Toward Knowledge-Centric AI: Discovering, Internalizing, and Using Explicit Knowledge Across Agents and Generations

PhD Qualifying Examination


Title: "Toward Knowledge-Centric AI: Discovering, Internalizing, and Using
Explicit Knowledge Across Agents and Generations"

by

Mr. Yisen GAO


Abstract:

Knowledge is how understanding accumulates across people and generations:
durable records that outlive their producers and that others can check, combine,
teach, and correct. AI agents are now built and run by many independent users
and organizations, and each agent is upgraded, replaced, or retired within
months. Yet what an agent learns often remains trapped in model parameters, a
transient context window, or a system-specific memory store that is appended to
and retrieved from rather than processed. This survey reviews research toward
knowledge-centric AI through a three-stage loop built around an explicit,
continuously evolving knowledge layer independent of any single agent. Agents
discover knowledge from their environments and trajectories; the layer
internalizes it by turning fragmented, heterogeneous records into composable and
revisable understanding; and agents across systems and generations use the
result and return new outcomes for revision. For each stage, we systematize
representative methods, knowledge types and forms, and application settings, and
assess maturity, scope, and remaining gaps. The landscape is uneven. Discovery
from unstructured text and symbolic use by large language models are
comparatively advanced. Procedural and metacognitive discovery from experience
and parametric use are advancing rapidly but remain weakly validated. Reliable
extraction from structured-data foundation models, persistent internalization of
large heterogeneous stores, and knowledge use by non-language foundation models
remain sparse. Within internalization, the survey covers cross-carrier
alignment, condition-preserving conceptualization, consolidation, repair and
completion, and deductive, inductive, and abductive reasoning across noisy,
heterogeneous representations, including text, tables, graphs, rules, and
programs. It closes with open challenges in common knowledge units, operator
invariants, repair and evolution at scale, cross-system and cross-generation
sharing, and knowledge interfaces across foundation models.


Date:                   Thursday, 3 December 2026

Time:                   10:00am - 11:00am

Venue:                  Room 3494
                        Lifts 25/26

Committee Members:      Dr. Yangqiu Song (Supervisor)
                        Dr. Binhang Yuan (Chairperson)
                        Dr. May Fung