SOUTENANCE DE THÈSE   

Vous êtes cordialement invités!   

Mardi 19 mai 2026 

9h30

Local: PK-2265  

TITRE : Cognitive and computational tools to improve reasoning and decision making in the intelligence community

Présenté par

Matthew MARTIN, personne doctorante en informatique cognitive  

RÉSUMÉ

Improving reasoning and decision-making in the intelligence community (IC) could reduce the chance of wars and other catastrophes.  This thesis describes three projects that develop and test cognitive and computional tools to improve reasoning and decision-making.  The first project conducted a content analysis of intelligence products to examine the relationships between techniques used, adherence to IC of Standardizing Probability Ranges and Timeline were predictive of IC norm adherence.  In the second project, two experiments used the Practical scoring rule as performance feedback to train better calibration of confidence.  The results showed no improvement to calibration from training, contrary to previous results.  However, actively open-minded thinking was positively related to overall calibration.  The third project usec recalibration and weighted aggregation in combination with Bayesian networks (BNs) and logistic regression to improve the accuracy of probabilistic forecasts made during a multi-year forecasting tournament.  Recalibration and aggregation alone produces the most accurate forecasts, followed by BNs, then logistic regression.  Future studies should use content analysis on a larger set of real intelligence products to address questions of external validity ans increase the power of the first project.  For the second project, the effects of participant motivation and comprehension should be probed to determine when calibration training with the Practical score is effective.  Finally, for the third project, more advanced machine learning techniques like hierarchical BNs and ensemble methods should be expoored to further increase the accuracy of recalibrated aggregated forecasts.

Mots clés : Reasoning, decision-making, intelligence community, probability, forecasting, content analysis, calibration, machine learning, Bayesian networks, logistic regression

 JURY D'ÉVALUATION

Mandeep Dhami, Université Middlesex de Londres, professeure au département de psychologie (membre externe)

Janie Brisson, UQAM, professeure au département d'éducation et pédagogie (membre interne) 

Hakim Lounis, UQAM, professeur au département d'informatique (membre interne et président du jury)

Serge Robert, UQAM, professeur au de philosophie (direction de recherche)  

Roger Villemaire, UQAM, professeur au département d'informatique (codirection de recherche)