Learning to Work with You
How should an agent learn to work with you?
This map annotates the research behind User-Specific Interaction Learning (USIL): work on learning about individual users, and work on when agents should ask, act, confirm, or defer. Each paper is labeled by what it learns, whether it learns from one person's evidence, and which interaction decisions it affects.
How to read the labels. Labels describe what each paper reports, not what its methods could support. Papers marked Full text were annotated from the full paper with quotes checked against the text; the fourteen in Closest predecessors were rated by hand. Other labels come from the catalog and may change. "Not established" means the paper does not say, not that the capability is absent.
The landscape at a glance
Click any bar or cell to filter the papers below.
Where the connections thin out
Papers that combine each successive USIL ingredient.
What is personalized, and for whom
Outcome vs. procedure, population vs. individual.
Papers per year
Most of the work dates from 2023 onward.
Interaction decisions studied
Papers that study each decision; a paper can count more than once.
How much is established
Share of papers by what they report.
Papers by research area
Each paper has one primary area.
Research areas against the USIL research challenges
P1 learning from sparse and evolving evidence · P2 representing and generalizing preferences · P3 when to infer, ask, act, or defer · P4 evaluation.
Closest predecessors
Ratings checked against each paper's full text (Table 1 of the position paper).
YesPartialNoNot established · "Cross-task transfer" means carrying what was learned about a user to new kinds of tasks.