ve read more of Guzey’s work besides this one piece”:
Note that I made a writing error in the relevant paragraph. It is possible this changed the meaning of my comment and this rightfully offended you. When I said:
I meant:
The first version could imply there is some “establishment” (yuck) and that these people share my negative opinion. This is incorrect, I had no other prior knowledge of opinions about Guzey and knew nothing about him.
_________________________
You said here:
This seems wise and thoughtful. I didn’t know about this.
_________________________
You made another comment:
I skimmed this, in the spirit of what you suggested. The truth is that I find reviews like this often on the internet, and I use reviews of this quality for a lot of beliefs.
If I didn’t know anything about Guzey, I would use his review to update in favor his ideas. But at the same time, I find many of Guzey’s choices in content and style different than people who successfully advance scientific arguments.
_________________________
I think that, combined with virtue and good judgement, being loyal to someone is good.
In non-EA contexts, I have tried to back up friends who need support in hostile situations.
In these situations, the truth is that I can become strategic. When being strategic, I try to minimize or even rewrite the content where they were wrong. This can lead to compromise and closure, but this needs coordination and maturity.
(deleted)
I hadn’t realised that your comment on LessWrong was your first public comment on the incident for 3 years. That is an update for me.
But also, I do find it quite strange to say nothing about the incident for years, then come back with a very long and personal (and to me, bitter-seeming) comment, deep in the middle of a lengthy and mostly-unrelated conversation about a completely different organisation.
Commenting on this post after it got nominated for review is, I agree, completely reasonable and expected. That said, your review isn’t exactly very reflective – it reads more as just another chance to rehash the same grievance in great detail. I’d expect a review of a post that generated so much in-depth discussion and argument to mention and incorporate some of that discussion and argument; yours gives the impression that the post was simply ignored, a lone voice in the wilderness. If 72 comments represents deafening silence, I don’t know what noise would look like.
[Edited to soften language.]
I find it surprising that your comment only provides one-sided considerations. As an intuition pump, consider reading this unrelated review, also by Guzey, and checking if you think it is also low quality.
The low quality of Guzey’s arguments around Doing Good Better (and his unwillingness to update in the face of strong counterarguments) substantially reduced my credence in his (similarly strong) claims about Why We Sleep, and I was confused about why so many people I know put so much credence in the latter after the former.
Here are two examples of Guzey updating in response to specific points:
https://forum.effectivealtruism.org/posts/7aqGFHirEvHTMD5w5/william-macaskill-misrepresents-much-of-the-evidence?commentId=mHnp8t97EfwrRA3vg
https://www.reddit.com/r/slatestarcodex/comments/9xluu2/william_macaskill_misrepresents_much_of_the/eby9cwz/
It depends what you mean by “updating”.
I think Guzey is very honest in these discussions (and subsequently), and is trying to engage with pushback from the community, which is laudable.
But I don’t think he’s actually changed his views to nearly the degree I would expect a well-meaning rational actor to do so, and I don’t think his views about MacAskill being a bad actor are remotely in proportion to the evidence he’s presented.
For example, relating to your first link, he still makes a big deal of the “interpretation of GiveWell cost-effectiveness estimates” angle, even though everyone (even GiveWell!) think he’s off base here.
On the second link, he has removed most PlayPump material from the current version of his essay, which suggests he has genuinely updated there. So that’s good. That said, if I found out I was as wrong about something as he originally was about PlayPumps, I hope I’d be much more willing to believe that other people might make honest errors of similar magnitude without condemning them as bad actors.
(deleted)
Is it so hard to believe reasonable people can disagree with you, for reasons other than corruption or conspiracy?
What is your credence that you’re wrong about this?
(deleted)