St John Grimbly
I study how intelligent agents decide what matters. My work asks how they learn uncertain models of the world, allocate limited attention, and turn bodily or task-relevant needs into adaptive behaviour.
Active inference, learning and artificial minds
My current work combines active inference and model-based reinforcement learning. I am interested in agents that maintain uncertain models, plan ahead, and direct limited computational or perceptual resources towards the problems that matter most.
This also raises a broader question. What can artificial agents tell us about biological regulation, affect and consciousness? My earlier research focused on causality in reinforcement learning, especially whether causal structure can make learning more efficient, robust and interpretable.
- 2026
Interoceptive Attention as Dynamic Homeostatic Prioritization in a Foraging Agent
From Animals to Animats 18 · SAB 2026
- 2026
Inferring Affective Consciousness in an Artificial Agent
Journal of Consciousness Studies · 33(7), 14–34
- 2025
Sophisticated Learning
SSRN preprint · Active learning during model-based planning
-
PhD Candidate · University of Cape Town
Active inference, interoceptive attention, model-based planning and computational approaches to affective consciousness. Supervised by Jonathan P. Shock and Mark Solms.
-
MSc · University of Cape Town, 2023
Research on trends, problems and solutions in causality and reinforcement learning.
-
Sponsored Associate · Syracuse University, 2023
A January–June research stint with Ferdinando Fioretto's lab, investigating bias and fairness in causal discovery.
-
Multi-agent reinforcement learning · InstaDeep
Contributed to the original Mava distributed MARL framework and implemented QMIX.
-
Honours · University of Cape Town, 2020
Research on counterfactual methods in sequential decision-making.