dvgrn

This is one of those tricky things, highly dependent on context.

If you're talking about Conway's Game of Life patterns, then "gliders" are the 5-cell spaceships that travel diagonally, and all other moving things are "spaceships" but not "gliders". If you call a Conway's Life non-glider spaceship a "glider" you'll mostly just confuse people.

But if you're talking about other CA rules -- especially rules where there isn't any 5-cell diagonal spaceship -- then "glider" is very commonly used to refer to other moving patterns.

For example, David Eppstein's "Gliders in Life-Like Cellular Automata" database was active for decades -- recording spaceships across a large rulespace, not just Conway's Life. It's an accepted generalization of the term, somewhat like saying "Xerox machine" for any old copying machine whether or not it was built by Xerox.

There are also patterns -- both stable patterns and oscillators -- that have an unbounded chain of predecessors, but still can't be glider constructed. That is, there are proven solutions to Conway's "Unique Father Problem" from over half a century ago:

  https://conwaylife.com/wiki/Unique_father_problem

To put it another way, there are certain patterns that, if you see them in the Life grid, you know they've always been there from T=0 -- because they are provably their own only predecessor.

Thanks, that's a big help! I think actually the Conway's Life community does quite regularly have "interesting things" that could fit the Show HN model.

The example I'm thinking most about is the fixed-cost 15-glider construction for absolutely anything that's glider-constructible at all --

for which we do have some eminently runnable code that showcases the entire process of 15-glider construction from beginning to end, with subtitles ...

  https://conwaylife.com/forums/viewtopic.php?p=153609#p153609

... running inside Golly, but that's a free download, and the Lua script version doesn't need any extra Python installation or configuration or anything.

So if I don't get too much "Nah don't do that" feedback here, I might try putting up a Show HN post for the new super-optimized RCT15 project, once the last pieces of that get completed.

People have definitely tried this kind of thing, but so far -- from what I've seen -- Game of Life problems seem to be highly resistant to neural-net types of solutions.

Take spaceships, for example. You can train a neural net to recognize spaceships, but there aren't any reliably recognizable features that can distinguish a spaceship from a non-spaceship. To find out if a never-before-seen pattern is a spaceship with period N, you really have to run it for N ticks and see if you get the same pattern back again at an offset. Visual similarity with other spaceships just plain isn't relevant, unless the similarity is 100%; a pattern with a 99% match on a 100-cell spaceship will almost always be ... not a spaceship at all.

A good analogy for this might be training a neural net against images of prime numbers up to 997, printed in decimal in some standard font. Sure, you can train a neural net to recognize prime numbers less than 1000, with great accuracy ... but primality isn't a visual property of a printed number, it's something that you have to do some mathematical tests to find out about.

So if you try your trained neural net on prime numbers above 1000, you're going to be rather disappointed with its performance. CA spaceship recognition is the same kind of problem... possibly worse, since you could at least have some hope of a neural net correctly recognizing non-primes by their last digits.

I took a day off of work yesterday to try this AMA idea out, and I'm thinking it turned out pretty well all in all. It was definitely a one-time experiment, though, not something I'll do again.

Something just occurred to me from a review of recent HN item titles: maybe what I _will_ try sometime is a "Show HN" post, for some shiny new Life discovery that seems particularly interesting.

Heh, I'm not actually sure whether Nathaniel meant to invoke Eric Raymond's hacker emblem on the book cover or not. The grid is there, and the glider orientation is right, but the cells are squares and not circles. That orientation of the glider is kinda canonical, independent of the hacker emblem -- e.g., it's the phase that shows up in the "glider" LifeWiki article.

I emailed back and forth a little bit with Eric Raymond when the hacker-emblem proposal first came out, but I don't remember that I had anything very interesting to say. Mostly I was hoping to get the Life Lexicon factoid about the unix oscillator into the "Anticipations" section on the official Hacker Emblem page --

Unix: ... The name derives from the fact that it was for some time the mascot of the Unix lab of the mathematics faculty at the University of Waterloo.

I'm thinking the most successful "automated exploration" so far has been Catagolue's method: simply generate a whole lot of random-soup initial configurations, run them until they settle, and then poke through the ashes looking for interesting stuff:

  https://catagolue.hatsya.com/census

Seems like that gets the most emergent-behavior bang for your buck. All the other "automated search techniques" that I can think of are too specifically tailored to some particular problem.

There's definitely plenty of material for a Volume II book (and several volumes after that, for that matter). At the moment, though, Nathaniel and I think that it might be somebody else's turn to write those!

  https://conwaylife.com/forums/viewtopic.php?p=136037#p136037

Definitely check out this blog series by Eric Lippert in the meantime, though:

  https://conwaylife.com/wiki/Tutorials/Coding_Life_simulators

My sense is that people don't usually play the Immigration Game for very long -- it just doesn't seem all that interesting to most folks ... and so there hasn't been much interest in developing a computer opponent for the game.

It seems to be rather difficult to convert cellular automata into any kind of playable game. If it's an arcade game then it's usually too arbitrary, and if it's a puzzle game then it's usually way too easy or way too difficult. There have been some good efforts, but they're mostly only playable by dedicated Lifenthusiasts, and that's ... well... not a very large market!

Re: the LIFELINE public service announcement -- no need to do the scanning and online-ing. That's been done already, though there's still some review and typing-up work left for someone to do:

  https://conwaylife.com/wiki/Category:Lifeline_issues

I don't want to say that there's no possible role for AI in Life research, but it's hard to see how something like ChatGPT can be helpful.

1) The placement of a single cell in a huge pattern will very often make the difference between a working Life pattern and something that catastrophically implodes. So making a generative AI like ChatGPT do any work on Conway's Life is very much like making it play chess: sooner rather than later, something really important will end up slightly out of place, and ChatGPT will have no way of knowing.

2) Unlike a lot of other subjects where ChatGPT really shines, Conway's Life is an incredibly niche subject. There simply isn't anywhere near enough training data for ChatGPT to give reliable results, even for fairly basic questions:

  https://conwaylife.com/forums/viewtopic.php?p=183306#p183306

3) However, there are definitely a number of areas of Life research where other types of AI might end up coming in very handy -- e.g., in monitoring and tuning parameters for very long-running and difficult searches. For this we need something much less like ChatGPT and more like Douglas Lenat's EURISKO, to try new experiments and learn what it can from the results ... EURISKO also happened to come up on Hacker News today:

  https://news.ycombinator.com/item?id=40128285

We just can't rely on generative AI to re-shuffle what is already known and make it into a nice new package, when what we're searching for is something that's never been seen before.

-- It's interesting to look at the different symmetries on Catagolue, where the idea is to generate a lot of random soups with different symmetry types, and then just run those patterns and see what comes out. The D8_1 and D8_4 symmetries are 8-way symmetric, which is as high as you're going to be able to get on a square grid.

  https://catagolue.hatsya.com/census/b3s23/D8_1
  https://catagolue.hatsya.com/census/b3s23/D8_4

Scroll down to the bottom of those pages and click on, especially, some of the higher-period "xq{N}" categories. These are objects that showed up "naturally", evolving from random soups.

-- There have definitely been a number of people over the years exploring various outer-totalistic rules on a hex grid, and (to a lesser extent) isotropic non-totalistic rules: see

  https://conwaylife.com/wiki/Hexagonal_neighbourhood

-- The smallest period at which gliders can follow one another is period 14. We don't have a true period-14 gun yet, though. The closest we have is a "pseudo-period" gun -- actually period 28, but it generates two gliders per period, so you end up with a period-14 stream:

  https://catagolue.hatsya.com/object/gun_14/b3s23