In developing artificial intelligence (AI) agents for the real world, we generally think that the artificial part will be the most difficult. After all, it has taken considerable technical advancements across several fields for us to begin building intelligent systems. Instead, the most challenging part is intelligence, a word that has proven ephemeral over decades of research. In this thesis, I define intelligence as the ability to adapt behaviours to achieve goals in a range of environments. Although this is a broad definition, in our quest to build AI agents we have an excellent source of intelligence that we can study: the human brain.
The brain is capable of producing intelligent behaviours in spite of a number of limitations, including the limited resolution of sensory systems and the need to discretize information. In this thesis, I demonstrate that we can study how the human brain produces intelligent behaviour by using rapid motor behaviour tasks, which require decision-making and motor control to occur in tandem. Using a large experimental data set, I establish that human rapid motor behaviours are fundamentally constrained by capacity limits. I use this insight to propose discrete-event systems (DES) models that incorporate capacity limits to model the brain’s success at producing intelligent behaviour.
Taken together, this thesis develops the argument that in order to build AI agents we must first understand intelligence. This requires us to account for an agent’s limited capacities and then ensure that these limits are incorporated into our models. This is the first work to use DES theory to model biological intelligence and I believe that building links between these areas will help us to develop control theory models inspired by the human brain, thereby allowing us to build AI agents for the real world.