Stellaris launched in 2016, since when it has followed the usual Paradox post-launch cycle of regular tweaking accompanied by substantial paid-for DLC expansions. The latest is The Machine Age and focuses on “synthetic ascension” and the promise of eternal life among the stars, adding various new elements to the machine side of the game alongside a synth space queen that says she’s going to get us all there.

The DLC’s store page spells all of this out then, right at the end, drops this disclaimer:

“We employ generative AI technologies during the creation of some assets. Typically this involves the ideation of content and visual reference material. These elements represent a minor component of the overall development. AI has been used to generate voices for an AI antagonist and a player advisor.”

Well well well, say some Stellaris players, doth mine eyes deceive me? The use of AI creation tools in an expansion about AI consciousness taking over the galaxy? In particular, the use of AI to generate in-game voices raised a few eyebrows.

It’s a fair cop, says the game’s director, before going on to explain how and why the team has been using AI to help with the production of the Machine Age. Stephen ‘Eladrin’ Muray was addressing comments piecemeal in Steam discussions before taking to the game’s subreddit (spotted by RPS) to offer further reassurance to players.

“The AI voice generation tools we use on Stellaris ensure that the voice actors that signed up and built the models receive royalties for every line we create,” says Muray. “Ethical use of AI technology is very important to us—we’re pretty good at exploring dystopian sci-fi and don’t want to end up there ourselves.”

Muray promises the devs will pull together a wider post on how it uses AI in a couple of weeks, but adds “we didn’t use it for concept art in The Machine Age” because there are “a couple of awesome concept artists on staff for that.” There are “a couple of AI generated pieces on the visdev exploration/mood board” among other thematic pieces, but these are being used as reference and not in any final art assets.

The director says he uses AI image generation tools to make basic representations of ideas he’s chucking around with the game’s designers, but when things get further along “the artists then take our ideas and might or might not use them as inspiration to make final assets. None of those design images go into the game.” With generative text it’s apparently the same, lots of “ideation” but “none of the results or generated text go into the game.”

Muray ends by saying Paradox has “strict guidelines” on using “AI tools legally and ethically” which naturally “we abide by”. It would certainly be interesting, when that dev diary on how it’s being used does appear, to know more about those.

All in all, the approach the studio’s taken seems eminently reasonable. Using AI for creating image references in the concepting stage is one step up from using google image search, the more traditional method of cobbling together some sort of mood board, and as long as the AI imagery isn’t being used in the final assets it’s hard to see the problem.

The bigger issue for most seemed to be the two AI-generated voices, one of which is a player advisor and the other being the DLC’s synthetic queen Cetana. But the kink here is that the AI is creating those lines based on work by human voice actors, who are then given royalties based on their use, and that the context is as fitting as it could be: An AI voice generator voicing a fictional AI character seems like a reasonable and indeed clever use of the tech.

So I’d say Paradox comes out of this one looking fairly decent. The times in which we’re living mean that any game or expansion with this kind of AI disclaimer is going to be subject to huge scrutiny from players, who by and large remain suspicious of the unstoppable push of the tech into the development process, but for much of the industry that ship has already sailed. The question now is less about whether games made at a certain scale use AI, and more how they use it, how any outputs are flagged, and whether they’re actually replacing any of us fleshbags within the process or simply part of a toolset.