The project and the person behind it

Leading AI systems now forecast world events about as accurately as superforecasters, the best human forecasters. This is an impressive and important achievement. Forecasting is nonetheless a narrow strategic skill, only as good as the question asked. Leadership Under Uncertainty is an applied research project to build AI systems demonstrably capable of strategic understanding and, ultimately, of judgment. It has potential application to higher-value strategy work and to AI safety, because judgment is a key attribute of responsible decision-making. Its model is the intelligence community, where strategic assessment and judgment are understood as a dynamic process: gathering and evaluating information, rolling individual assessments up into broader judgments and, above all, systematically updating those judgments on new information and assessments. The project asks whether AI systems can replicate that process, whether it enhances their strategic understanding, and whether integrating a judgment layer into agentic AI measurably improves decision-making.

The project is led by Christopher Berzins, Visiting Professor at the Graduate School of Public and International Affairs, University of Ottawa, where the research is conducted. The project lead wrote the launch paper and the six companion manuscripts and designed the reference software that implements them.

The project’s next stage is open to collaborating researchers, practitioner communities and institutional partners. It begins with the four first studies set out in the launch paper, none of which needs a resolved outcome. The decision study follows, comparing AI agents with and without the judgment layer while holding forecasting ability constant, and its historical track, on declassified cases disguised so that a model cannot recognize them, would run first at small scale in the first half of 2027. The stage also includes a study with a participating board and independent replication of the initial findings.

How the project reports results and handles data

1. 1New studies are pre-registered and sealed before they run.
2. 2Results are published whether or not they support the design, and any departure from the registered plan is stated.
3. 3The launch paper labels every claim about the project by level, from specified in a manuscript to measured in a study.
4. 4No gain in accuracy is claimed until sealed judgments scored against outcomes show one.
5. 5The design keeps the authority to act with accountable people. In the public judgment the author fixes the question and the rule in advance, and every change is on the record with its reasons.
6. 6Partners’ documents stay with them, results are released only in a form they approve, and research with people follows ethics review.

The launch paper reports its results whether or not they favour the design, and its appendices give the methods and limits of each.

Write to the project

For collaboration, data contributions, rater participation, funding discussions, or comments on the papers, write to Christopher Berzins at cberzins@uottawa.ca.

Christopher Berzins
Visiting Professor
Graduate School of Public and International Affairs, University of Ottawa