The types of cognitive abilities
There is a certain intuition often discussed about the increasing use of large language models (LLMs): that outsourcing cognitive work that requires effort leads to a decrease in our cognitive abilities. In other words, we become less capable of doing things that require sustained attention, effortful thinking, and similar cognitive attributes.
This reminds me of the argument that, as the Industrial Revolution unfolded and led many people to move away from agricultural lifestyles embedded in nature and requiring great physical effort to do even the activities we consider most basic today into cities and the comfort of modern life, we became less physically active, incapable of manual work, and out of touch with our physicality and the environment all around us.
There is truth to these situations. If you don’t use something, you “lose” it, until you practice again and can gain it back over time. The use of LLMs for writing, for example, significantly decreases activity in germane cognitive load, which is the process responsible for turning short-term information into long-term skill.
According to Cognitive Load Theory, there are three types of cognitive load: intrinsic (the intrinsic complexity of the information), extraneous (superfluous elements to learning the topic of interest), and germane (the effort required to transfer short-term information into long-term knowledge). All of these “consume” cognitive/mental effort whenever you are learning something or in a novel situation.
Germane cognitive load is particularly negatively affected when you use LLMs to aid writing and other tasks that require a high degree of cognitive effort. From the study’s abstract: “EEG revealed significant differences in brain connectivity: Brain-only participants exhibited the strongest, most distributed networks; Search Engine users showed moderate engagement; and LLM users displayed the weakest connectivity. Cognitive activity scaled down in relation to external tool use.”
This is interesting to know. At first sight, it may seem like an undesirable result of the increasing use of and reliance on LLMs. Maybe it is so. We don’t know for sure until we discover whether there are also any positive effects resulting from LLM use that may overturn such shrinkage in cognitive load. Maybe the chance for higher volume of deliberate practice, or the faster feedback loops.
Most people don’t care about negative effects discussed in niche scientific publications that they don’t directly experience themselves in some significant manner that impacts daily life. We are addicted to convenience, and continue to vow for convenience with our choices every minute. Yet, convenience comes with downsides and long-term consequences hidden in plain sight.
A human is foremost an animal, an ape, a mammal, not to be confused with an artificial creation made in a tech lab. Animals like us have some fundamental biological truths that don’t go away if we pretend they don’t exist. Instead, they roar to be seen and acknowledged, increasingly strongly, until some kind of explosion occurs (whether internally or externally, large or small and incremental).