Advancing the Search Frontier with AI Agents

TL;DR

The paper argues that search is not solved and complex tasks require assistive AI agents. It presents a vision of AI copilots integrated with search engines, powered by a layered stack and grounded in a task-centric framework including task modeling, alignment, grounding, and task completion. It articulates a comprehensive opportunities framework spanning model innovation, next-generation experiences, measurement, and broader implications, along with challenges such as hallucinations, bias, and user control. The work emphasizes responsible AI, privacy, and economics as central to practical deployment, and outlines a roadmap for unifying traditional search with agent-driven dialogue to push the frontier toward end-to-end task completion.

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.

Figures (6)

• Figure 1: Task tree representation for a complex task involving planning a vacation to Paris, France. The tree depicts different task granularities (macrotask, subtask, action) and different task applications (decomposition, prediction, recognition) as moves around the tree. Time progresses from left to right via a sequence of searcher actions (queries, result clicks, pagination, etc.). Only actions are observable in traditional search engines. Aspects of subtasks and macrotasks may be observable to AI agents when searchers provide higher-level descriptions of their goals in natural language.
• Figure 2: AI agent stack depicting the various layers and the important role of AI safety and security across the stack. Foundation models can be either large language models (LLMs), with trillions of parameters, or small language models (SLMs), with just a few billion parameters. The star (*) symbolizes that there can be multiple cooperating agents as discussed later in Section \ref{['multiagent']}.
• Figure 3: Advancing the search frontier. Visualizing the set of possible tasks that can be tackled with search only today (i.e., finding, learning, and investigating) plus the expansion in the frontier into support for higher-order task activities with the addition of AI agents (e.g., adding emerging AI support for creative inspiration, synthesis, and summarization).
• Figure 4: Information interactions in a traditional search engine versus an AI agent.
• Figure 5: High-level overview of the typical generative AI search process in search engines. The query and the context are passed to the orchestrator, which coordinates with foundation model to create internal queries and generate answers. The orchestrator may also integrate content (e.g., search results, direct answers) from the search engine.