Adopt a directed evolutionary approach

The current evolution algorithm of organisms is completely random. But we know that each entry into the editor represents 100 million years of evolution. During this time, natural selection should have eliminated many unfavorable evolutions, so a directed evolutionary approach should be adopted to AI evolution.

Currently, no matter how AI evolves, after entering the editor more than 30 times, the other species are basically wiped out by the player. After that, the player can only rely on finding free phosphate clouds to replicate. The process is very long and boring. Through this method, the competitiveness of AI will be greatly improved, and the game will be more interesting.

This can be achieved by hard-coding or encoding biological evolution paths with json. AI, like the player, gains 100 evolution points every 100m years. For example, for plant cells,

{
    "name": "plant",
    "path": [
        {
            "modifications": ["chloroplast", "mitochondrion"],
            "weight": 1,
            "condition": [[]]
        },
        {
            "modifications": ["mitochondrion", "mitochondrion"],
            "weight": 0.8,
            "condition": [["CompundBalance", "Glocose", "nl", "0.1"]]
        },
        {
            "modifications": ["pilus", "pilus", "pilus"],
            "weight": 0.5,
            "condition": [["ATPBalance", "nl", "10"]]
        },
        {...}
    ]
}

achieves continuous growth in size and improved defense capabilities.

An alternate solution to the lack of competitiveness is ‘simply’ to improve Auto-Evo. This is much more interesting and adaptative, in my mind, than predefining evolution directions, which will too heavily rely on life as we know it.

(Also, Thim is already working on such improvements)

2 Likes

That is caused by the player species not being affected by auto-evo at all. Other species can’t compete when it is literally hard coded so that auto-evo numbers don’t affect the player’s population at all:

edit: Also regarding the evolution paths, I’d just much rather increase the number of attempted mutations, that way the species are much more likely to find a really good mutation, without all getting stuck evolving in the same direction.

2 Likes

I have been under the assumption that a system of evolution by natural selection was a non-negotiable on Thrive; that even if we want the auto-evo to do something specific, it needs to be able to figure it out by itself, and I think that there’s some reasoning behind that. Emergent systems can do some pretty cool things that you can’t think of in planning.

If that’s not the case though, we can make things quite simple. There are probably about 2 dozen meaningfully different “kinds” of microbe designs that make sense, and if we wanted to we could just consider those the targets that would show up on every planet, and have the auto-evo just move the closest species to each of those. The fact that new species branch off from what the player has chosen to evolve means that the algorithm can’t be completely on rails, and would mean that there would always be a few species running around that are different every game (if only temporarily).

The biggest downside to that kind of algorithm is probably just the fact that we can’t put “contains real evolution” on the box if we deviate this far from Darwinism. Speaking as the guy who is currently working on trying to make the evolution model more accurate, I doubt a fully scientific algorithm would actually produce different results than the idea suggested here. In later stages, however, the game might become complex enough with physics systems and social behaviors that this starts to not be viable.

3 Likes