Fractal agency and confidence

One of the most interesting frames I've come across sees the mind as made up of multiple distinct agents, each with their own set of subagents and influence over the whole.

I first came across this idea about a year ago, reading Kevin Simler's post "Neurons Gone Wild" on his blog Melting Asphalt. In it he describes how different subagents in the mind compete for influence over the 'self.' He even goes further and claims that agency in the brain is hierarchical, in that components of subagents in the mind can individually demonstrate agency, and so too can the subcomponents of those components, all the way down to the level of a single neuron.

A lot of his ideas were directly influenced by Daniel Dennett, who had some really interesting things to say about neurons. As Dennett sees it, neurons have evolved into powerful computational blocks because they (along with every other cell in our body) descend from unicellular eukaryotes which had to fend for themselves, and so they've learned to be scrappy and "slightly feral," in his words.

Each neuron struggles to survive; if it isn't in use, it won't get neuromodulated, its receptors will disappear, and it will die. So the selfishness of a neuron motivates it to join a useful circuit and stay useful. An interesting example of this theory's explanatory power is in neuroplasticity. Under this lens, neurons struggle for resources, leading to selfishness, scrappiness and ultimately a high adaptiveness which we call neuroplasticity.

This idea that individual neurons are agents might be one of the most interesting I've encountered, and one of the most important in neuroscience (I'm no expert, but it certainly seems substantial enough). On top of the myriad of implications it has for the field, it also supports the idea that the brain is not composed of just a "simple computational" substrate; its far more hospitable to agency than a computer because its very building blocks are agents.

I've found that this frame fits neatly with Scott Alexander's descriptions of motivation in his Astral Codex Ten post, "Towards A Bayesian Theory of Willpower." In it, he describes how willpower may arise from the sum of influences of each subagent in the mind. This theory explains natural phenomena really well – since I'm not nearly as great a writer as Scott, let me take from him in giving an example:

A tradition originating in psychotherapy ... interprets willpower as conflict between mental agents. One "subagent" might want to sit down and study for a test. But maybe one subagent represents the pressure your parents are putting on you to do well in school so you can become a doctor and have a stable career, and another subagent represents your own desire to drop out and become a musician, and even though the "do well in school" subagent is on top now, the "become a musician" subagent is strong enough to sabotage you by making you feel mysteriously unable to study.

It also works well with the idea that dopamine is the currency of evidence the brain -- whichever subagent presents the strongest evidence for a particular action wins, and you end up taking that action.

Scott's theory differentiates between a few types of subagents: one is biased towards inaction (why do anything at all?), one is biased towards maximizing reward (like a reinforcement learner), and one contains your high-level, conscious calculations on what you think you should do. These subagents, among others, compete to convince your brain to take a certain action.

An interesting caveat here is that

sufficient evidence doesn't necessarily make you do something, but overwhelming evidence sometimes does.

A cig addict might know the risks of lung cancer and death if they continue smoking, but this evidence isn't felt strongly enough in their brains for them to quit. This is why successful quitters often mention facing "rock bottom" before resolving to quit – it provides strong enough evidence in their brain to override the reinforcement learner craving another hit.


So, what can we do with knowledge of this theory? In my last post I talked about confidence and its importance. In this post, I want to explore how the above theory of mind can help us understand aspects of how confidence is manifested in the brain, and how this understanding can help us troubleshoot a lack of confidence.

My hunch is that a person's confidence is determined in part by the extent to which the subagents in their mind are aligned.

Consider again the example I quoted from Scott Alexander above. A student might find themselves procrastinating on their studies without understanding why. But internally, the lack of action arises from conflict between the "do well in school" and "become a musician" subagents in their mind.

The former sets an intention to study, while the latter proposes practicing guitar. Since the school subagent is "on top of mind," it is able to shut down the proposition to practice guitar. However, the musician subagent is just strong enough to overpower the school subagent's intent for action using their proposition to practice. So the brain stalls as the two fight each other's intentions, and the result looks a lot like procrastination.

Setting a conscious intent for an action and then failing to act is highly demotivating, and certainly delivers a blow to one's self-confidence. On the other hand, if the student in the example was actually a student of music, and their musician parents supported their musician dreams, then the "do well in school" subagent wouldn't be at odds with the "become a musician" subagent at all. In fact, the two would be in perfect harmony with setting an intention to practice guitar. With no subconscious conflict to bog them down, the student would actually practice. And, having carried out a conscious intention of theirs, their self-confidence would increase a bit because they just reinforced their sense of agency and willpower. This is the basic idea behind my hunch.

If we were to consider a person's mind tabula rasa, with no conflicting subagents, we should assume that their sense of self-confidence is at a maximum. I know confidence and self-esteem are learned and not innate, but this is still pretty consistent with my experiences with toddlers and small children – they DGAF! They, for the most part, act on their immediate desires and speak their mind with no filter. It's only when their parents start instilling them with social constructs and etiquette that we see their behavior start to change. They learn to say "please" when asking for things, and "thank you" when they receive them. Or, they learn that at a house party it's more polite to ask the host where the bathroom is than to simply state that they need to pee. As this social training continues, a sort of "conform to society" subagent begins to form, and moderate the intentions of other subagents in their mind.  

I think the way to apply this knowledge in order to improve your own self-confidence is to resolve conflict between conflicting subagents. How to go about doing that could be an entirely separate post, but thankfully that work has already been done for me by another (Kaj Sotala). As I understand it, my hunch falls pretty much in line with the predictions of the Internal Family Systems model (IFS). If you're unfamiliar, Sotala does a great job building up the intuition behind IFS in his LessWrong post about it. More presciently, he has written about integrating disagreeing subagents, which I think is a key aspect of regaining self-confidence. I highly suggest checking it out if you're interested in this concept of subagents, and want to explore practical methods of resolving inner mental conflict.


That is all! If you enjoyed, please consider subscribing below. It's free (unless you want to pay)!

P.S. if any of this resonated with you, I'd love to know! Hit me up.