Big Walk and the Cost of Disclosure

On November 9, 19 jurors for the Independent Games Festival published a statement saying that “this year we are declining to vote for any game that uses generative AI in any capacity.” They didn’t name a game. Within hours, the discussion online had turned to Big Walk, from House House, the Melbourne studio behind Untitled Goose Game.

House House describes its use of AI in the game’s FAQ. “In the final stretch of Big Walk’s development, LLMs were used sparingly by House House to help work through difficult technical challenges and bugs.” In some cases, snippets from a chatbot’s response were copied as a starting point and then modified into game code. The studio estimates that “well under 0.1% of Big Walk’s code had anything to do with this usage,” and says that “generative AI was not used in any other context when making the game.”

The studio gives a specific example. Unity’s render pipeline couldn’t cull shadow casters by their distance from the camera, so those behind the player were still rendered no matter how far away they were. This made the game run about 50% slower for around 20 seconds at sunrise and sunset. The team researched the problem extensively, read through the pipeline’s code, and spoke to an engineer at Unity, who asked around and concluded that what they wanted wasn’t possible. They then asked a chatbot. “After a few dead ends, it suggested a possible fix, and while we had to modify the suggestion extensively, it worked.” The fix involved about 10 lines of code inside Unity’s pipeline.

Under the pledge, that is enough for the jurors who signed it to rule Big Walk out. This is a game that took six years to make and sold more than a million copies in under a week. Aftermath’s report calls it “one of the best video games of the year.” A game like that isn’t slop, and I don’t think the jurors would say it is.

Their objection is to the tools themselves, and the statement refers to “this trend and the games embracing it.” I don’t think that describes what House House disclosed. But “any capacity” excludes Big Walk along with a game whose art, dialogue, and voices were generated and published with little review.

In a previous note, I asked what “made with AI” actually tells us about a game. In this case, we’re talking about a fraction of a percent of the code.

The team had already done the research and asked the people who make the engine. When the chatbot offered suggestions, the developers were the ones who recognized the dead ends and reworked the promising idea until it fit their game. The model helped them reach a fix, but the decisions about what worked and what needed to change stayed with them. That work removed the slowdown players would otherwise have encountered at sunrise and sunset.

It’s also the kind of problem where a small team has the fewest options. A large studio may have engine programmers who can spend weeks inside a rendering pipeline. A small independent team that’s told something can’t be done can ship the slowdown, cut something, or keep looking for a solution. I wrote in that note about using AI to “attempt something beyond the resources of a small team.” This is a modest example of that.

What I find sad is what came next. House House doesn’t defend the decision. “This is not an endorsement of, or argument for, the use of AI tools, however minimal.” The studio says it supports people working to limit the harm of generative AI, and it apologizes “that Big Walk’s LLM usage may not meet their standards.” It also says that “Big Walk was developed without an internal AI usage policy in place, and we consider that lack of policy a failure.”

Those are the studio’s values, and it isn’t my place to argue House House out of them. If the team wants a stricter policy for its next game, I respect that choice. But after six years of work, the conversation about the game this week is whether its makers should be sorry for a few chatbot conversations and the code that came out of them. I don’t think they owed anyone that apology.

House House also explains how it came to disclose the use: “We originally wrote a version of the above disclosure as it was a suggested part of our IGF submission.” It notes that it hadn’t disclosed anything on Steam “as the platform does not request disclosure of code or chatbot usage.” The studio was asked how it made the game and gave a careful, complete answer. As far as I can tell from the reporting, that answer is why Big Walk became the center of this debate.

In that earlier note, I argued that penalizing disclosure creates “an incentive to say less about your process.” I was writing about storefront labels, but the same thing has happened here. A studio that decided a few chatbot conversations weren’t worth mentioning would probably have faced no questions. I worry other developers will conclude that they’re safer explaining less about how they work.

I want to be fair to the jurors. Their statement says, “The IGF has affirmed that we may cast our votes according to our own tastes and values,” and I agree that they may. I’ve said before that nobody owes AI a place in their creative process, and that applies to what a juror chooses to celebrate. Their concerns about labor, consent, the environment, and developers compelled by their employers to adopt these tools are serious. They also acknowledge disagreement within the jury: “Many jurors had differing perspectives and significant reservations about this statement.”

Where I disagree is with turning that position into festival policy. The jurors “urge the IGF to adopt a strong anti-AI policy for future years that disallows games containing any AI-generated work.” They’re explicit that code is no exception. That would turn a personal conviction into a condition of entry for every developer who submits a game.

I also don’t know how a festival would apply that rule. Would a fix that a developer rewrote from a chatbot’s suggestion count as “AI-generated work”? What about an answer summarized at the top of a search page, a line completed by a code editor, or a third-party library whose maintainers used a coding assistant? A studio can’t always know whether a library’s maintainers used one. The pledge covers generative AI used “in any capacity,” while the proposed policy covers games “containing any AI-generated work.” I’m not sure those describe the same set of games.

A festival has little way to verify any of this from the finished game. The rule would depend on developers reporting their own process, so it would mostly catch the ones who report it honestly. That’s the same problem I raised with storefront filters: someone willing to lie avoids the penalty, and someone who answers in good faith takes it.

I do want festivals to ask how games were made. House House’s disclosure is close to what I meant by useful transparency. It explains the problem, what the tool contributed, and what the team changed. An answer that specific deserves to be read for what it says. A rule based on “any capacity” doesn’t allow for that.

The jurors can vote according to their values, and House House can hold itself to whatever standard it believes in. I want the next small team asked how it made a game to feel able to answer as openly as this one did.