This post was contributed by Nick Sexton, PhD student in the Department of Psychological Sciences
Prof Rick Cooper
How can computer simulations help us understand the human mind? That was the main topic of the Rick Cooper Inaugural Lecture, in which Professor Cooper outlined 15 years of research on cognitive computational modelling.
Cognitive computational modelling boils down to designing computer simulations of how the mind processes information. While computers that appear to think in a human-like-way (whatever that means) are increasingly commonplace in our everyday lives – driverless cars, the Google Deepmind model which learns to play Atari games, and intelligent personal assistants, are all examples – the talk revealed that a more difficult challenge is not only to mimic (or improve on) human behaviour, but to produce it in the same way that humans do – using the same types of mental process.
For example, certain computer programs have succeded in being indistinguishable from humans on Alan Turing’s classic test of artificial intelligence: however, when one digs under the surface, it is readily apparent that their responses are generated in a not remotely human-like way.
So if modelling how the human mind actually works is tricky, how does one go about doing it? Cooper’s approach is to build on theories of how the mind works, from cognitive psychology, often pieced together through painstaking use of behavioural experiments on human participants. These theories, describing how the mind processes information, often resemble flow-chart-like schematics – but often the details are left vague.
This is where cognitive modelling comes in – a fully operational computational model must provide exact details on the inputs, outputs, and algorithms computed, at every stage of mental processing, so the modeller must fill in details that the theorist has left blank. It is a test of whether the psychological theory really is sufficient to explain what it purports to explain, and if not, suggest what details it might be missing.
One element that makes Cooper’s research stand out is his focus, not just on abstract tasks conducted in a sterile psychology or neuroscience lab, or even on a less defined realm of behaviour, as in the Atari game player – but on distinctively human, often startlingly everyday behaviour.
For instance, a large amount of what we consider normal human behaviour is routine – habitual actions, like preparing meals or hot drinks, dressing, commuting. One particular branch of Cooper’s modelling work has been on developing a computational theory of how the mind accomplishes routine actions with minimal attentional oversight, and how this mental apparatus can be applied to non-routine situations.
One model of routine everyday actions simulated preparing drinks. It manipulated objects in its (virtual) environment, like utensils (cups, knives, juicers) and resources (such as hot water, coffee, tea, milk, sugar, oranges )- to achieve an end goal – such as preparing coffee(milk no sugar). The model needed to account for normal human behaviour – successful preparation of the drink most of the time, with occasional lapses – sometimes forgetting to put milk in the coffee, or adding sugar when it wasn’t required.
So what is interesting about a model which prepares drinks (sometimes badly)?
Well, the model was also able to explain what happens when normal mental processes break down – say, in the event of brain damage. With certain setttings, the model not only simulated the lapses of neurotypical people, but also the more extreme lapses observed in
patients with particular types of brain damage – putting butter in the coffee, or forgetting to add water, say.
The model was also able to simulate the behaviour of patients with specific conditions – Ideational apraxic patients struggle to retain a sense of an object’s purpose – say, trying to use a fork to cut an orange. Patients with utilisation behaviour tend to perform actions
appropriate to a given object, but inappropriately to the current situation – take off your glasses and hand them to the patient, and they are liable to put them on.
Here, a cognitive model is rather more use than more everyday artificial intelligences which perform everyday tasks, such as Siri – because Siri might ‘think’ in a way completely differently to humans, there is no reason to believe that if we deliberately damage part of the program, she will produce behaviour typical of people with brain damage. However, because Cooper’s model was based on neuropsychological theories where routine actions depend on the correct interaction of different cognitive processes – simulating damage to specific processes in the model was able to account well for the
differrent patterns of behaviour typical of different neural conditions.
This approach isn’t just useful for understanding what might be damaged in people unfortunate enough to suffer brain damage, then – it is also a powerful tool for trying to understand what role those cognitive processes play in the human mind when it is functioning normally, and whereabouts in the brain they might take place.
The hour-long talk gave a fascinating glimpse into how – as the knowledge gained from the brain and mind sciences continues to accelerate – computational cognitive modelling has an important role to play in drawing together different disciplines – taking cutting-edge research in psychology, neuroscience, and machine learning – showing how the individual pieces fit together, to give us a better glimpse of the overall picture of how our minds work.
Find out more
