Big Walk used LLMs for code, House House says

Nineteen jurors for the upcoming 29th IGF Awards say they will not vote for any game using generative AI, while House House says LLMs contributed to under 0.1% of Big Walk’s code.

The jurors say programming is creative labor that deserves respect equal to that given to other parts of game development. Their objection also extends to models trained on creators’ material without consent, leaving no stated exception for AI used to help with code.

House House says it disclosed its limited use after the issue became a matter of public interest. The studio’s explanation of the disclosure says LLMs were used in the final stretch of development to tackle technical challenges and bugs.

The studio says its main use was asking chatbots to research problems with third-party code. In some cases, it adapted snippets from LLM responses as starting points for game code; it says generative AI had no other role in making Big Walk.

One example was a slowdown bug that House House says its staff and a Unity engineer could not solve. The studio then turned to a chatbot for a fix.

House House says it prepared a version of the disclosure for its IGF submission, but did not disclose the use on Steam because the platform does not ask about code or chatbot use. The studio stresses that disclosure is not an endorsement and worries that voluntary disclosures could make generative AI seem normal or inevitable.

The studio supports efforts to limit generative AI’s harms, apologized that its use might not meet others’ standards, and called the lack of an internal AI-use policy a failure it plans to address. The jurors’ all-or-nothing rule is blunt when the disclosed use ranged from chatbot research to modified code, though their concerns about programming labor and consent deserve serious consideration.

Sam Barlow said discussion among IGF judges about what the awards should represent was more intense this year.