When
Nadja R. Ging-Jehli, PhD
Assistant Professor
Neurocomputation, Adaptive Intelligence & Agency Dynamics (NAIADEL Lab)
Cognition and Neuroscience Systems, Department of Psychology, University of Arizona
Adapting Under Partial Control: From Neural Circuits to Interactive Worlds
Abstract: People rarely have full control over their circumstances. We must decide how much caution a choice deserves, learn what our actions are worth, switch strategies when conditions change, and judge when to keep trying and when to let go. I use the term adaptive intelligence for this capacity to adjust how we learn, decide, and exercise control. It is a requirement for any agent that must act with incomplete information and partial control, whether biological or artificial. My research asks how it is implemented, why it breaks down in different ways in different people, and what it offers for building artificial agents. The talk follows this question across three time scales.
Within a single choice, adaptive intelligence means deciding how much evidence to require before acting. In patients who have electrodes implanted for the treatment of Parkinson’s disease, I combine direct recordings from the basal ganglia with models of evidence accumulation. Two structures play complementary roles in setting this requirement. The subthalamic nucleus raises caution when options conflict. The globus pallidus coordinates adjustments as uncertainty grows. Deep brain stimulation shows that the relationship is causal. High-frequency stimulation lowers the evidence requirement and makes responses faster but less accurate. Low-frequency stimulation has the opposite effect. The circuit that supports adaptive choice is therefore also a point of clinical intervention.
Across repeated choices, adaptive intelligence means revising a strategy in response to feedback. Combining reinforcement-learning models with EEG, I show that learning impairments in depression and in bipolar disorder can look alike while arising from different processes. The same logic applies to classical laboratory phenomena. The cost of switching between tasks is usually treated as one quantity, but game-based environments can separate it into distinct sources of error.
In a dynamic environment that responds to our actions, adaptive intelligence means judging how much control we actually have. I introduce Gearshift Fellowship, a game-based platform in which the environment adapts to the participant and in which rewards, threats, and opportunities for control change over time. Computational models examine how persistence and disengagement depend on emotional appraisal, beliefs about one’s own competence, explanations for past outcomes, and estimates of available control.
One theme recurs across all three scales: similar behavior can arise from very different underlying processes, which matters for mental health because effective intervention depends on identifying the mechanism rather than the behavior. I will close with the research program of the NAIADEL Lab, which brings together circuit neuroscience, computational psychology, game-based experimentation, and artificial agents as model systems for adaptive intelligence.