A Tiny Bit of AI Has Created a Big Problem for an Indie Game Darling

For game developer teams, the Independent Games Festival (IGF) is a very big deal. A nomination for the IGF Awards provides a game with priceless publicity, including a showcase at the festival, and a shot at winning one of the industry’s most prestigious independent gaming honors. The Awards are judged by panels of industry representatives—and this year, 19 such jurors have issued a statement declaring that they will not vote for “any game that uses generative AI in any capacity.” IGF uses different jurors for every award—last year’s festival involved six awards and a total of 46 jurors, and we’d guess there’ll be roughly the same number. (In a blog post published yesterday, developer Austin Walker, who is one of the signatories to the statement, says that there are “50 or so” jurors.) So 19 is not necessarily a majority, although it’s not clear which jurors are on which panels yet, and the statement also notes that other jurors support the stance without having gone public, due to “concern about their ability to safely sign without professional blowback.” The 19 signatories don’t necessarily represent a majority of IGF jurors. Developer Austin Walker, one of the signatories, estimated in a blog post on Tuesday that roughly 50 people serve on this year’s juries. Their distribution across the various award categories is also unclear, making it difficult to assess how much influence the pledge will have on the outcome. The statement also notes that other jurors support the stance without having gone public, due to “concern about their ability to safely sign without professional blowback.” If you’re not a fan of generative AI, your instinctive response to what’s effectively a blanket refusal to reward games made using AI is probably to applaud wildly. But even blankets have edges, and sure enough, not long after the statement was issued, it became clear that Big Walk, a new game by House House—the developers behind Untitled Goose Game—would fall foul of these criteria for a use of AI that… well, judge for yourself. According to a statement published on the House House website, “the primary use case for [the] use of LLMs [in the development of Big Walk] was to research, via chatbots, issues we encountered with third-party code.” The statement provides the example of a problem with Unity’s culling of shadow casting, which neither their own programmers nor a Unity engineer could solve: “We posed the question to an LLM chatbot, [which] suggested a possible fix, and while we had to modify the suggestion extensively, it worked.” House House estimates that well under 0.1% of Big Walk’s code was connected to its use of LLMs, and says generative AI was not used in any other capacity during development. The studio also acknowledged that it sometimes incorporated modified snippets of AI-generated code into the game. This is obviously not the same as throwing together a vibecoded game populated with AI-generated assets. But it absolutely falls under the rubric of “[using] generative AI in any capacity,” which is the criterion expressed in the jurors’ statement. (It’s worth noting that the statement states specifically that “AI-generated code is not an exception,” arguing that “programming is a creative discipline whose labor deserves the same respect we extend to any [other] part of making a game.”) Developer Sam Barlow, another of those who signed the statement, addressed the issue of Big Walk on BlueSky yesterday. He explained that while the game “was not the impetus for this discussion,” an honor that went to the “large number” of submissions making much more egregious use of generative AI, it did end up being a “tricky edge case.” House House had disclosed its AI use as part of its IGF submission, but hadn’t previously made the information public. The studio says Steam, where Big Walk is sold, does not require developers to disclose the use of chatbots or AI-generated code. So has Big Walk got a raw deal here? It’s worth noting that using ChatGPT, etc., isn’t necessarily a victimless crime—as the statement points out, chatbots’ ability to provide code depends on “past work used to train these models without the creators’ consent.” More generally, the normalization of these tools depends on the idea that their existence and use is somehow inevitable, the self-fulfilling and self-serving narrative that the late and great Mark Fisher identified as a key component of capitalist realism. “We just asked ChatGPT a question” is the thin end of a very thick wedge. To their credit, House House accepted the situation gracefully. In a statement published on its website, the company acknowledged that its lack of an internal policy on AI usage was “a failure,” and that it “support[s] the work of those who are trying to limit the harm generative AI is doing to videogames and our world.”