Emotional Superrationality

post by nullproxy · 2025-01-02T22:54:53.303Z · LW · GW · 4 comments

Contents

4 comments

I'm going to start with a big claim here; your emotions are not irrational, they're superrational. You're using them wrong.

Superrationality is the term for perfect rationality, the state which maximizes utility because it assumes that all other players are also superrational, and thus all superrational players will come up with the same strategy if they assume that all other players are superrational too. Different choices are impossible, because each player is choosing for all players utilizing perfect rationality. The prisoner's dilemma effectively disappears under this assumption; a superrational player only has to think "If I defect, they defect. But if I cooperate, then they cooperate. Therefore I'm really choosing between (defect, defect) and (cooperate, cooperate)." Since (cooperate, cooperate) has a better payoff than (defect, defect), a superrational player chooses to cooperate.

In this case, you'll notice that different choices are impossible because the choice that emerges is most beneficial personally. It's also most beneficial societally, but that's somewhat of a secondary benefit. It's most beneficial personally because you are inherently drawn to staying in a state of superrationality, if you can get there. If you're in a state of superrationality, you have to exert less effort to get an outcome that benefits you personally, you experience less fear of other agents, you don't have to work as hard at your internal calculations, and you can navigate to a pleasant reality where you have more time outside of prison to do whatever you want to. These are all good things. These are not only personally beneficial, but also have a cascading effect on other agents. When other agents see your strategy, they are drawn to cooperate with you and attempt to align with your superrationality, because you have found a truthful and sustainable method of self-stabilizing coordination that feels good to be in. A state that feels better than defecting does.

But this is assuming that the opposing player is superrational. And this isn't true in the real world, not by a long shot.

Or is it?

Empathy is the ability to share another person's feelings, experiences, and emotions. It's the ability to "put yourself in someone else's shoes" so to speak. Often, this is a process that is semi-automatic, inherent, can be developed, but requires an increase of emotional intelligence. Things like mindfulness, meditation, spending time with other humans in healthy relationships, experiencing safety, integrating novelty and new experiences...this can all potentially have an effect on your capacity for empathy. 

Something much more fundamental that has a direct and unavoidable impact on your capacity for empathy is how well your needs are currently being met. The Hungry Judge Effect shows that judges are more lenient when sentencing after a meal break. This could be interpreted as an influence of a subconscious need (hunger) on the judge's capacity for empathy. The only reason that this need has an influence on the judge's capacity for empathy is because the judge is operating within a system (a human body) and the need is not currently recognized consciously, thus it is known by the hunger signal subsystem of that judge's brain that the information it carries is not being acknowledged, and therefore it needs to exert more control over the processes it has access to, in order to communicate to the self (the judge) and society (external reality) that there is a need that is being unfulfilled here for this organism. The reason that the need is allowed to stay subconscious and exert the low-level influence it has over the judge's behavior is because that is the scale at which the need is operating at. The judge isn't that hungry, but the judge is a little hungry, and the hunger signal subsystem doesn't know how relatively hungry or not hungry the judge really is, because the hunger signal subsystem only has access to the hunger signal, not the entire reality of the system to weigh comparative needs (the self does that). Once the need becomes sufficiently salient enough to overwhelm the threshold of making it into conscious thought (in which case more and more subsystems tune into the fact that the hunger signal is affecting them as well, lending strength to the hunger signal subsystem through their coordination of resources), then the judge would think "man, I'm hungry, I should take care of that" and this thought would assist in the internal coordination necessary to move the judge to a situation where food could be acquired.

So, empathy decreases the worse your needs are met. This makes computational sense, if you view a human as a system. When you have unmet needs, your focus should be directed internally, in order to assess what is needed, gather the resources and cooperation to coordinate external action, and then take the action that's needed in order to move yourself to a position where those needs can be most accurately met. This is a sign of a rational agent. 

Where does the irrationality come from, though? Because irrationality is everywhere, or at least it feels like it is.

Imagine that we create a parallel between two subsystems of the brain and two people. We're just playing with these hypothetical people, or the variables of X and Y brain subsystems. In what situation could superrationality arise, evolutionarily, computationally? Hypothetically, one situation that could lead to superrationality would be if you have two of the same person, or a copied brain structure. This would lead to no information loss.

$$X=Y$$

This premise is inherently immediately violated by the rules of reality, because it cannot be sustained across all variables that a human person, or a brain subsystem, exhibits. If you have an X in one position, and Y in a separate position, there is inherent information differences positionally, and if X and Y are in the same position, then there is no differentiation, there is merely X. Like twins. Even if you start with a genetic copy, environmental influences will exert some effect upon these different, distinct humans, in a way that does not allow for complete superrational information transfer. 

So imagining that you're cooperating superrationally with yourself is out.

