Personality Basins

This article approaches human psychology from the perspective that, we are all neural networks and our output (actions) are all a learned function of our inputs (experiences).

This is a common (and convenient) perspective, especially among engineers, but doesn't reflect reality particularly well. We know large swathes of a person's personality is directly linked to their genetics.

The article extrapolates this neural network perspective onto other topics like, mental disorders and depression. The solution is made clear then - just learn how to not be mentally ill! Again, convenient. But not really reflective of reality.

That’s not how I read it, I think you’re missing some nuance here.

The article doesn’t imply genetics have no effect, it just treats them as a baseline which are then adjusted over time according to the person’s lived experiences.

Likewise with mental disorders and depression, the “solution” you claim it states as “not being mentally ill” is the outcome of a process, not the process itself.

But what if it's possible to alter your influential genes, through some powerful mechanism? Whether it be through insane willpower or anything else. In that case, you have an analogy like something like artificial general intelligence or recursive self-improvement. We get to approach the discussion of questioning natural values and instinctive goals with this line of inquiry. We get to eventually question the metaphysics of God, morality, and aesthetics, by introducing fantastic elements like radical self-modification.

> But what if it's possible to alter your influential genes, through some powerful mechanism? Whether it be through insane willpower or anything else.

Sounds like epigenetics, where the environment actively influences the genes themselves: https://en.m.wikipedia.org/wiki/Epigenetics

> just learn how to not be mentally ill!

you can also learn to cope with mental illness with more or less self-destructive responses. not everybody gets a chance to learn healthier coping mechanisms.

The process itself, as far as I can tell from the article, seems to be "increase your learning rate", "change your environment", "meet new people", "take psychedelics".

My point is not that doing these things is never beneficial (well, one may argue about the psychedelics lol), just that it oversimplifies the problem space (and solution space) to the point of not being useful advice.

> My point is not that doing these things is never beneficial (well, one may argue about the psychedelics lol)

In our experience, psychedelics are very hit-or-miss depending on the person. Some (like us) take high doses regularly without much consequence, others can suffer terrible damage after just a single dose. It can be difficult or impossible to predict, anyone who's unwilling to take the risk probably shouldn't.

I Am a Strange Loop tangentially covers this.

>The solution is made clear then - just learn how to not be mentally ill! Again, convenient. But not really reflective of reality.

And you know this because?

Cutting edge theories of depression link it to alterations in the reward learning system. There is some evidence that training persons with depression to attend to certain aspects of the reward learning mechanism can reduce depressive symptomology [I am involved in this research]. But speaking more broadly, cognitive behavioral therapy, one of the most successful non-pharmaceutical treatments for depression, involves people "learning how not to be depressed" by unlearning problematic patterns of negative thinking and coping with negative events: first by recognizing what those problematic thoughts and behaviors are, and working to adjust those ... to move you out of that basin.

The main issue with this article imo is that it does not consider the meta-problem: how the reinforcement learning system can be altered by experience as well.

Not everyone can just learn such coping mechanisms. Some people have physical problems that need physical help/remediation. Of course, our current psychiatric drugs are their risky attempts to help such folks. It is all very tricky, but we should all learn how to have better attitudes and behaviors.

CBT can work, sure. It can also not work. As with any treatment.

And depression is only one mental illness, there are countless others. And there are also many different forms and causes of even depression itself.

As I mentioned in another comment, my point isn't that the article's advice is necessarily harmful, just that it oversimplifies a lot of things by assuming that all psychology can be boiled down to learning and unlearning. Ignoring the role of biology may also cause one to ignore possible paths to progress.

> This is a common (and convenient) perspective, especially among engineers, but doesn't reflect reality particularly well. We know large swathes of a person's personality is directly linked to their genetics.

I really am not an expert in any of this. Just my quick thoughts of the idea about genetics and "being born with it".

If we're attempting to create a mental model of how machine neural networks relate to human brains, would it be useful to think of genetics as the basis that determines your neural network's architecture? Maybe there's even some pre-trained weights that are communicated through genetics.

I think it would be oversimplification to say we're all born with the same neural network and pre-trained weights because like you've mentioned: large swathes of a person's personality is directly linked to their genetics.

Doesn't the environment already affect genes albeit on huge timescales like millions of years? Then the more clear question becomes how to accelerate mutations and for one individual person instead of through delicate and fallible processes like generations of species made through costly reproduction acts like civilizational projects.

Idle musing - maybe some genetics is for stored environment?

Eg, perhaps some of your genes’ purpose are to encode memories in DNA.

Kind of, but the data isn't that great on that. There's some doubt about even SSRIs being net positive long term.

TL Dr: yes, some people have low levels of "X", but we have insufficient data about why that is.

I agree, and the profit motive for the pharma corps seems to have really compromised their ethics (to put it mildly while giving them more benefit of doubt than I think they deserve).

Ultimately, medical science didn't even know the brain had a lymphatic system until this century. Futzing with the subtle biochem of neurotransmitters and hormones is quite beyond their abilities, looks to me. That doesn't appear to prevent them from making a solid off their profit, regardless of any negative results.

And, always, RIP Chris Cornell. Man, we miss that man's voice.

Bro genetics is what determines the neural network in the human brain. You ARE a neural network.

This is noted and considered out of scope: >Obviously some traits are more genetic, and thus inherent, than others, but that is not the scope of this post as even highly-heritable traits will result in a large distribution of outcomes.

Natural selection encodes adaptive responses to the environment in DNA (and other molecules), so memories can be encoded to the extent that they are adaptive and can be encoded (i.e., mechanisms may not exist to encode everything using only standing natural variation).

> it just treats them as a baseline which are then adjusted over time according to the person’s lived experiences.

So like the randomization process that seeds the values for RNN weights?

I've made the jump from physics to neuroscience, so I can talk to the engineers here (I've taken a lot of EE and worked professionally in it too).

The linkages between neurons is somewhat similar to how and RNN looks. But you must remember, there are electrical and chemical elements going on here. It's not just one neuron spiking another. There are many different biochemical processes that modify the behavior of little parts of a neuron, stoichiometrically. And there are many different types of neurons and they all change over time, sometimes drastically so. Most of the goings on is biochem. It's not digital, or even analog. You really need to go down to the field equations at times, finite elements will get you far, but only just so.

RNNs thinking certainly will help you understand better what is going on in a brain, but, like, these things are millions of years old, and optimized just to make more of themselves, not to be understood. It's tough going, and we as a species are only at the very beginning of hundreds of years of study of the brain.

If you'd like to learn more, I can recommend some texts.

And the architecture which determines what units have weights between them.

And the training rules, that determine which weights are adjusted in response to what units.

Hargeysa

Not the OP but I'd be very interested to hear text recommendations on this!

> and optimized just to make more of themselves, not to be understood

There is also pressure for it to avoid outputs such are too predictable, as part of a constant arms race against predators and same-species competitors. (For example, how many predators are instinctively key to certain prey behaviors, and when you violate their expectations they might back off.)

That goal doesn't guarantee that the mechanics themselves will be obfuscated, but it does trend in a similar direction.

Bear's Neuroscience: Exploring the Brain

Purves's Neuroscience

Kandel's Principles of Neural Science (grad level, but really the bible for neuro)

Dayan and Abbot's Theoretical Neuroscience (good for compneuro)

The Art of Electronics, 2nd edition (cheating here, but it's good to go back through the fundamentals before going into the edge cases that is neuro)

In general, AI is so new that there really isn't a good classical text between AI thingys and neuro. It will take time to suss one out and write one.