{"id":52765,"date":"2025-10-08T00:01:00","date_gmt":"2025-10-07T15:01:00","guid":{"rendered":"https:\/\/automaton-media.com\/en\/?p=52765"},"modified":"2025-10-07T17:24:02","modified_gmt":"2025-10-07T08:24:02","slug":"ishii-jiros-200-ending-roguelike-visual-novel-doesnt-use-generative-ai-but-human-designed-decision-trees-based-on-over-850-stage-plays","status":"publish","type":"post","link":"https:\/\/automaton-media.com\/en\/news\/ishii-jiros-200-ending-roguelike-visual-novel-doesnt-use-generative-ai-but-human-designed-decision-trees-based-on-over-850-stage-plays\/","title":{"rendered":"Ishii Jiro\u2019s 200-ending roguelike visual novel doesn\u2019t use generative AI, but human-designed decision trees based on over 850 stage plays"},"content":{"rendered":"\n

Jiro Ishii, game designer best known for creating cult classic 428: Shibuya Scramble, recently announced his brand-new social deduction style visual novel Depth Loop<\/strong>. Taking on the \u201croguelike visual novel\u201d genre pioneered by Gnosia, Depth Loop will be offering an endless cycle of replayability and at least 200 different possible endings<\/a>.as With each \u201cloop,\u201d the roles of the Mafia-inspired game\u2019s participants will be randomized.<\/p>\n\n\n\n

Though the devs have been utilizing AI technology to create behavioral patterns for characters and support all the endless game outcomes, talking to Famitsu<\/a>, Ishii states that the game doesn\u2019t actually rely on generative AI. Instead, he opted for a more down-to-earth approach.\u00a0\u00a0<\/p>\n\n\n\n

\"Depth<\/figure>\n\n\n\n

The idea behind using AI for Depth Loop\u2019s characters is to give the players a sense of participating in a social deduction style game against real people. Ishii states that AI was majorly used to realize a decision tree of thinking patterns and distinct routines for all 12 characters in the game. This is so that players can enjoy \u201ccompelling, strategic gameplay,\u201d he explains. Throughout development, Ishii has been studying different tactics that actual players use in social deduction games, and with the help of AI, the characters will be able to apply that knowledge, adapting to different situations and following the path that is optimal for ensuring their victory.<\/p>\n\n\n\n

However, that sounds like a lot of work, given that the devs will have to manually churn out all the possible tactics and behavior patterns. \u201cI\u2019m not the one doing system development, so it may be easy for me to say this,\u201d Ishii jokes, \u201cbut it\u2019s not like we\u2019re trying to make generative AI, so I believe it won\u2019t be detrimental to development. But we will be trying our best to incorporate as many possible strategies and gameplay trends as we can.\u201d<\/p>\n\n\n\n

\"Depth<\/figure>\n\n\n\n

Notably, Depth Loop is based on JINROU: The Playing Theater (TLPT) \u2013 a blend of Mafia-styled social deduction game and improvisational theater, in which professional actors are assigned characters and roles, and the audience gets to see the suspense unfold in real time. In order to achieve the same feeling of suspense offered by TLPT, Ishii got ahold of all of the past stage performances (more than 850 of them!), and checked them one by one, picking out all of the interesting situations, spoken lines and story beats to base the game\u2019s scenario off of. As he explained, \u201cthe main principle of the title\u2019s development is not to just make a game using AI, but to actually think how to properly incorporate the know-how we earned from more than 850 different stage performances.\u201d<\/p>\n\n\n\n

\n