Matthew Botvinick represents an eye opening cup of java. Building on the foundations of cognitive psychology, Botvinick’s laboratory works at the intersection of neuroscience, psychology and computer science, seeking to clarify the computational and neural foundations of human behavior. They employ a diverse set of research tools, including functional neuroimaging (fMRI), behavioral techniques (reaction time, error, and decision analyses), and computational modeling (neural networks, reinforcement learning models, and belief nets), typically applying multiple techniques to a single problem.
They are leveraging these tools to investigate a range of specific research questions, spanning the topics of cognitive control, working memory, decision making, sequential action, and language processing. Current projects include the monitoring and control of cognitive processing, the control of sequential behavior, and the representation of sequential order in working memory.
At the May 6 Lunch ‘n Learn seminar, Botvinick placed his work within the context of developments within Psychology. At the onset of Psychology as a field, he began, one of the central developments was introspection, observations and reports on the contents of the unconscious mind. That work generated interesting findings and got the field rolling, but Psychology concluded that this early approach provided no meaningful mechanism to confirm or deny such subjective hypotheses. Simply put, there was no objective data to analyze.
Behaviorism emerged, providing data from observable actions and reactions, an attempt to apply the methods of the natural sciences to the study of human and animal behavior. B.F. Skinner and his contemporaries worked with animals, mainly rodents, placing them in boxes with levers, and rewarded them with various treats. It was an article of faith, says Botvinick, that behavioralists would talk about observables, the stimulus conditions and the resulting behavior, but not what was happening inside the subject animal. The key theoretical construct in behavoralism was the construct of the stimulus-response association. It was a view of the function of the mind and the nervous system that had to do with direct observation of the relationship between stimulus events and observable responses.
Voices of dissent, even in those early days, notably Edward Tolman, pointed to empirical phenomena that could not easily be reconciled without thinking about what was happening within the subject animal. A rodent in a maze, for example, appeared to maintain a cognitive representation of the maze, not a simple stimulus or observable response.
And then came the computer, spurring a cognitive revolution for Psychology says Botvinick. Here was a machine that contained internal representations, patterns of electrical activity within the machine representing information and which determined the machine’s responses.
Early on, the new field of cognitive psychology used the computer as a metaphor for the mind. It could store short and long term memory, it had control elements, a central processing unit. And within its software, it had the ability to achieve tasks algorithmically. Some have even held that the machine well mimics human decision making processes. Do computers and neurons equivalently process information; Can we write programs that mimic human response to stimulate? These questions spawned research into neural network models and several other approaches in computational neuroscience.
To give a sense of what researchers can derive from computational models, Botvinick presented one case study of computer modeling from his own work, a study of the seemingly mundane task of making a cup of coffee. Using the computer more as a tool than a metaphor, modern machines implement models of such behavior.
Looked at in detail, says Botvinick, the study of coffee-making provides significant insight into how people function, and why and how they occasionally fail to accomplish even simple tasks.