Home
Publications
Research
Teaching
Biography
Talks / Outreach

Murray Shanahan



Research

I am interested in the principles that underlie sophisticated cognition, both as it is found in Nature and as it might be realised artificially. Because I'm committed to the view that cognition and embodiment are intimately related, my work is focused on agents that are coupled to complex environments through sensorimotor loops, such as robots and animals. I'm also interested in consciousness, and I see consciousness and cognition as closely related. Furthermore, in my opinion, to understand these themes properly entails a degree of engagement with philosophy, particularly philosophy of mind. Finally, I am interested in the long-term implications of artificial intelligence, both its potential and its risks.

My early career work (late 1980s and 1990s) was in the area of classical, symbolic AI, and mainly concerned how to represent and reason about actions and their effects. I then moved into cognitive robotics in the mid 1990s, while retaining some commitment to the tenets of classical AI. In the early 2000s I shifted away from classical AI and began to study the biological brain and cognitive architecture more generally, with a particular emphasis on the relationship between cognition and consciousness. By the early 2010s, my research focus had moved to brain networks and neurodynamics, which continue to fascinate me. I have recently become interested in applying what I've learned from neuroscience to machine learning, especially to reinforcement learning.

Here are some of the specific themes I'm interested in. For details consult my publications.
  • Neurodynamics. The biological brain exhibits a variety of exotic dynamical phenomena, including metastability, dynamical complexity (a balance of segregated and integrated activity), and criticality. I am interested in building models of such systems, especially based on complex network topologies.

  • Global workspace theory. This is Bernard Baars's theory of consciousness, which is based on a cognitive architecture comprising a set of parallel specialist processes and a global workspace. The specialists compete and co-operate for access to the global workspace, from which information is broadcast back to the whole set of parallel specialists.

  • Internal simulation. This is the idea that the brain internally simulates interactions with the environment in order to plan ahead. This capacity can be likened to the operation of the "imagination" or of a predictive model. I am interested in the idea that it works by rehearsing trajectories through sensorimotor space.

  • Machine learning. I am most interested in reinforcement learning, particularly as the basis of an architecture for general intelligence.

  • The frame problem. Originally this was a problem that arose in logic-based AI, but my recent interest relates more to the philosophers' interpretation of the term.

  • The technological singularity. Without subscribing either to alarmist or to evangelist views, I am interested in understanding the long-term implications for humanity of AI technology.


>
Home
Publications
Research
Teaching
Biography
Talks / Outreach