Advancing the field of cognitive science: using cognitive models to help us better understand patients
Picture a typical visit to the doctor’s office. A patient walks into the office, describes their symptoms and comes out with a prescription which they take to their neighborhood pharmacy to be filled. A few weeks later, they come back in for a check-up. The doctor asks: Did you take your medicine twice daily in the morning and evening as instructed? Is the medicine helping to solve what you came in for? I think so, a little…maybe?
A major limitation of the traditional medicine model is that once the prescription is written, doctors typically have very little insight to how patients are adhering to the treatment and often have to rely on subjective reports to assess whether the medication is working.
Digital Therapeutics (DTx) – clinically-validated treatments delivered through a digital interface – are uniquely positioned to help close this gap and provide health-related insights derived from rich data on exactly when and precisely how patients are engaging with a treatment. At Akili, we build DTx that are designed to assess and train attention function. The rich data we collect make it possible for us to infer changes in our patient’s cognition (and by association, their brain function) each time a patient engages with the treatment.
In the past few decades, the field of Cognitive Science has made tremendous advances in our understanding of how our brains give rise to our everyday behaviors. One of the approaches that has made this possible is the development and application of mathematical models to explain how cognition works.
Descriptive cognitive models are typically written down in plain English and thus subject to many different interpretations depending on the reader.
Computational cognitive models (CCMs) are written down in mathematical expressions which are identical regardless of who is reading them. Typically, the goal of these cognitive models is to explain how a cognitive process – like attention or memory – works, and how that cognitive process relates to human behavior.
Both formal and informal cognitive models are just theories about how the brain works. However, with recent advances in machine learning and the availability of enormous amounts of computing power, CCMs are poised to help us make enormous strides to understand the inner workings of the human mind and how we might fix them when things go awry (for example, in the case of cognitive/behavioral disorders like ADHD).
Over the past few years, Akili has invested in building and improving these powerful tools to better understand how exactly our digital therapeutics improve cognition today as well as how we might improve them to be even more effective in the future.
We’ve partnered with PyMC Labs, a Bayesian consultancy firm with deep expertise in CCMs, to help us build state-of-the-art tools that allow us to test different kinds of models in a very efficient and scalable way.
Our most recent collaboration focused on solving two things: speed and flexibility. We now have a very flexible and very fast set of tools to use to develop nearly any cognitive model that we can write down, which means we now have the toolkit to better understand and address patient needs, motivation and behavior.
Increasing speed: These models can be complex and thus are often very (VERY) slow to fit properly, which limits how quickly we can iterate and ultimately how we’re able to use them.
Improving use cases and flexibility: Historically we’ve been very limited in the kinds of models we can use (see this research article for more information on this topic) and now we have the tools to fit a much (MUCH) wider variety of models.
The work Akili is doing today will help advance the field of cognitive science.
"While this work provides many benefits to Akili, its total impact extends far into the broader cognitive science community. The ability to fit this rich class of models allows us to answer scientific questions about the mind that were previously unattainable to ask,” said Thomas Wiecki of PyMC Labs.
We’ve already seen the tremendous potential of this approach in building better digital therapeutics–and this is only the beginning! As we continue to make progress, we’ll continue to update you on how CCMs can help us build and iterate on digital therapeutics.