Partial retraction of 'Misinformation on Marginal Revolution'
o3 still not reliable though.
Tyler Cowen wrote a post titled Berthold and Emanuel Lasker, saying that ‘In the 1880s (!) he [Berthold] developed what later was called “Fischer Random” chess’, with the link going to ChatGPT (o3).
I replied in a post called Misinformation on Marginal Revolution, calling this claim a hallucination, with o3’s sources either not being specified or being hallucinated.
Cowen replied to me on Twitter, saying ‘Nice try, but it is all confirmed in volume three of the Lasker biography. Good luck finding a copy of that book, but I have mine at home.’
Having now acquired a copy of that book:
I retract the title ‘Misinformation on Marginal Revolution’.
I retract ‘The claim is hallucinated. It is irritating that Cowen shared it as though it is true.’
I retract ‘the wrong information would not have been introduced to the public at all if not for Cowen uncritically accepting it as fact and actively sharing it.’
I retract the reference to ‘Cowen’s gullibility and poisoning of the information environment’.
I would still have written a post on the subject, which would still have been scathing about the o3 text: I stand by my analysis of its output, which is full of hallucinations. The comparison between o3 and GPT-5 is still interesting, albeit in a less rhetorically forceful way. If I knew then what I know now, the framing of my post would have been about citation and hyperlinking practices. There was misinformation linked to on Marginal Revolution, but the vast majority of readers would not have clicked the link.
Here is the relevant passage on page 13 of Emanuel Lasker Volume III: Labors and Legacy: Chess, Philosophy, and Psychology, edited by Richard Forster, Michael Negele, and Raj Tischbierek. It’s at the end of a section about Lasker’s opinions on how to counteract an increasing number of draws in top-level chess:1
In later years, at Nottingham 1936, Lasker advocated the random setup of pieces, a proposal which has gained traction in recent years. The idea, which is over two hundred years old, had also been popular in the Berlin Café Royal in the 1880s, championed by Lasker’s brother Berthold (DSZ, October 1936, p. 289).
This is close enough to what Cowen wrote.2
A key error I made, which I do not regret, was assuming that Cowen linked to ChatGPT because that’s where he got his information from. The three-volume Emanuel Lasker biography is meticulously well-researched, the result of many hundreds or thousands of hours sifting through archival sources. I infer that Cowen supplemented his reading of Volume III with o3, and then linked only to the latter without mentioning the former.
This is bad practice! (And honestly quite bewildering to me, hence my non-regret.) It is bad for the practical reason that the o3 text is littered with errors and references that do not exist. But it is also bad for moral reasons: I think that the chess historians who brought this interesting nugget about Berthold Lasker to readers in the 21st century should be credited for that work.
An error that I do regret is not caveating my remarks with the acknowledgement that even though none of o3’s sources supported its claims, it was nevertheless possible that Berthold Lasker had proposed shuffle chess in the 19th century. I had such a sentence in my first draft, but deleted it as my irritation increased to make the post punchier. I am usually more careful.
The o3 text mentions shuffle chess being ‘tried informally in Berlin cafés’, which sounds like an echo of Volume III saying that it was popular at Berlin’s Café Royal. I still don’t think Cowen should have linked to o3 here, but I have a little sympathy if this phrase led him to think that the o3 text was generally accurate. The alarm bells that I mentioned in my earlier post wouldn’t have been ringing so strongly on a first read.
Two hints could have suggested to me that there was more to Cowen’s post than an o3 chat. One was wondering why Cowen had asked in the first place if Berthold had had the idea for randomised piece positions. The other is that Cowen’s post says ‘the 1880s’, while the o3 text says ‘c. 1890-1892’, a discrepancy that I didn’t notice until checking the book.
It turns out that if I had been slightly more thorough in my checking of sources, then I could have discovered that Volume III contained a passage relevant to Berthold Lasker and randomised pieces. The biography (mistitled) had been mentioned in the o3 summary reasoning trace, and I had found an excerpt of one of the first two volumes online. The excerpt didn’t support the o3 claim, but since it was only an excerpt, and since o3’s reasoning at that point was about 1936 rather than the 19th century, I neither pursued the lead nor mentioned it in my post.
If I had found the publisher’s previews of all three volumes, however, then I would have found that the third volume’s index contains an entry ‘Lasker, Berthold (brother): … random piece setup, advocate of, 13’. Then I would have ordered the book and eventually written a post with something of the framing that I now wish I’d used. Alas!
The ‘DSZ’ cited at the end is Deutsche Schachzeitung, the 1936 editions of which are not available online.
I still have a somewhat substantial quibble over the verb ‘developed’, when it’s clear that Berthold wasn’t the first to suggest randomising the pieces, and it’s unclear if he came up with the idea independently or instead took a liking to it after hearing it from someone else. This quibble would not be worthy of a reply post on its own, only a comment on the original post.
(I also have a minor nitpick about calling this ‘Fischer Random’ when it didn’t necessarily have Fischer’s castling rules and constraints on the randomisation; this isn’t even worthy of a comment.)
The main point is that it’s interesting that one of the world’s strongest players in the late 19th century ‘championed’ starting the game with shuffled pieces, and imperfectly highlighting this does not come close to something that I’d aggressively call misinformation.
Kudos for correcting!
tyler's response is still pretty douchey but good to have this follow up