Herpers, I’m happy to believe you but I don’t understand. Presumably every organism after the second generation has ancestors = $gen-1?

My run’s seemingly hit a local maximum at basically the simplest mobile form you could imagine, a square body with a rectangular head that waves back and forth to drag and/or wheelie-hop through rotational inertia.

And it reminds me of a genetic algorithm to make the best bicycle.

I’ve personally played around a bit with genetic algorithms. One to create a shortest route (travelling salesmen problem) and one total rip off of the EvoLisa project.

Huh? If the real world was anything like this game, the vast majority of organisms that we observe would be dysfunctional cripples. It’s obviously just a fantasy.

This reminds me of fooling around with the Neuralbot in Quake 2, back in the day. That had a number of bots you could set (according to performance) each with its own evolving neural network governing behaviour. The behavioural loadings were all random at first. You could run around and execute the particularly stupid if you wanted to as well.
It was bizarrely fun to fidget around with. It took a long time to get truly notable results though (not just stuff you thought was intelligent). It turned out that, despite having proper fields of view and being able to utilise them (theoretically anyway. It was a complex neural network by default and you could even expand on the number of neurons and synapses if you wanted to) the bots couldn’t really cope with up and down, or behaviours that dealt with that were never passed on. The best breeding grounds were big, flat levels all on the same plane with some obstacles (on more complex levels you’d get all sorts of amusing things like one I did where the level was a bit overcrowded and the bots often spawned in the same place and some evolved the great skill of grabbing the grenades and holding them until they exploded, killing others around them). On those flat levels you really saw results. One guy had demos from various generations showing the bots at the start randomly jumping around and firing like some demented dance party. Then a few thousand later they’re running around perfectly, zipping between obstacles collecting weapons, killing each other (they favoured hitscan weapons for fairly obvious reasons). It’s quite cool. You could also dig through the bot files and inspect the neural nets youself and edit them if you wanted to. The failures in bot behaviours were illustrative of just how many variables you need to get good complex creatures in the world (I forget exactly how many neurons these things generally had, but it was a lot and it was plainly not enough) and also the necessity of hard coded behaviours in order to survive and clearly separate, demarcated systems for handling the various kinds of stimulus (as an open system can too easily not pass on good behaviours).
Anyway, I agree there’s a lot of room for this stuff in gaming. Some sort of Spore-ish spin off where you breed your own creature and then make them fight (or even live) with others on the net or something. Endlessly amusing.