Barrett-Concepts are Levin-Memories
A synthesis of some ideas found across the work of Lisa Feldman Barrett and Michael Levin.
When you find yourself in any affectively relevant scenario, you rely on your past experiences to choose actions to address the present situation based on their anticipated consequences. Doing this successfully in a complex environment requires two things, concepts and memories, which are actually the same thing.
A concept is basically a goal-oriented featural assessment of an object, system, or structure. The concept of a vegetable in regards to, e.g., a tomato emphasizes features of the tomato such as its taste and texture in regards to the goal of using it in a salad or a sandwich, whereas the concept of a berry in regards that same tomato might entirely neglect its taste and texture while emphasizing instead the tomato’s botanical origin as being produced from a flower with one ovary, features that are entirely neglected by the concept of a vegetable.
Concepts are really useful for forming plans for how to address an affectively relevant situation because they greatly simplify what needs to be specified at a high level. For example, if you’re attacked by a tiger and want to run away, the specific act of running is highly detailed, consisting of every single aspect of the trajectories of your arms and legs, as well as controlling the necessary changes in your lungs, heart rate, blood pressure, sweat glands, etc. This is a lot to think about. It would be better if you could just form a high-level concept of running and let that concept propagate downstream to produce the relevant outcomes. Not only is this simpler and frees up your brain to stick to its comparative advantage in high-level planning while other systems in your body deal with the details, but it also allows for a more context-sensitive instance of running to emerge as the various parts of your body do the best they can based on available internal resources and the conditions of the environment—e.g., your feet are tuned to the slope of the ground in a way that your brain isn’t.
It’s intuitive to think that concepts pertaining to a tiger attack, like tiger, attack, predator, run, fear, death, etc., are stored in the brain somewhere, waiting to be activated by a relevant situation. But this isn’t true. Instead, concepts are assembled in an ad hoc manner, built out of psychological primitives in the brain. This is very useful for a few reasons. First of all, stored concepts are a waste of space, taking up valuable neurons dedicated to storing them that could be doing something else instead. Second, stored concepts are context-insensitive. Concepts that are stored from a previous situation are based on that previous situation, but previous situations are always significantly different from the current situation. Deploying an instance of a concept that was built to address a situation in the past won’t be optimal for addressing a situation in the present or an anticipated situation in the future.
It’s also unclear what a stored concept of e.g., running even is. To understand why, it’s useful to distinguish between a concept and an instance of the concept. Running is a concept, while any particular act of running is an instance of the concept. The brain can’t just store the concept of running because the concept alone doesn’t specify anything that would help the brain form a specific plan to address the current situation because surviving a tiger attack doesn’t call for the general concept of running, it calls for an instance of running. But there is an exponential amount of instances of running that all differ from each other in various ways, like one instance of running may be faster, while another may be quieter, as well as subtle but important differences in terms of limb trajectory, muscle activation, and even which neurons synapse to which muscles. There wouldn’t be enough room in the brain to store all instances of even a single category, let alone multiple categories. Also, each instance of running depends on environmental conditions as well as internal ones. You can only deploy an instance of running uphill when there is a hill to run up. Activating the correct stored concept therefore requires already having the conceptual knowledge as to which concept is the right one to activate, which defeats the whole purpose of storing concepts in the first place.
The alternative to using stored concepts is to construct an ad hoc category that didn’t exist until it was created. These concepts are assembled out of psychological primitives whenever they are useful and can be constructed efficiently. Because these concepts are constructed in an anticipatory manner rather than a reactive one, they’re able to be constructed in time to address the oncoming situation, and because these concepts are constructed based on available internal resources, they will be feasible.
Which concepts will the brain construct ad hoc? This depends on memory. Basically, the idea is that the brain constructs concepts that worked well in situations similar to the situation the brain is anticipating. But since the concepts the brain can construct aren’t stored anywhere, it seems almost paradoxical as to how the brain can know which concept to construct before the concept is present to be evaluated for its fit for the situation at hand. However, cells have memories that can be exploited by cellular communities to do work, which suggests a more organic view of how concepts are constructed. Similar situations will be experienced similarly by neurons, so the large, densely connected neurons that exist at a high level in the brain will remember what they did in previous situations and generate combinations of signals that are not necessarily the same but are expected to produce functionally similar outcomes. Lower-level neurons will remember how they responded to similar prediction signals in the past and construct functionally similar motor behaviors and perceptions. Thus, the memory/concept constitutes a biasing of the action and perception landscape for the brain as a whole, determining which motor behaviors and which perceptions are constructed and facilitated without having to micromonage them.
Just like concepts, memories aren’t stored in the cell to be accessed upon need but are constructed on the fly whenever they are expected to fit the situation. The fact that memories aren’t fixed objects allows them to be translated from one substrate to another, and from one context to another, allowing for memories to be used to determine optimal behavior even in novel situations.
Concepts and memories are similar in at least the following ways:
Constructed, not stored. Concepts and memories are not stored in the brain to be accessed later but are constructed on the fly whenever they are expected to be relevant for a situation. This allows for the brain to economize on space while also producing concepts and memories that are more appropriate to the specific and virtually always somewhat novel situation, and producing them more quickly as well.
Degenerate, not one-to-one. Concepts and memories are degenerate, meaning there is no single fixed way of making them, be it specific neural pathways, specific neurons or neuronal ensembles, or even specific neural patterns. Degeneracy allows for much more flexibility and robustness in creating concepts and memories.
Confabulated, not recreated. Because concepts and memories are created anew for new situations, and because the internal conditions of the body are never the same to what they were previously, the concepts and memories created are never identical to previous concepts or memories respectively. We lump them together with previous instances because they are functionally similar, but they are never identical. This allows for solving new problems and addressing new situations.
Built for action, not for thinking. When the brain forms a concept of a situation, it does not do so for accuracy in whatever nebulous “purely intellectual” sense but does so based on its utility for acting to successfully regulate affect. Similarly, memories are not constructed to accurately reproduce previous conditions but again to optimize for anticipated conditions in the future.
Concepts and memories are two different words for the same thing: systems of relationships constructed by the brain (or some analogous structure) that fit the situation at hand in terms of biasing action and perception to regulate affect (or at least achieve allostasis if we are unsure whether all systems have affect). Concepts and memories are based on the past but not limited to it; they draw on the flexible, context-sensitive behaviors of the parts to achieve familiar yet novel problem-solving capabilities in new situations.