Goals provide constraints
Higher-level objectives give subgoals priority and sustain a governing state under competition.
Cognitive control · Humans & AI
Doctoral Candidate in Cognitive and Comparative Psychology
CUNY Graduate Center
I investigate proactive cognitive control, Bayesian learning, and zero-shot generalization in humans and large language models (LLMs). My work draws on the brain’s cortical hierarchy and shallow, parallel subcortical routing to develop supramodal control architectures for rapid, uncertainty-weighted belief updating.
Eight valleys. Select an idea to raise its peak.
A brief phasic burst marks adjustment; a sustained tonic field marks maintenance. The peak, waves, and timing are visual analogies. Control networks contribute across timescales. Dual-network account · Later findings
Evidence and consistency gain influence. Inspecting evidence becomes a higher priority.
Higher-level objectives give subgoals priority and sustain a governing state under competition.
Means–end and compositional links determine how a goal recruits shared representations.
Changing the goal reshapes the landscape and the trajectories through it. The underlying nodes remain.
Behavioral experiments, explicit computational proposals, and comparisons between human cognition and language models.
In a color Stroop task, tested language models show increasing difficulty resolving conflict as word lists grow. These findings motivate our proposal for explicit executive-control mechanisms.
We argue for cognitive control as an explicit computational object, with testable roles in task-state inference, implementation, and adaptive regulation.
A comparison of language-model responses on measures of emotion identification, description, and empathy with human benchmarks, examining how these capacities vary across models.
I examine goal maintenance, interference, and the mechanisms that make a task state effective under competition.
Current work explores proactive control, Bayesian learning, and generalization as systems infer and update task states.
I investigate how behavioral evaluation and architectural proposals can make control mechanisms identifiable and testable.
Writing on agency, cognitive development, regulation, and the relationship between minds and artificial systems.
Read the essaysThe Stroop findings have prompted a broader conversation about the limits of attention in AI.