Human-Like Artificial Intelligence through modelling Motor Decision Making

In psychology, there is a growing view that 'higher order' human cognition is built upon sensorimotor foundations. Two hundred million years of our evolutionary history was driven through natural selection of those species best able to interact with their physical environment. It is only in the last 6.5 million years that humans have developed higher-order cognitive abilities. The brevity of this evolutionary period suggests that the foundations of human cognition rest upon the neural architecture that underpins the sensorimotor system. Ontogeny appears to recapitulate phylogeny in this regard, with a wealth of evidence showing that a child's physical interactions with the world lay the foundations for their higher-order cognitive capabilities. For example, the decisions made when selecting actions have components of inhibition, working memory and set shifting - elements that are generally agreed to be the hall mark of human cognition when applied to abstract computational tasks.

From this viewpoint, we reason that studying human sensorimotor decision-making can create the necessary foundation to build a general understanding of human cognition. Activities such as deciding which pots of yoghurt need to be removes to allow access to the milk bottle in the cluttered fridge, and then generating the appropriate movements to grasp the pot safely, could present important insights into the mechanisms driving human planning and action. The student will develop computational models based on empirical data from decisions on human motion so that these models can be used for planning robot motion. This work will help inform the development of robotic systems that can plan and manipulate objects in a human-like fashion and, importantly, are able to explain their actions.

While this research will be limited to sensorimotor decision making in the domain of cluttered environments, the findings will have broader implications for our understanding of human cognition and the feasibility of human-like AI. To this end, we have identified three research questions which will guide the programme of research:
1. What information do humans use to plan and act upon objects in cluttered environments?
2. How do these capacities develop through childhood?
3. Is it possible to assimilate parameters from these findings into a hierarchical model so that robots exhibit more human-like behaviour in novel reaching tasks?