Overview
- Erik Wolpaw, a writer behind Portal and Half-Life, disclosed on the MinnMax podcast that a small group at Valve has been experimenting with generative AI for text and audio.
- He emphasized the work is informal, not company-wide, and not attached to any announced or unannounced game.
- His tests suggest a narrow use in real-time reactive dialogue, as current models struggled with creative writing and humor but handled on-the-fly responses to player actions.
- He said the goal is to improve the player experience with reactions too vast for humans to script, and he noted AI voices trail real actors and are too costly to ship at scale now.
- Coverage from Kotaku, GamesRadar+, and PCGamesN is based on his podcast remarks rather than a Valve statement, with industry figures like Nobuo Uematsu and Patrick Söderlund voicing skepticism that AI can replace human-made art.