Naive Psychology

how a human who makes ai thinks humans think, feel and interact

Revisiting the Id, Ego, and Superego

Freud's model

The id, ego, and superego model is a psychiatric model theorized by Sigmund Freud. The human mind is factorized into an animal-like and impulsive part (id), a rational action taker (ego), and the need to satisfy external rules (superego).

Revisiting the id and ego

Here I present my own interpretation of these terms, which differ slightly from the usual definition.

Id – the animal part

The id is our animal-like part or child-like part, and to me, it's what distinguishes us from cold-blooded machines. It is the receptacle of all our emotions and feelings, and transforms those mental concepts into physical sensations, which we experience through our body (if we are connected enough, that is). When we say that we our happy, we really mean that our id is happy. Likewise, when we say we are angry, afraid, surprised, sad, or disgusted, we mean that our id is experiencing those emothings. Therefore, to be happy, we must make our id happy.

Here are some more characteristics of the id:

  • The id is directly connected to our body sensations, and is what ``feels''. It experiences physical sensations and reacts to various emotions by causing physical sensations which it feels in return.
  • We feel alive because of the physical sensations experienced by the id, both external (outside stimuli) and internal (emotional stimuli).
  • The id can transform emotions into physical sensations, which we experience in our body in the same way as we experience external stimuli.
  • The id drives all our actions, directly through spontaneous desires leading to impulsive actions, and indirectly through the action of the ego runs simulations on the best way to satisfy the id.
  • The id doesn't "think", and certainly not in logical terms. It cannot plan, structure, or organize. In fact, it has no notion of time and cannot model the future. Rather, it reacts emotionally to situations (recalled, present, or projected). This means it can remember the past indirectly by associating various emotions to certain situations (recalled, present, or projected).
  • The id changes slowly, and is a product of our genetics and past experiences. It can be influenced, but only gradually and over time.

Ego – the problem solving part

The ego is the rational problem-solver inside our mind. When we say that we need to make our id happy to be happy, we really mean that our ego needs to make our id happy. The ego finds the sequence of actions (strategy) to make the id happy, within the physical and social constraints of our reality. When the ego cannot make the id happy, we feel frustrated (the id or the ego?)

Here are some more characteristics of the ego:

  • The ego is a pure problem solver. It has no notion of feelings or ethics, as those are provided by the ego and the superego.
  • The main purpose of the ego is to make the id happy.
  • The ego's rational thinking and problem-solving abilities can be used for solving external problems, which may of may not be aligned with making the id happier (e.g. working a painful job to make money).
  • When we say someone is intelligent, we actually mean their ego is. The IQ attempts to measure the pure problem-solving abilities of the ego.
  • In school, students are expected to develop their ego's rational and logical thinking abilities. They are graded purely on the basis of their ego's performance (except for creative subjects).
  • Laws, rights, and responsibilities are about defining the scope of our actions, hence they are about regulating the egos, but not the ids (at least in a democratic country).

RL model – problem solver vs. reward setter

Basically, the id feels a bunch of emotions all the time, which are either perceived as positive (e.g. joy, excitment) or negative (e.g. frustration, anger). All those emotions aggregate (think–weighted sum) into a single scalar, which expresses our level of happiness/well-being. The ego notices the current level of happiness, and attempts to improve current and future levels of happiness, according to simulations it runs inside our brain. Those simulations might not be optimal, nevertheless they

The following model may contradict some of the previous points, but I like it for the reason that it factorizes id and ego into reward function setter (id) and problem solver (ego). An important problem in reinforcement-learning is how to define interesting rewards that will lead to learning interesting agents. Indeed, which RL researcher hasn't wondered what is the ultimated reward function that will lead to the emergence of human consciousness and intelligence?

Here is an illustration of how the id and ego fit together to generate human behavior.

Superego – the ethical part

So where is the superego in the previous model? The superego embodies our moral and ethical values: what out to be, what out to be done in each situation. The superego could be modeled as another value to optimize, some moral cost to minimize.

However, it is also possible that the superego is not a final cost to optimize, but rather a proxy objective for making the id happy in the long term. In fact, ethical values are not just abstract concepts, they are implicitly enforced by society around us. Feelings of guilt, shame, the need for validation and acceptance from our peers keep us in check through some kind of emotional blackmail.

Some characteristics of the superego:

  • The superego influences the actions of the ego, but the ego –being the sole captain– has the last word.
  • The superego embodies our values–and by extension our definitions of right and wrong. When we say we ought to do something, we mean our superego considers it to be positive.
  • Our everyday interactions with family, peers, school, society and culture slowly shape our superego and what we think to be right and wrong.
  • Value is what we want. Ideology is how everyone should behave in order to achieve that value.

Published on 2022-02-01 by Gabriel Huang