Hacker News Reader

Comment by kevinpet | Hacker News Reader

You can't exactly use "digital signatures" and "cryptographic proof" as distinguishing characteristics among crypto researchers. Many of the phrases noted are similar overly verbose academic writing. "It should be noted" at 5% and "can be characterized" at 1% do not add up to a 0.05% chance of non-Nick Szabo writing the paper.

Even if we agreed with the statistics, the "one in a thousand" would be relevant if we had some other reason to suspect Szabo. He could simply be that one in a thousand who matches.

The article is somewhat persuasive, but more certain than the evidence presented seems to justify.

View on HN

A paper gives you dozens of phrases through which you could fish for the ones that point towards your desired author, if you wanted. It's sort of like how given a list of my inheritable traits and two potential fathers, you could make a list with a few traits that match with either one of them. If the list is long enough (say you have thousands of SNPs to root through), you might even be able to find a few _rare_ traits that match.

[Edit: speaking of which, interesting to see that a past attempt to out Satoshi claimed a combination of textual and other evidence pointed at a particular group of researchers who filed a patent right before bitcoin.org was bought: https://en.bitcoin.it/wiki/Satoshi_Nakamoto]

The poster's suggesting they ran an automated analysis against a bunch of authors, and that it clearly pointed to Nick. Particularly if _still_ it points that way even if restricted to content-neutral phrases (so it effectively picked Nick out of all crypto researchers, not just out of those that talk about timestamp servers etc.), that's more persuasive: programs won't fish (though you could still repeatedly change the algorithm until it gets the result you want). Publishing it would be interesting.

As is, this is interesting but seems preliminary or shallow--doesn't deal with any other text Satoshi or Nick wrote, doesn't deal with code, and doesn't give us full raw data or scripts. I'm not sure Satoshi is Nick Szabo.

Also, the post-dating had occurred by the Oct 17, 2009 Wayback Machine crawl, but the post content didn't visibly change, only the date. I have no idea if it was anything interesting or not.

Publishing the program would be interesting.

And persuasive, if it's all as suggested.

Ha, I don't think this has any value as evidence, but it's kind of funny that in a comment on one of the eventually-postdated blog posts (originally April 2008), Nick explained that the mechanics of the bit gold market would be nonobvious to most people and therefore the idea could use a demonstration project, and asked, "Anybody want to help me code one up?"

http://unenumerated.blogspot.com/2008/04/bit-gold-markets.ht...

That's evidence against Satoshi=Szabo; IIRC, somewhere Satoshi says he started coding up the prototype in 2007, so if Szabo is suggesting someone write it in 2008...

Or it could be intentional misdirection, as with "favour". Not that I'm particularly invested in Satoshi==Szabo, but if I were interested in deflecting suspicion, that would be one obvious-looking way to do it.

If I saw strong evidence Szabo wrote some of Satoshi's words (and I haven't yet!), then maybe someone took him up in Apr 2008, they pretty quickly refined some things and got 0.1 out the door, Szabo contributed some text, and they changed the post date to after Bitcoin's release/threw in some British spellings/made incorrect claims about when work started [or the implementer really had started earlier, or something] to add noise for anyone trying to trace Satoshi back to actual people.

Still, that all kind of assumes there was 11-dimensional chess going on to conceal who worked on Bitcoin, and I think the reality was likely simpler.

However, that scenario would allow for some of Szabo's post-Bitcoin statements to be sincere even if he was involved: could be true that he hasn't audited the code and that he's not entirely sure if certain bit gold -> Bitcoin changes are features or bugs, since any refiner/implementer presumably had some independence and collaborators never agree on everything. And Szabo being involved is at least a saner notion than previous ones: we at least know he found proof-of-work e-gold interesting.

But I'm speaking without having dug deeply into what they both wrote, which would probably probably a prereq for developing useful priors. I don't think I know much here, and I certainly don't think the comment alone is evidence for Satoshi=Szabo (or Satoshi-involved-Szabo).

Another thought 'cause I can't stop myself: if Satoshi is a collaboration, anyone outing themselves as part of it is making it harder for others to keep their roles secret. So if Nick + friend(s) were Satoshi, you couldn't well expect him to talk about it.

>>Even if we agreed with the statistics, the "one in a thousand" would be relevant if we had some other reason to suspect Szabo. He could simply be that one in a thousand who matches.

Except we aren't dealing with a massive amount of people who may use those phrases just a couple of people. If we already have reason to suspect something and we are given independent persuasive evidence this should be taken together to make it very unlikely that our hypothesis is not true.

The great thing about textual stylistic analysis is that everyone has a specific way of communicating that can be identified unless one takes steps to actively hide it. This happens in all languages and can usually identify authorship regardless if someone tried to copy someone else's style.

> "The great thing about textual stylistic analysis is that everyone has a specific way of communicating that can be identified "

How scientifically well supported is this assertion?

http://bits.blogs.nytimes.com/2012/01/03/software-helps-iden...

Stylometry's a real thing. It gets harder the shorter the sample is, and it's much easier to say "this text matches X's verbal tics better than it matches Y's" than to confidently point at a specific person like this post is doing. Fishing for a similarity measure that supports your desired conclusion is a risk. Identification from style (phrases, quirks, vocab size) is a thing, though.

The program or more-raw data would be interesting.

I agree there the two phrases are not independent, but the joint probability of both events is mathematically guaranteed to be lower than the probability either event alone. As we add more phrases (events) to this distribution, it becomes exponentially less likely to be someone else.

Guaranteed to be less than or equal to.

* lower or equal

The problem is also a sample bias I think. He clearly hand picked low probability words as evidence.

In other words, the probability that there exists a subset of words in a pair of papers that are "globally rare" among cryptography papers is likely high.

So to answer P(Nick|Evidence) we can't use ~P(selected subset|crypto), we better use ~P(all words|crypto) and compare that to all other plausible sources.

As a clarifying anecdote, remember low probability events happen all the time if the sample space is simply large. Throw a dice ~8 times in a row an the probability of that event is lower than winning the lottery or getting hit by lightning, yet it happens every time. If you pick every word from the source paper and multiply their relative frequencies out, you will get a number astronomically low, and that happens for any paper.

personally, this kind of thing seems obvious. i would be surprised if the person who wrote the paper didn't take cute things like this into account

Reply on news.ycombinator.com