People have a phenomenal capacity for learning about the world, allowing us to store concepts that tell us the meaning of words and the properties of objects (e.g., that pianos come in many different shapes and sizes but can all be played to create music). This conceptual knowledge is critical in everyday life. Without it we would be unable to interact with objects or communicate with others. Yet simply storing this knowledge is not sufficient, it is critical to access and use the right bits of information to produce behaviour that is appropriate in a given situation (e.g., understanding the weight of the piano in order to move it but not when playing it). In a recent collaboration between scientists from the MRC CBU and the University of Wisconsin-Madison, Dr Becky Jackson built a computer simulation able to do both these things; to store concepts and to use them flexibly.
Instead of simply building one model and seeing whether it worked, we built several models, with different features. This allowed a systematic comparison of the models with and without these features. If the model with a given feature was better at learning and using concepts, then that feature is considered helpful for processing conceptual knowledge. Remarkably, the features that were found to be useful were similar to the anatomy of the brain system responsible for learning and using concepts. For example, the simulations only did well if information from all senses could be brought together and processed in one region. The same thing happens in the brain; after basic processing in visual and auditory regions, the same brain area accesses conceptual knowledge of a dog based on its sight or the sound of the word ‘dog’. The best model was able to simulate different aspects of this brain network, most critically the effects of brain damage. Damaging this model in different ways gave it similar problems in understanding concepts as those caused by stroke or dementia. This research helps us understand how the brain is structured to support our phenomenal ability to comprehend the world and how this breaks down after brain injury.
The paper can be read here: Jackson et al. (2021) Reverse-engineering the cortical architecture for controlled semantic cognition, Nature Human Behaviour