Shafik Quoraishee and Joanne Song contrast traditional state machines and learned policies with agents that interpret context and choose actions. Game examples illustrate perception, prediction, decision and action, while crossword constraints show why an agent still needs structured reasoning.
Shafik Quoraishee and Joanne Song emphasize local inference's privacy and offline advantages alongside practical limits. Model memory competes with rendering, and latency, battery consumption and hardware acceleration constrain what can run within a smooth frame budget.
Shafik Quoraishee and Joanne Song connect adaptive interfaces to accessibility: input tolerance, target sizes, keyboard escape paths and changes to layout can respond to changing player needs. These are exploratory designs requiring testing, not a claim that existing New York Times games already use the proposed agents.
Watch on YouTube