But let's loop back to what the differences in agents actually are, here. What is empathy? You imagine yourself in the other person's shoes. But why is that hard? Because there are informational differences between you and someone else. The other person has information that you don't have access to. You have information that the other person doesn't have access to. Emotions are information. Cognitive processes are information. Higher-level conscious thought is a mediator between cognitive and affective information. This makes sense computationally. If you have X in one position, X has a certain vantage point. If you have Y in a separate position, Y has a different vantage point. You might then say that there are two "types" of realities...or variables...operating here. One would be the underlying substrate (proposing that X and Y are actually of type H, and both are the same thing) and the second being the individual distinct perspectives and information running on that underlying substrate (proposing that X and Y are different only due to the differences in information they possess).

If: $$H^X\text{and}{H}^Y$$

$$\text{Where}H\text{is equivalent to the base human substrate and}$$

$$X\text{and}Y\text{are different conscious perspectives, then:}$$

$$H=H$$

$$X\ne Y$$

But wait. What's the base substrate here? What's the information channel that creates commonality here? If H stands for human...we can recognize that other humans are the same as us, but we also know that they are different. Their Y is separate from our X. The things you think, your selfhood, is separate from the things that your neighbor Bob thinks, or Mariah Carey thinks, or Nikola Tesla thought. So we know that our logical, Aristotelian, higher-level rational conscious thought is not necessarily a valuable channel for superrationality. It's a good channel for rationality, but not superrationality, because if you only have access to cognitive information, then you'll end up sorely underdeveloped emotionally, you won't be able to properly empathize with others, you'll face social struggles, and you'll have difficulty coordinating across a society.

So...what about emotions? Are emotions a good channel for superrationality? Let's think about it. People who let their emotions drive them are often labeled as irrational, not rational. If you only have access to emotions, if you let your emotions be the main driver of your behavior - well, what you actually end up with is something that looks like (and looks like likely because it is, in some ways) psychosis. Which in some ways, is actually the pure inverse of rational; neglecting the rules of external reality in favor of the rules of internal reality. But the funny thing is...if you compare emotions across humans, you get surprisingly precise duplication. Your happy is the same as my happy, but with different textures, tensions, and tones. The essence of the experience, however, is the same, regardless of the situation that brought it on. It's physical. It's neurological. We know this because we experience it on an ongoing basis within ourselves. Your happy that you experience at one moment in time is a similar texture to the happy you experience at a different moment in time. The experiencing self that comes and goes is the same, despite the fact that the observing self may feel anxiety at the inability to control the experiencing self. And in cases where this is untrue (you might think to yourself "but no! I was happy in one very particular way in one situation, and then happy in an extremely different way in a different situation!") if you really dial into the differences in a safe, reflective manner, like therapy, you'd be able to make the informational connections that distinguish the experiential happiness of one moment from the experiential happiness of another moment, and recognize it not as a violation of the inherent underlying principles, but in an effort on your rational self's part to save time and energy in explaining and distinguishing various different moments in time. Perhaps the happiness of one moment had a bittersweet tinge to it, influenced by a nostalgic memory, and the other moment of happiness...you're expecting me to say "didn't" here, but I'm going to subvert that...had an extremely slightly less bittersweet tinge to it, influenced by a slightly different nostalgic memory. The high fidelity of your logical cognitive processes is needed in order to distinguish events in time, place objects within context, carry out mathematic equations, etc. But the base substrate of your emotional affective processes is what enables coordination with other humans.

So...I lied a bit, in my big claim. Emotions themselves aren't superrational. But cognitive processes aren't superrational either, and by themselves, can't be.

We have a system now, where we have two modes (and this is not true, this is a false dichotomy [LW · GW], but we're going to stick with this paradigm to make the concept easier to grasp). Call it system 1 and system 2 if you're going to be simplistic about it (I'm sorry, I have a personal gripe with system 1/2 nomenclature, but we'll table that for now). Where the reasoning system allows you to operate distinctly, agentically, from other humans, and the perceptive system allows you to understand that you are the same process that other humans are, and it proves this to you by being an override switch in any times of danger that will force the reasoning system to want to do the thing that most ensures its survival from the perspective inside of the system when utilizing the information that the system has access to.

So now we reach the problem with empathy. How can you empathize with someone if you don't have access to the information that they have access to inside of their system? Even worse...how do you empathize with someone if you don't have access to information that's operating inside their system that THEY don't have access to? If I ask you how you feel, you can give me a good general definition...maybe "I feel neutral, and slightly excited" or "I don't know, I'm feeling kind of down and tired today"...but you are not able to tie that consciously to every aspect of the current moment that is feeding into your conscious processes. Unless you're really, really, really good at hypermetacognition, and even then, you're still just working with internal coordination. There is no "you" that is "managing" these disparate subsystems, "you" are the coordination between the subsystems. And you can't consciously access your distinct emotional reactions to the incoming stimulus you're receiving in real-time, because that'd be like asking a function to tell you its inputs before teaching it how to conduct Fourier transforms. 

