June 8, 2015, 2 pm
A central goal of both artificial intelligence and neuroscience is to understand human intelligence that possesses a natural ability to deal with a trade-off between performance, energy, and time. In this regard, it is becoming widely recognized that having several different learning and decision systems may be the optimal design for an artificial intelligence that operates under constraints on energy and time, and this may be both the way the human brain actually works. Based upon a combination of functional magnetic resonance imaging (fMRI) and computational modeling, I will discuss a theoretical account of how the human lateral prefrontal cortex (lPFC) influences the brain's various learning systems, placing the lPFC as the brain's “meta-controller”. I will present first evidence suggesting that the inferior lateral PFC allocate control to learning systems associated with either goal-directed (model-based), or habitual (model-free) learning. I will also show evidence supporting the view that the ventrolateral PFC plays a crucial role in switching between incremental and one-shot causal learning. These findings may help to explain how and why control processes breakdown in various psychiatric disorders, including obsessive-compulsive disorders and addiction. In turn, a deeper insight into these mechanistic anomalies may permit further development of neuromorphic algorithms for restoring stability to prefrontal circuits, as well as intelligent systems that make precise predictions about humans’ suboptimal behavior.