Abstract
As many of us in the information retrieval (IR) research community know and appreciate, search is far from being a solved problem. Millions of people struggle with tasks on search engines every day. Often, their struggles relate to the intrinsic complexity of their task and the failure of search systems to fully understand the task and serve relevant results. The task motivates the search, creating the gap/problematic situation that searchers attempt to bridge/resolve and drives search behavior as they work through different task facets. Complex search tasks require more than support for rudimentary fact finding or re-finding. Research on methods to support complex tasks includes work on generating query and website suggestions, personalizing and contextualizing search, and developing new search experiences, including those that span time and space. The recent emergence of generative artificial intelligence (AI) and the arrival of assistive agents, based on this technology, has the potential to offer further assistance to searchers, especially those engaged in complex tasks. There are profound implications from these advances for the design of intelligent systems and for the future of search itself. This article, based on a keynote by the author at the 2023 ACM SIGIR Conference, explores these issues and how AI agents are advancing the frontier of search system capabilities, with a special focus on information interaction and complex task completion.
Introduction: The Challenge of Complex Search
Traditional search engines are highly effective for "lookup" tasks, such as finding a specific fact or navigating to a known website. However, they often fall short when users attempt to complete complex, multi-step "macrotasks." Research suggests that roughly half of web searches go unanswered because systems fail to grasp the intrinsic complexity of the user's underlying goal. The gap between a user's high-level intent and the granular queries they must issue creates a significant hurdle in information retrieval (IR).
Conceptual Frameworks for AI Agents
To address these challenges, the paper proposes moving beyond simple query-response cycles toward assistive agents, or "copilots." These agents leverage generative AI and Retrieval-Augmented Generation (RAG) to bridge the gap between intent and action.
The Task Tree
A central concept is the Task Tree, which models human activity at three levels of abstraction: Macrotasks (high-level goals), Subtasks (specific components), and Actions (granular steps like queries or clicks).
By understanding this hierarchy, AI agents can assist in task decomposition—breaking a large goal into manageable steps—and task recognition—inferring the broader goal from a series of individual actions. This allows the system to support the user across time and space as they work through different facets of a project.
Expanding the Search Frontier
AI agents are pushing search capabilities up Bloom’s Taxonomy ladder. While traditional search focuses on "Remembering" and "Understanding" (lookup), agents enable higher-order cognitive activities like "Evaluating," "Synthesizing," and "Creating." This expansion allows users to describe goals in natural language, shifting the system's role from providing a list of links to generating comprehensive, actionable outcomes.
Comparing Search and AI Agent Workflows
The fundamental shift lies in who performs the cognitive labor of synthesis. In traditional search, the user must decompose their goal, issue multiple queries, and manually aggregate information from various search engine result pages (SERPs). In an AI agent workflow, the agent takes on the burden of retrieval, aggregation, and synthesis, presenting the user with a direct answer or a completed task.
Limitations and Future Directions
Despite their promise, AI agents face significant hurdles that the IR community must address:
- Reliability and Grounding: Hallucinations and bias remain critical issues, necessitating robust grounding through RAG and clear provenance (citations) for all generated content.
- Human Agency: There is a delicate balance between automation and control. Users need to understand why certain information was included and maintain the ability to intervene.
- The Content Ecosystem: A "paradox of reuse" exists: if agents summarize content so effectively that users never click through to original sources, content creators may lose the incentive to publish, eventually starving AI models of new data.
- Evaluation Metrics: Traditional metrics like Precision and Recall are insufficient for agents. New frameworks are needed to measure "task success," "user effort," and "generation quality."
As search evolves into a more agentic experience, the focus must remain on augmenting human capability while preserving the learning and discovery inherent in the search process.
Summary updated