Knowing Isn't Understanding: Re-grounding Generative Proactivity with Epistemic and Behavioral Insight

Kirandeep Kaur; Xingda Lyu; Chirag Shah

cs.CYFeb 16, 2026

Knowing Isn't Understanding: Re-grounding Generative Proactivity with Epistemic and Behavioral Insight

Authors: Kirandeep Kaur, Xingda Lyu, Chirag Shah

Organizations: Paul G. Allen School of Computer Science, University of Washington Seattle, USA · Information School, University of Washington Seattle, USA

Abstract

Generative AI agents equate understanding with resolving explicit queries, an assumption that confines interaction to what users can articulate. This assumption breaks down when users themselves lack awareness of what is missing, risky, or worth considering. In such conditions, proactivity is not merely an efficiency enhancement, but an epistemic necessity. We refer to this condition as epistemic incompleteness: where progress depends on engaging with unknown unknowns for effective partnership. Existing approaches to proactivity remain narrowly anticipatory, extrapolating from past behavior and presuming that goals are already well defined, thereby failing to support users meaningfully. However, surfacing possibilities beyond a user's current awareness is not inherently beneficial. Unconstrained proactive interventions can misdirect attention, overwhelm users, or introduce harm. Proactive agents, therefore, require behavioral grounding: principled constraints on when, how, and to what extent an agent should intervene. We advance the position that generative proactivity must be grounded both epistemically and behaviorally. Drawing on the philosophy of ignorance and research on proactive behavior, we argue that these theories offer critical guidance for designing agents that can engage responsibly and foster meaningful partnerships.

Figures & tables

Figure 1: Epistemic proactivity under uncertainty. A proactive agent surfaces gaps within a user’s epistemic landscape, reorganizing known and partially known regions (KK, KU, UK) and incrementally engaging the epistemic frontier under uncertainty.

Figure 2: Proactivity regimes organized by the variable governing initiative: prediction in anticipatory systems, regulation in mixed-initiative systems, and commitment in autonomous systems.

Category	Representative works	KK	KU	UK	UU	Explanation
Anticipatory IR / proactive retrieval	( Lieberman, 1995 ; Rhodes and Maes, 2000 ; Liebling and Dumais, 2012 ; Shokouhi and Guo, 2015 ; Yang et al., 2016 ; Song et al., 2016 ; Bahrainian et al., 2016 ; Luukkonen and Kekäläinen, 2016 ; Samarinas and Zamani, 2024 )	✓	✓	✗	✗	Forecasts or initiates retrieval over a fixed information space; uncertainty over known targets.
Sequential / basket recommendation	( Adomavicius et al., 2011 ; Hidasi et al., 2016 ; Li et al., 2017 ; Kang and McAuley, 2018 ; Sun et al., 2019 ; Yu et al., 2016 ; Le et al., 2019 )	✓	✗	✗	✗	Proactivity is prediction/selection among known items under a fixed catalog and representational schema.
Web / OS / embodied agents	( Nakano et al., 2021 ; Yao et al., 2022 ; Zhou et al., 2024 ; Xie and others, 2024 ; Liu et al., 2024 ; Mialon et al., 2024 ; Ahn and others, 2022 ; Driess and others, 2023 ; Zitkovich et al., 2023 ; Wang and others, 2024 )	✓	✗	✗	✗	Acts within benchmarked environments and task specs; success is defined by predefined objectives and action spaces.
Planning + tool-using LLM agents	( Yao et al., 2023b ; Shinn et al., 2023 ; Yao et al., 2023a ; Bhatia and others, 2024 ; Schick et al., 2023 ; Patil and others, 2023 ; Qin and others, 2024 ; Guo and others, 2024 ; Jimenez and others, 2024 ; Yang and others, 2024 ; Wu and others, 2024 ; Li and others, 2024 ; Hong and others, 2024 )	✓	✓	✗	✗	Optimizes actions with known tools/goals/metrics; does not reframe what should be modeled or surfaced.
Proactive conversational agents (human-centered)	( Nothdurft et al., 2015 ; Wu et al., 2019b ; Kraus et al., 2020 ; Chen et al., 2023 ; Bairi and others, 2024 ; Liu et al., 2025 ; Chen et al., 2025a ; Deng et al., 2024 )	✓	✓	✓	✗	Models when/how to intervene and user-facing appropriateness/constraints via interaction, but within predefined dimensions.
Mixed-initiative clarification and sensemaking	( Horvitz, 1999 ; Horvitz, 2007 ; Deng et al., 2023 ; Kraus et al., 2021 ; Chen et al., 2024 ; Sekulić et al., 2022a ; Mass et al., 2022 ; Wu et al., 2023 ; Yuan et al., 2024 ; Rahmani et al., 2024 ; Lei et al., 2020 ; Kang et al., 2023 ; Ye et al., 2025 ; Chen et al., 2025b ; Overney et al., 2025 ; Mei et al., 2025 ; Radensky et al., 2024 ; Shankar et al., 2024 )	✓	✓	✓	$\sim$	Surfaces latent intent/constraints through interaction and mixed-initiative workflows, but may not surface missing task dimensions (UU).

