FWIW, I don't believe at all that it's an AI, it's rather an algorithm with a ruleset that determines how the thargoids expand. This means that Frontier doesn't have to manually decide which system the thargoids will attack, it's automatic and follows some predefined rules.
Thargoid War: The Beginning of the End!
- Thread starter Seljdon
- Start date
They implemented an AI, but probably not a machine learning algorithm. In computer games, "AI" has generally referred to static sets of rules, state machines, and stimulus responses to give NPCs behaviours, not the more specific modern marketing meaning. Machine learning algorithms need huge amounts of input data to stop them taking completely random and nonsensical actions, which one "how well did this work" a week won't get in any useful amount of time.
It's detectably changed its behaviour twice since initial implementation, both of which have been reductions in its aggression, and both of which were from Frontier making those changes.
We figured out that bit: yes, but not particularly well. Each Maelstrom gets 20 points a week to place Alerts, which must be placed within 10 LY of an existing Control system (plus some other restrictions).
- prior to 2nd Feb: inhabited Alerts cost 4 points, uninhabited Alerts cost 1, the Maelstrom picks fairly randomly among the systems in range giving a tendril-like branching pattern of reasonable rapid but sparse expansion. Expansion "away" from the bubble takes place as much as expansion into it.
- 2nd Feb to 9 March: the Maelstroms switch to incredibly strongly prioritising the systems closest to themselves, giving a solid-sphere like expansion, which is extremely slow. The direction (or rather lack thereof) doesn't change
- 16 March onwards: the cost for uninhabited Alerts is increased to also be 4. This decreases the number of both inhabited and uninhabited alerts substantially (the increased cost of uninhabited alerts doesn't seem to make it less likely to pick them in the first place). Target selection on the substantially reduced budget is unchanged.
It was only winning before because it could place about 80 Alerts a week and we could take 10 systems back - brute force, rather than strategy. Now it places 40 and we win about 25 back, which is technically still an eventual Thargoid victory but not one that's going to happen this century, especially given how many of those 40 are strategically irrelevant.