Using AI to build character dialogues

What the talk is about

Oleksii Ilchyshyn looks at how artificial intelligence could change dialogues with non-player characters. He briefly walks through the history of game dialogue, from branching trees and browser-like topic menus to the dialogue wheel, which he feels often fails to convey what the player meant to say. The speaker outlines a classification of game AI and the kinds of player input a character should understand: questions, statements, commands, compliments and insults. He compares cutscenes and dialogue trees with natural language processing, which gives players freedom and makes characters feel more autonomous, though scenarios still have to be written by hand. His ideal future is characters that learn while talking. In the Q&A he discusses the risk of novelty fading, toxicity, localization and model training time. Useful for game designers and narrative designers.