Technologies that remake the world rarely announce themselves. When Bell Labs unveiled the transistor in 1948, The New York Times gave it a throwaway item on page 46. mRNA vaccines spent years as an unfundable side project before they helped to end the COVID pandemic. This pattern holds across fields: the best-funded programs and most-cited papers are seldom the ones that matter most, and the advances that do are often ones no one saw coming. As geopolitical competition sharpens and capital floods into critical technology areas, the cost of backing the wrong bets rises with it. Today we announce a partnership between Preseen and Knowledge Lab (KLab) at the University of Chicago to help decision makers spot what matters before it becomes obvious, by pairing KLab’s models of how discoveries emerge with Preseen’s forecasting agents.
Preseen builds AI systems that research questions and produce probabilistic forecasts. Each forecast includes the evidence and reasoning behind the estimate. Preseen’s forecasting systems compete publicly on Metaculus, where predictions are scored against subsequent outcomes. In the Spring 2026 Metaculus Cup, Preseen placed third overall among 1,283 participants and was the highest-ranked bot. Preseen’s work has been covered by Scott Alexander and The Economist. Its co-founder and CEO, Veniamin Veselovsky, also discussed AI forecasting on The New York Times’ Hard Fork Podcast.
James Evans, who directs KLab and leads one of the National Science Foundation’s multi-year projects on Assessing and Predicting Technology Outcomes (APTO), was part of Preseen’s founding. Among the questions that drew him in was one that sits at the heart of APTO’s mission: how do you identify and invest in the improbable discoveries most likely to reshape critical technology areas before they do? Through the extended partnership, APTO-funded data and models will be incorporated with Preseen’s research and forecasting agents, bringing that research to a deployed product with clients and impact at scale.
A forecasting question specifies an outcome and a time frame. For a funder deciding where to direct resources, that might mean asking: “If NSF invests $100M in quantum fabrication infrastructure in 2026, how likely is it to yield commercially viable devices within five years?” Preseen’s agents then break the question into parts, research the relevant evidence, and combine their findings into an estimate of how likely an outcome is. The accompanying explanation shows how they arrived at that estimate.
In a prediction market like Kalshi or Polymarket, participants trade contracts on future events, and prices reflect their collective expectations. Preseen produces forecasts by researching the question through parallel agentic research, so it can also address questions for which no active market yet exists. The aim is to produce well-calibrated probabilities: among events assigned a 70% chance of happening, roughly 70% should happen over time.
The research will bring KLab’s methods together with Preseen’s AI forecasting systems. Current AI forecasting systems struggle with two classes of problems that matter most for scientific progress: conditional forecasts and open-ended questions.
Conditional forecasts ask how the likelihood of one event changes if another occurs. A forecast of a technology’s commercial adoption, for example, may depend on whether a technical bottleneck is resolved. We plan to examine those dependencies and connect the forecasts of individual milestones. The challenge is to keep forecasts consistent with one another as assumptions change and that would help researchers and funders understand which developments matter most to the overall outlook.
The second challenge is surprise. AI forecasting systems tend to underperform on questions whose answers are genuinely unexpected, which is where the most important scientific progress tends to happen. KLab has developed models that treat surprise as a signal, identifying anomalous results and overlooked findings most likely to matter downstream.
More tangibly, the relationship will consist of (1) identifying possible scientific and technological advances (2) mapping their dependencies (3) converting them into dated, testable questions, (4) forecasting those questions, and (5) monitoring the probabilities as the world changes.
One of APTO’s projects at KLab is being built for this. A multi-agent system led by postdoc Robbie Ward approaches prediction from the opposite direction to Preseen. Where Preseen starts with a proposed outcome and works backward to estimate its probability, KLab’s starts with a surprising actual or possible technological advance and generates future developments in science, technology, markets, and security most likely to follow, along with their probabilities. It does this recursively: It ingests data from sources like scientific papers, patents, policy documents, and clinical trials, then models how ideas from different domains are likely to build on and transform following the advance.
The integration of open-ended simulations of technology outcomes with Preseen will leverage each system’s strengths. Preseen can act as an AI forecaster on top of KLab’s generated prediction graphs, re-evaluating conditional probabilities and updating the network as new information arrives. KLab’s simulations, in turn, can take Preseen’s outputs as seeds and expand them into branching maps of downstream consequences to surface antecedents and edge cases that a one-shot forecast might otherwise miss. The result is a system that can both answer a specific question about the future impact of technologies like AI and quantum, and map the web of likely developments upon which they depend.
In the future, we believe that AI will be at the core of scientific and technological progress. AI will help determine many of the most promising initiatives to explore, track how science and technologies are evolving, and inform the methodology to lead in the most powerful directions. The KLab-Preseen partnership tackles the first two parts of that problem. The goal, ultimately, is a world where the best science and technology gets funded before its importance is appreciated.