Or if you don't want to have to download anything, both copy.sh/life and Chris Rowett's LifeViewer are awesome, in different ways.

The first link goes to a complete self-constructing spaceship, the Speed Demonoid -- copy.sh/life is an online HashLife implementation that's fast enough to simulate the whole thing in a browser.

The second link shows a LifeViewer "Exploratorium" display of a number of Life discoveries, old and new. Point of interest #20 shows the new doo-dah spaceship, which is another of the discoveries from 2020 that the article mentions. LifeViewer also handles most of the animations in LifeWiki articles, and on the conwaylife.com forums.

There's an ongoing distributed project, "Catagolue" (not a misspelling, think about it...) to census the results from random 16x16 seed patterns. It's very close now to censusing a quadrillion objects, and it has indeed found a large number of very unlikely things, such as loafers and Schick engines appearing spontaneously from random asymmetric soups.

It would take many, many orders of magnitude more soup-searching before anything rivalling the most impressive "engineered" patterns is likely to appear, though.

It's all part of the editors' nefarious plan to make you go and read the actual article, no doubt. The latter part is a series of quotes from people like Bill Gosper, Melanie Mitchell, Stephen Wolfram, Susan Stepney, and Rudy Rucker. Hard to summarize, at least if you're trying to go a little deeper than "something something emergent complexity"...!

Definitely worth a watch, especially for anyone not familiar with Conway's Life. The link is to Alan Zucconi's amazing short documentary from early October -- there's a reference to it in the article's title animation. I really liked the in-video Life simulations starting from Bob Wainwright's old typewritten LIFELINE newsletters.

Yup, the proof is the hard part. It's definitely possible to design a universal-constructor-based pattern that almost certainly remains chaotic indefinitely. For example, build a fast quadratic-growth replicator using Speed Demonoid technology, but don't put in any collision suppression.

Descendants of the original replicator will crash into each other with horrific results in the crowded center, making a chaotic mess of ever-increasing diameter, while the replicators at the ever-expanding frontier will probably be able to outrun the mess behind them indefinitely. But turning that vague hand-waving into an airtight proof is a very tall order.

The Speed Demonoid's design is simple enough that online simulators can do a pretty good job of displaying them in action:

c/5 -- https://copy.sh/life/?pattern=...
c/8 -- https://copy.sh/life/?pattern=...
c/137 -- https://copy.sh/life/?pattern=...
3c/14 glider synthesis produced by the generator script -- http://copy.sh/life/?pattern=s...

I was hoping someone would ask that. Let me start out with a comparison to other cellular automata. Conway's Life is B3/S23 -- "born if 3 neighbors, survives if 2 or 3 neighbors". There are other rules, such as HighLife (B36/S23, very close to Conway's Life) in which a 12-cell pattern can replicate itself -- after 12 generations there are 2 copies, after 36 ticks there are 4 copies, and so on. This pattern regularly evolves from random starting states.

There's even a rule, Fredkin's parity rule (B1357/S1257) where every possible pattern is a replicator -- an extreme example of replication being "just the way the universe works". But these replicators are, in some sense, too simple to be very interesting! They replicate the way crystals grow, and it's hard to harness that kind of low-level behavior. If you wanted a HighLife replicator with 13 cells, or one that would replicate in 13 ticks, instead of 12, you'd be out of luck. By comparison, the Gemini spaceship is extraordinarily adjustable.

No to all of the above. Conway's Life is not amenable to error-correction of this kind, because small changes have such huge consequences. Kind of like building machinery out of chunks of sub-critical enriched uranium: you can design it so that during normal operation the various pieces never come close enough together to start a chain reaction, but if any little thing goes wrong, you end up with high-energy particles flying all over the place, spreading the reaction to other nearby machinery, which then contributes to the explosion.

Well, I didn't say "in spite of the things around it" -- it was "in spite of the way the universe normally works." The Life universe, for random patterns anyway, normally settles into a scattering of stable or P2-oscillating ash after a few hundred or a few thousand generations. There are any number of "lucky" self-perpetuating stationary and moving patterns that are exceptions to this general rule, but they're all very delicately balanced on the edge of chaos.

The Gemini spaceship contains a large amount of data in its glider channels that is recognizably information about its own structure. Change that data, and the replicator unit will (usually) build something different. Most other gliders and spaceships in Conway's Life don't have anything like this -- all the other hundreds of patterns in Golly's Spaceships folder, or the tens of thousands in Koenig's Life Object database, are "naturally" self-perpetuating, because a future generation of the pattern happens to be identical to the original.

The Gemini spaceship has a significantly higher degree of control over its environment: with the right change to its program, a Gemini replicator unit could construct anything that can be built by colliding gliders, in any empty space in the Life universe. The Gemini contains most or all of the construction tools that a Conway's Life self-replicator will need; it's just a few short steps away from being a true replicator. Mostly it just doesn't have the right program -- yet.

There are a few other large patterns, especially Gabriel Nivasch's Caterpillar, that blur this line to some extent. However, the pi-climber "data" in the Caterpillar is much more difficult to reprogram than the gliders in the Gemini. Several new variants of the Gemini with different speeds and angles of travel have already been built -- with a lot of help from the Python scripts that Andrew Wade made available along with the pattern..

This is a tricky point. The people who say that this new pattern is not ultimately different from a glider are correct, in a sense -- the Gemini spaceship is technically a spaceship, not a replicator.

It _does_ make two copies -- but the copies are of the two replicator units at the ends of the glider channels, not of the entire spaceship.

But replicator units replicating themselves, even with the help of an outside stream of instructions (which is then reflected on to the next newly-created copy of a replicator unit) are still something that hasn't been seen before in the Life universe. So this is a much more impressive technical accomplishment than, say, finding a new variety of spaceship using a search utility.

Gliders and spaceships "replicate" themselves in somewhat the same way that salt crystals or wildfires do -- that's just the way the universe works, you might say. But the Gemini pattern keeps itself going by continuously reconstructing itself, in *spite* of the way the universe normally works.

The replicator units are like robots that include all the tools needed to make more robots exactly like themselves -- but they're radio-controlled and have no internal memory, so you have to pipe the actual construction recipe in from somewhere else. That means they're not self-contained self-replicators, true -- but they're a darn sight closer than a salt crystal or a glider!

Eventually someone will build a pattern with an internal memory that can hold a complete self-construction recipe -- but the Gemini is an important milestone along the way to that goal, and the first true Life replicator will probably contain ideas taken from the Gemini, just as the Gemini contains ideas and mechanisms from preceding patterns.

All of these patterns -- the Deep Cell, a 512x512 version of the original Unit Life Cell, and several samples of the OTCA metapixel -- are included in the pattern collection that comes with Golly. Check the "HashLife" folder for the metapixel examples, and the "Signal-Circuitry" folder for the others.

There's also a "metafier.py" Python script that lets you convert any selected pattern into an OTCA-metapixel pattern, and then run the results. Shouldn't be attempted on too large a starting pattern -- meta-metacells are a bit too much to ask for, and meta-Gemini-spaceships won't finish building in your lifetime! -- but it's surprising how big a metapattern Golly can handle, once it's been constructed.