Working memory enables humans to manage and juggle multiple pieces of information simultaneously in short-term scenarios. For instance, it allows individuals to mentally create a grocery list before going shopping or recall phone numbers when making calls. Despite the agreement among scientists about the limited capacity of working memory, there is still debate over its underlying mechanisms.
New research from scientists at Brown University’s Carney Institute for Brain Science has shed light on why these limitations exist. Michael Frank, a professor affiliated with the institute, and Aneri Soni, one of his graduate students, developed a computer model that replicates brain functions associated with working memory—the basal ganglia and thalamus.
Their study published in eLife reveals that learning is the reason behind these limitations. Their simulations show that if individuals attempt to hold more than just a few items simultaneously, it becomes difficult for them to learn how to manage this information effectively. This results in confusion and hindered usage of stored data.
The research also indicates that when faced with such constraints, the brain learns strategies to conserve space by utilizing mechanisms similar to those used during “chunking.” Chunking is a process where related pieces of information are compressed together, making them easier to store and manage.
Because dopamine plays an important role in learning-related working memory functions, these findings offer new insights into dopamine-related disorders such as Parkinson’s disease, attention deficit-hyperactivity disorder (ADHD), and schizophrenia. The team reached their discovery through building a new computer model of the brain that mirrored experiments conducted with humans by researchers from Frank’s lab and Matt Nassar’s.
The previous experiment demonstrated human capability to “chunk” information for better management in working memory. Soni confirmed her model was successfully replicating brain-like functions when she tasked it with recalling colored block orientations after showing different colors on a screen, resulting in the strategic grouping of similar colors over multiple trials.
By running these trials on models both with and without chunking mechanisms but ample storage space, Soni discovered that while those capable of chunkting utilized their full capacity strategically, others lacking this mechanism did not recognize available storage effectively. This led her to conclude that learning rather than mere capacity is the real driver behind working memory.
A crucial component enabling the model’s learning process was a system emulating the human brain’s dopamine delivery. When effective recall performance occurred due to chunking similar items together for space conservation, this artificial dopamine release reinforced continued use of such strategies in future trials under identical constraints.
In additional experiments, Soni manipulated the model’s dopamine delivery systems to reflect conditions observed in patients with Parkinson’s disease, schizophrenia, and ADHD. These modifications resulted in poorer learning capabilities and less frequent chunking when faced with similar tasks.
Frank concluded that findings such as these could significantly advance psychiatry through computational brain science: “Take Parkinson’s disease as an example,” he stated. “Most people think of it mainly as a movement disorder due to obvious changes in motor function, but actually there are also significant issues with working memory among patients.
Currently, treatments typically target the prefrontal cortex for such cases. However, our findings suggest that we might want to explore whether drugs targeting basal ganglia and thalamus could enhance symptoms better.”
Frank believes a deeper understanding of dopamine-related processes within these brain regions in patients with disorders may lead clinicians towards adopting alternative treatment strategies.
This research was supported by the Department of Defense (ONR MURI Award N00014-23-1-2792), the National Institute of Mental Health (R01 MH084840-08A1, T32MH115895) and computing hardware was supported by NIH grants.
