Losing critical knowledge when employees leave is one of the biggest hidden crises organizations face today. But what if there was a way to capture and share expertise effortlessly, even when you don’t know exactly what to look for? This is the challenge that the Knowledge Management at Scale and Speed (KMASS) program, led by DARPA, aimed to solve—with powerful contributions from the University of Southern California’s Information Sciences Institute (ISI).
In August 2025, DARPA gathered top international research teams in Arlington, Virginia, to showcase the final outcomes of KMASS, a cutting-edge initiative tackling the costly problem of lost institutional knowledge as staff depart or move on. Jay Pujara, the director of ISI’s Center on Knowledge Graphs, spearheaded USC Viterbi’s involvement, investing two years developing innovative tools that don’t just wait for users to know what to search for but proactively deliver the insights they need.
Traditional information systems require users to have a clear question or understanding before they can find the right data. This common issue often blocks learning and problem-solving, especially in high-pressure or complex operations. KMASS breaks this mold by creating technology that understands context and pushes relevant information to users before they even ask for it—eliminating guesswork and speeding up critical decisions.
Pujara, a research associate professor at USC Viterbi School of Engineering, unveiled "Knowledge Needed in Context" (KNIC), a smart assistant that observes how experts perform their tasks, records their workflow, and then offers precise guidance to others tackling similar challenges later on. Experts in fields like intelligence analysis and resource management who saw KNIC in action immediately recognized its potential to fit seamlessly into real-world workflows.
DARPA captured the essence perfectly in their post-event headline: “Don’t just ask a chatbot. Have it push useful info right when it’s needed.” But here’s where it gets controversial—should we rely on AI to decide when and what knowledge to deliver? What if it gets it wrong or oversteps?
KNIC approaches knowledge transfer as if it were a digital apprentice, silently watching over seasoned professionals, noting how they navigate their tasks, and preserving those methods for newcomers. The system tackles knowledge management in three stages: First, it observes experts in real time, extracting patterns in how they think and act. Second, it organizes this captured knowledge into a structured, accessible format. Third, KNIC delivers targeted examples and advice exactly when a user needs help, avoiding the usual delays or reliance on guesswork.
"Trying to write down all your know-how for the next person is exhausting and often incomplete," says Pujara. "Our system watches your actions, summarizes what you do, and even asks questions to fill in missing pieces." Unlike typical chatbots or AI tools that generate answers based on probabilities (and can sometimes hallucinate incorrect info), KNIC only provides real examples from your own organization. This means zero fabrication, real-world context, and true learning by doing, not just copying AI responses.
A key innovation is that KNIC minimizes interference—it carefully measures its confidence level before stepping in, behaving like a courteous apprentice who knows when to speak up and when to stay silent, thus supporting users without disrupting their workflow.
KNIC was first tested in data science environments, where it observed users working with Jupyter Notebooks—a widely-used tool for writing and running code. It learned how analysts write scripts and examine data, then guided novices to replicate those processes successfully. Later, KNIC integrated with technology from InferLink, assisting government analysts in extracting and analyzing data from documents like contracts and tables.
With KMASS concluding officially on September 20, 2025, the ISI team is shifting gears from research to implementation. "The exciting, hands-on phase begins now," Pujara explained. The goal is to move beyond demos and prototypes to real-world deployment, not just for government but for large companies, universities, and any organization wrestling with high turnover or complex training needs.
This transition highlights a bold vision: rather than knowledge disappearing when personnel change, it can evolve and grow within organizations. For sectors where experience is invaluable and errors costly—like military, healthcare, or critical infrastructure—this could be a game changer.
But here’s the provocative question worth debating: Could giving AI this kind of autonomy in knowledge transfer introduce risks? Who ensures the digital apprentice’s guidance stays accurate and unbiased over time? How do organizations balance trust in smart tools without losing human judgment?
We invite you to share your thoughts—do you believe systems like KNIC will revolutionize how we learn and retain expertise, or do you see potential pitfalls in handing over too much to automated assistants? Join the conversation below!