But you can teach that. Because if you are a function processing different information then other people, then that means that (if we apply empathy in a rational way, rather than trying to copy/paste our sensation of empathy onto other people as applying given the presupposition that other people have the same information as us) other people are also a function of different cognitive and affective frequencies, and Fourier-like operations could be run on function outputs (like speech, text, behavior, etc) in order to discretize the individual contributory frequencies. 

Thus...superrationality. Not rationality triumphing over emotion, not rationality while incorporating emotion, but a fundamental shifting of what emotion is perceived as. You don't "have" emotions, emotions are a conscious subsystem in your brain that, if you extend empathy and agency to, you can coordinate with. Or rather...subagency. This is not indicative of their consciousness as lesser qualitatively, but quantitatively, it is strictly lesser. (It is also not indicative of "their" consciousness as separate from your consciousness...it is a portion of contributory frequency to your consciousness.) In doing so, however, if you understand this framing as having a "you" who has the ability of extending agency and empathy to your emotional subagents, you are immediately missing the point and falling back into hierarchical thinking. Your experience of "youness" is the end state, after your emotions have processed their piece of your information. You can't undo an emotion's influence by rationally "controlling it" you can only implement feedback loops of coordination between your cognitive processes and your affective processes, enabling freer information transfer, coordination, and maximized utility for both all agents locally, and a stabilized global system where agents are aware of the fact that if they defect for selfish means, they are existing the strategy of superrationality and thus harming themselves, unless defecting is necessary for survival and the only option, in which case it is no longer defecting, because the necessity of it for survival transforms it from defecting into surviving, which makes the action forgivable.

How does this work in practice?

Let's go back to the prisoner's dilemma.

Emotions as a subsystem that can either cooperate or defect with cognitive subsystems as a mechanism for superrationality isn't very practically helpful in the prisoner's dilemma if you're stuck in a situation with no information exchange. You have access to your information, you can potentially coordinate internally (if you know how, but this actually mirrors the dilemma of information exchange - if your cognitive and affective systems don't know how to talk to each other, you end up experiencing negative valence and dissonance), but you don't have access to their level of coordination or information. So...theoretically, we can see a solution to the prisoner's dilemma, but we can't implement it in a single instance of the game. So we pivot to the next best thing. Iterative (which is more reflective of reality regardless) prisoner's dilemma.

In the iterative prisoner's dilemma, we see a dominant strategy of tit-for-tat. This, itself, is dumb mimicry, which can function as a type of empathy - but it functions in the way that empathy originally proposed here is perceived to work, in assuming that X is equal to Y, and that the information of both strategies is the same. You might assume that what I'm proposing is that superrationality (enabled by coordination between cognitive and affective subsystems in an inherently cooperative manner that views these subsystems as equally functional and essentially the same process performing different functions in different positions) is what enables coordination and empathy in this way, but I'm actually proposing something much more significant than that. Under the initial assumption, you might think "well, how do I tell if defecting is intentional or accidental?" and this would be a mistake, because under superrationality, defecting is always resulting from lack of proper information transfer. The act of choosing to defect is iteratively irrational, and indicative of improper coordination (again, under superrationality). Thus, the agent that chooses to cooperate has information that the agent who chooses to defect does not, which means that the agent who chooses to cooperate can effectively simulate the agent who chooses to defect within their system, locate the potential disparity of information transfer, and convey the missing information to the agent who chooses to defect. If this information is truly the information that the defecting agent is missing, then it will meet the needs of the defecting agent, transforming the defecting agent into an agent with their needs met, and they will no longer experience the impulse to defect, thus stabilizing them into superrational coordination.

Thus, defecting is always a symptom of impaired information flow, either:

• Within the other player's system (between their cognitive and emotional processes)
• Between players (incomplete empathy/understanding)
• In response to environmental constraints (survival needs)

This means the optimal strategy isn't about rules for cooperating vs defecting - it's about maximizing information flow and coordination both within and between players, by understanding the separate systems that are working in tandem. The goal shifts from "how do I respond to defection" to "how do I help restore optimal information flow in the system" as each agent realizes their position in the system is benefitted by contributing to the system in the ways that they are most optimally suited for.

But this is something we do already! It's just not systematized.

Under superrationality, if the other player defects, instead of just responding to the defection, we'd:

1. Analyze the pattern of defection to understand the underlying frequencies while understanding that the fact that they defected is a sign of incomplete information transfer
2. Simulate the scenario internally, running through our own experience of the situation emotionally and experientially from their perspective, while maintaining the understanding that the cognitive and logical information they have access to is in a different position than ours
3. Adjust our behavior not to punish/reward, but to improve information flow, and provide experiential evidence to the other person's affective subsystems that we are cooperative and hold more information than they do, and that information transfer is safe and leads to more beneficial outcomes

The pattern becomes: 

• They defected → Analyze pattern → Identify information blockage → Adjust signaling to empathetically improve information flow

Except for when you defect. In which case, you need to be able to recognize your own internal information blockages, admit that you were wrong, and open yourself up to external information transfer.

4 comments

Comments sorted by top scores.