Recently, my colleagues and I published a “commentary” titled “Beginning with biology: ‘Aspects of cognition’ exist in the service of the brain’s overall function as a resource-regulator.”
In a commentary, authors may challenge, elaborate, or extend ideas that have been recently published in an academic article. A commentary is similar to a book review for academic articles, though commentary authors are expected to riff - not just reflect- on the ideas presented in the parent article.
In our case, we were commenting on a new article by Falk Lieder and Thomas L. Griffiths published in the journal “Behavioral and Brain Sciences” titled “Resource-rational analysis: Understanding human cognition as the optimal use of limited computational resources.” Lots to unpack there.
In their paper, Lieder and Griffiths address a critical challenge in cognitive modeling. I’ll explain this challenge in a moment, but first let’s take a step back. What is cognitive modeling?
Cognitive models are computerized models that attempt to simulate cognitive processes. The computations underlying cognition remain poorly understand - in part because it’s hard (and expensive) to observe the brain working in real time. Cognitive modeling is useful because researchers can make hypotheses about what they think the computations underlying a specific type of cognitive process are, formalize those hypotheses into algorithms, simulate that process, and then compare the results of their simulations with data from human subjects. If the data produced by their simulations approximates the data produced by human subjects, then the researchers gain some evidence in favor of their hypotheses.
Lieder and Griffiths address a critical challenge in cognitive modeling: simply that any given cognitive process may result from seemingly infinite causes - how do you choose which ones to test? To limit the hypothesis space, Lieder and Griffiths introduce a new framework: “resource rational analysis.” The central argument of the resource rational approach is that cognitive processes are constrained by resource usage, and that this insight can help researchers to select among potential cognitive models.
Here are the five steps to Lieder and Griffiths’ “resource rational analysis:”
1. Start with a computational-level (i.e., functional) description of an aspect of cognition formulated as a problem and its solution.”
2. Posit which class of algorithms the mind’s computational architecture might use to approximately solve this problem, the cost of the computational resources used by these algorithms, and the utility of more accurately approximating the correct solution.
3. Find the algorithm in this class that optimally trades off resources and approximation accuracy (Equation 3 or 4).
4. Evaluate the predictions of the resulting rational process model against empirical data.
5. Refine the computational-level theory (step 1) or assumed computational architecture and its constraints (step 2) to address significant discrepancies, derive a refined resource-rational model, and then reiterate or stop if the model’s assumptions are already sufficiently realistic.
Our commentary homed in on steps 1 & 2.
In steps 1 & 2, Lieder and Griffiths advocate that researchers begin by identifying a cognitive process (for example, memory) before then considering how that “problem” may be solved in a way that efficiently uses resources (for example, metabolic resources) and approximates accuracy. The problem, as we propose in our commentary, is that the brain did not evolve for cognition… the brain evolved to help regulate the physiological resources of the body (for a cool conversation focused on the point about ‘accuracy’ - check out Anil Seth’s recent interview with Aeon). What we call “cognitive processes” are merely mechanisms the brain uses to achieve this fundamental prerogative.
“At its biological core, life is a game of turning energy into offspring” (Pontzer 2015), meaning that if Lieder and Griffiths wish to model how a brain works, then resource usage must be a central (if not the central) computational concern. Brains did not evolve for animals to think or see or feel – they think, see, and feel because doing so regulates a body with resource-hungry systems.
Consequently, we argue that researchers ought to begin with the biology, and then consider how cognitive processes help achieve biological goals rather than vice versa.
Psychology has a long history of studying what are called “folk psychological categories.” Folk psychology (also called common sense psychology, naive psychology, or vernacular psychology) defines a set of categories in everyday language and based on everyday experience (e.g., emotion, cognition, perception). We experience these mental phenomena as having different qualia (e.g., we may experience emotion as different than “pure rational thought”). But just because we experience these phenomena as different, does not mean that they are indeed fundamentally different biological processes. Studying folk categories thus risks oversimplifying, or indeed misrepresenting the concepts we are trying to examine. As humans (subject to human biases) studying humans, researchers are vulnerable to these oversights. A great example of this comes from the world of motor movement.
For a long time, researchers attempted to understand how infants learn to walk by having children complete what’s called the “straight-path” task. The “straight-path” task is exactly what it sounds like: children are asked to repeatedly and continuously walk a straight line over a uniform ground - it stems from the common intuition that “successfully walking” often feels (to adults) as being tightly controlled and linear. However, when Karen Adolphs, a researcher at NYU, observed infants walking spontaneously during free play she observed several characteristics of infant walking that were obfuscated by the straight-path task. Infants moved in short bouts, in curved lines, and in multiple directions…. not really in straight lines. In fact, these patterns continued across development. It turns out, people don’t really walk in straight lines. Adolphs wrote in a paper published in 2017 titled, “The cost of simplifying complex developmental phenomena: a new perspective on learning to walk:”
We propose that a focus on spontaneous walking, the phenomenon we ostensibly wish to explain, yields important insights into the problems infants solve while learning to walk.
A lot sits in that “ostensibly” - a frustration in scientists studying phenomenon that they experience as being “real” without taking a step back to observe that phenomenon “spontaneous[ly] unfold.”
Our commentary makes a similar argument: if we “ostensibly” want to understand resource-rational cognition, a focus on biology offers important insights into the metabolic problems “cognition” is evolved to help solve.