Hacker News Reader

Comment by 6ren | Hacker News Reader

Adjacent cells might self-select, but you can adjust the probability of self-selection to lower this risk as much as you like. Once selected, the cell owns the adjacent cells, by inhibiting their self-selection. (Maybe the inhibitor spreads further over time?). As self-selection continues, any gaps are filled in.

It reminds me of git's hashes. There's no absolute guarantee that different objects will have different hashes (since there are more possible objects than hashes), but it's pretty good in practice (and I guess/hope git has a secondary check for collisions). I really don't like these kinds of algorithms - they are just wrong - but I have to admit they work pretty well, by taking a likely guess instead of working it out exactly. eg ethernet also uses randomnness.

View on HN

No, git doesn't have a secondary check: http://progit.org/book/ch6-1.html § A SHORT NOTE ABOUT SHA-1

This claim of wrongness is a pretty large claim. Care to back it up? (e.g. define wrong)

From context, I'd guess that wrong means "nondeterministic".

Or, inductive versus deductive.

I agree that a clearer definition of wrong here. My guess is that it has to do with the solution being empirical and not-guaranteed, rather than analytic and airtight.

Just because the solution isn't perfectly analytic doesn't mean it's wrong, though. In fact, it seems hard to argue that the solutions which work well in natural systems are wrong.

Really, we need a lot more research like this that takes well-working natural systems and distills the underlying model. It seems a bit arrogant to try and independently solve problems when existing solutions are sitting all around us.

It's not meant to be a claim at all: the italics were meant to convey a sense of wrongness, that it feels wrong, not that it's actually wrong. Like noticing that a dog's "knee" bends the "wrong" way (that joint actually corresponds to our ankle). Or like fractions of bits, as in Arithmetic Coding. It's just wrong. How can you have a fraction of a bit!? Ridiculous! (note: my masters involved arithmetic coding; I actually do understand it). Another one is an existence proof, as opposed to a constructive proof (ie. where you demonstrate that something must exist, without actually constructing it).

joeyo's right: non-deterministic. It doesn't actually nail the problem, it just "usually" works out "OK". Although, you can prove things about it, like how often it works, and how well. And that's very useful. Actually, I think it's to do with the algorithm not embodying a perfect solution, but rather a method for arriving at the solution. It's an engineering "how-to" hack, rather than a science "what-is". EDIT ie. a heuristic

Note my explanation of the algorithm, including a way to "lower this risk as much as you like", which is saying that it does work, within tolerances. Any interpretation of "wrong" could take that context into account.

Well, that is much clearer, thanks :)

Reply on news.ycombinator.com