I recently got an email about Crescent Loom asking about opening it up with an API or something to fine-tune the parameter space of its bodies & neurons, and I put enough thought into writing a response I thought it'd be worth sharing here too:
In short, as a biologist, I've found myself more interested in making a game about intelligent design than evolution (lol).
My thoughts have evolved somewhat since the initial "scope" issue — my party line for years has been that I'm making this thing in order to let people get their hands into the guts of biological nervous systems, not to let them press a button and have the computer give them a funny animal. Crescent Loom as a game already struggles with being too close to being a fishbowl screensaver maker (you make your little guy… and then what?) and trying to automate more of the creation process only worsens that problem.
I also think that "evolution" games that use genetic algorithms as their primary mechanic are honeypots that trap developers working in this field but never produce compelling gameplay because of a fundamentally cursed problem that the most interesting thing the program is doing is not directly visible to the playe. "It's getting better at doing stuff? I guess?" — it's a fun mechanic to program, not play. And weirdly people almost always only think of doing it for biology-themed games, not ones like Kerbal that are doing the same damn thing but the idea of evolution isn't as close at hand (though there's been some cool demos done for driving games).
But I hear where the idea is coming from that searching the parameter space is not a fun process, and the story that "centaurs" of humans running things with a computer taking care of the details outperforming either working alone is an alluring one. Getting an open API with CL handling the UI of weaving a nervous system and allowing it to be modified or plugged into whatever you want would open up a lotta possibilities — genetic algos, sure, but also stuff like CL-made networks driving robots or something.
And if you had emailed me like two months ago, that's where my email would have ended, but I recently connected with someone who's done basically that: check out FEAGI and Neurorobotics. Mohammad's been working on a very much more implementation-agnostic neural-net-genetic-algorithm series of projects. Definitely less "pick up and play" than CL, but it's about as close to that open API idea that I think we're ever going to see. He's doing it better than I could ever do with CL, so it's kind of nice to be able to say that that dream's taken care of so I can focus on education & accessibility rather than making it a general-purpose tool.
