Level-5 Apologizes for AI Use, Blames Human Error

Developer Level-5 has apologized for causing distrust through its inexperienced use of AI and admitting to human errors in its "VISION 2026 II: Dreams" showcase. Fans online criticized inconsistencies and elements suggestive of generative AI in trailers for games including Professor Layton and the New World of Steam, and Yokai Watch 2: Haunted Domain.

CEO Akihiro Hino initially stated that AI had been experimented with to make the presentation flashy, but stressed that core game elements like story and character designs were human-created. He mentioned AI tools were used to digitize artists' creations and speed up development cycles from five years to two.

However, this statement intensified backlash, with many questioning the focus on flashiness over creativity and the ethical implications of AI training data. The developer's subsequent statement on September 17 reiterated apologies for distrust caused by AI use and human mistakes, promising greater attention to accuracy and quality.

Level-5 announced it would remove AI-generated elements and human errors, releasing corrected trailers. They specifically addressed concerns about "unnatural" telephone lines in Yokai Watch 2, explaining it was a simplification of a hand-drawn reference photo, and similar issues with vending machines and guttering were human errors.

The studio clarified that differences in character models, like Kenta/Nathan's face between trailers, were due to updates before its own showcase deadline, not AI alterations. While assuring no AI was used in Professor Layton and the New World of Steam, they remained silent on the Curious Village remake.

Level-5 concluded by promising to limit AI to enhancing efficiency and avoid its direct use in fan-released videos. Despite this, public sentiment suggests a continued lack of trust, with some critics feeling the company is downplaying the fundamental issues of generative AI and shifting blame to staff. Level-5 has been experimenting with AI tools for efficiency for at least three years, using Stable Diffusion for layout ideas and 3D model creation, with AI outputs serving as guidelines for human artists.