My principle scientific passion is understanding cognition, particularly as it relates to explaining human culture, but also natural intelligence more broadly. My main methodology for doing this is designing intelligent systems to model and test scientific theories. Most of my research has focused on the unintentional and non- or proto-linguistic aspects of human intelligence, and how intelligence evolves more broadly. From 2000–2007 I worked primarily on understanding non-human primate behaviour. Since 2008, my group has been more focused on characteristics of human cognition such as consciousness, artificial intelligence “feelings” (emotions), language, religion, and especially cultural variation in cooperation.
Understanding Natural Intelligence
Natural intelligence (NI) is the opposite of artificial intelligence: it is all the systems of control present in biology. Normally when we think of NI we think about how animal or human brains function, but there is more to natural intelligence than neuroscience. Nature also demonstrates non-neural control in plants and protozoa, as well as distributed intelligence in colony species like ants, hyenas and humans. Our behaviour co-evolves with the rest of our bodies, and in response to our changing environment. Understanding natural intelligence requires understanding all of these influences on behaviour and their interactions.
One of the best methods for understanding how NI systems work is to try to replicate their behaviour in simulation. Just as learning to paint forces you to understand the details of what you are seeing, building a working model forces you to understand the intricacies of what the target intelligent system is doing. For example:
What environment does it work in?
What aspects of that environment does it rely on?
What does it need to do itself?
How much does it need to learn and remember?
What can it learn just from its senses?
How much does it need to innovate?
An AI model of an organism is a very-well-specified hypothesis about how that organism thinks and behaves. Like any hypothesis, we assess an AI model by testing its predictions against the performance of the real system and by evaluating the plausibility of its assumptions. The predictions of a model are its behaviour, which we simply record after running simulations. Its assumptions are its components; for example, the computations it makes, the information it has access to, the things it perceives and remembers. We can use standard statistical tests to see how close we come to modelling behaviour in order to argue the validity of our assumptions.
Earlier Work: Artificial Models of Natural Intelligence (AmonI) (my research group at Bath) was dedicated to understanding natural intelligence through the use of simulation and modelling.