Table 1: Epistemic reach of prevailing proactivity approaches, summarized by the highest epistemic state supported by each category. The absence of explicit UU support reflects a structural gap: proactivity is operationalized as action selection within assumed task frames, rather than as discovery of missing dimensions.

\sim

: partial / limited support

Figure 3: Inverted doughnut model of proactive behavior

Figure 4: Epistemic–behavioral coupling space.

Figure 5: Epistemic partnership through proactive gap surfacing. Illustrative example of how proactive agents can support inquiry by surfacing latent epistemic gaps rather than executing premature action. Known facts do not by themselves determine the governing relationship; progress emerges when missing relations are identified and articulated through interaction. Proactive surfacing of such gaps expands the inquiry space, generating newly articulated unknowns and enabling joint discovery.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Panel of Figure 6 — Figure 6: Anticipation as an act-ahead pipeline.

Axis of initiative	Available choices
	Option 1	Option 2	Option 3	Option 4
Who acts	Human	System	Negotiated	–
When to act	Timing	Interruption	Turn-taking	–
How much to act	Suggest	Clarify	Execute	Defer

Appendix

Table 2: Core axes along which mixed-initiative systems regulate proactive behavior. Each axis defines a discrete choice set governing when and how initiative is exercised.

Proactive Assistance

Proactive assistance studies how AI systems can anticipate users’ needs and decide what help to offer, when to offer it, or whether to stay silent.

See all 46 papers

Foundations of Proactive Agents: Principles, Technical Layers, and Proactivity-Gym
Sep 29, 2026Jio Oh, Seunghyun Do, Young-Jun Lee +2LLM Agent Evaluation Long-Horizon Agent Evaluation
KAIST · University of Minnesota
Asking for What Was Never Requested: Horizontal and Vertical Proactivity in Agents
Sep 29, 2026Ido Levy, Asaf Yehudai, Segev Shlomov +2Proactive Agents Proactive Assistance
IBM · Weizmann Institute of Science
Proactive Service Agents: A Unified Decision Framework, Methods, and Evaluation
Sep 3, 2026Yan Tang, Tingyu Cao, Yuanbo Tang +2LLM Agent Evaluation Proactive Assistance
UniCone Team

Sep 29, 2026J/K move · Enter open · S save

Foundations of Proactive Agents: Principles, Technical Layers, and Proactivity-Gym

Jio Oh, Seunghyun Do, Young-Jun Lee, Steven Euijong Whang+1

KAIST · University of Minnesota

LLM Agent Evaluation Long-Horizon Agent Evaluation Proactive Assistance LLM Agents Trust in AI Proactive Agents

Open paper AskPDF arXiv

Proactive LLM agents can turn idle compute into useful support before users ask. Yet even correct work can misread user context, impose review costs, or undermine trust. This work proposes foundations for designing, realizing, and evaluating proactive LLM agents around three joint principles (3T): Task Capability, anticipating relevant needs and correctly performing useful work; Temporal Allocation, allocating compute according to resource availability and when results are needed; and Trust, sustaining users' confidence and appropriate reliance on the agent. We connect these objectives to a design space organized around five dimensions: task scope, anticipation horizon, activation trigger, processing timing, and intervention depth, and specify the situation and system modeling needed to support its choices, including user and environment representations, backbone LLMs, and agent harnesses. Lastly, we propose PROACTIVITY-GYM, a simulation-based evaluation testbed including multi-day scenarios, stateful environments, and persona-conditioned simulated users that can evaluate the consequences of proactive assistance across interactions. Evaluations across 23 model-harness configurations uncover substantial performance gaps across 3T and reveal that LLM judges often conflate task capability and trust. A human study with 30 participants demonstrates the importance of the joint 3T optimization: participants show sharp trust declines after intervention misalignment despite correct outcomes, and prefer sleep-time assistance, even when imperfect, to preserve ongoing focus. Together, these findings support designing and evaluating proactive agents through the joint consideration of useful work, compute allocation, and evolving user trust.

Abstract

Proactive Assistance

Explore similar work

Foundations of Proactive Agents: Principles, Technical Layers, and Proactivity-Gym

Asking for What Was Never Requested: Horizontal and Vertical Proactivity in Agents

Proactive Service Agents: A Unified Decision Framework, Methods, and Evaluation