Note: This is a draft, under active editing. The end-goal is to get across the mindset of, “holy shit, you can first-principles figure out your brain if you’re good at science. You can gradually build better and better models. And eventually, you can derive how to develop superpowers from those models, and then develop them.” The Jhanas (meditative states of bliss surpassing all sober experience) and memory techniques are examples of narrow superpowers. But the brain can learn anything. If you can practice it, you can get better at it. And general skills help in many instances. This is basically the TL;DR. The rest of the post is currently fairly bad. Written very quickly. I’m going to totally rewrite it soon. There’s difficulty in communicating exactly what I mean, since many people don’t know how to do proper science here (or often, anywhere). The “standard scientific kit” that I have is, in all likelihood, something only 0.01% of the population has in comparable quality. Navigating a domain where every abstraction is preliminary, the act of observation changes what you observe (in the mind; you can’t “step outside” to “cleanly watch”), and you’ve got very partial observability of an extremely complex system… Point is, it’s hard. Like, actually hard. You can make progress on it, but it’s totally possible it would take me + others 5-50 years full-time to develop a (meaningful and real) theory of everything for the mind, that fully describes what extreme capabilities are possible with what learning curves, with what ideal form/technique looks like. (Or: a process to ratchet towards better form/technique.)
So, it’s very plausibly an impractical hobby given the rate AI seems to be advancing. That said, partial progress into understanding my mind from the inside seems to yield improved intuitions around AI & AGI. Though I’ll have to wait to say for certain. There’s a big difference between “reading about active inference” (as an example, though it’s a very high-level, incomplete theory) and “internalizing active inference as your intuitive model for yourself”. The latter takes time, but analogous to doing math without anything becoming intuitive, the former path kinda sucks.
(It’d be funny if I ended up helping build artificial superintelligence after meditating a lot. lol.)
But yes, towards an engineering grade understanding of the brain! Things like internal family systems therapy are (in my opinion) more akin to alchemy. They work, and are very valuable. But technology looks weirder. In my opinion, the depths of Jhana that monks have developed over thousands of years are more akin to technology. Memory techniques similarly. I’m very curious in developing more technology. And theories that allow you to derive a great deal of similar technology from them. (I’m also very interested in learning to use existing technology, of course.)
My current personal practice is a combination of (1) learning to use existing technology (and alchemy) and (2) attempting to derive what’s actually going on. Models of my brain that fit all the data (i.e. everything) while still being juicy enough to extract insights. (3) practicing seeing my experience through the lens of various models (these models are intuitive, not just conceptual. very important!) (4) occasionally experimenting with small innovations.
Long-term I’m hoping to develop some interesting and useful personal technology. And derive the stuff that some people get lucky and do intuitively. I expect there are a lot of “latent factors” not described by top performers in various fields, analogous to Feynman and the hairy ball story, that are actually critical bits of mental tech. I’m very interested in pursuing and understanding exceptional people for this reason also. There’s so much that’s not articulated and needs to be absorbed. (That said, there’s no reason why the implicit cannot be made into an explicit training program.)
This all circles back to my interest in deliberate practice. I would not be where I am if I had not gotten interested in the science of learning at 15. Now, there’s a sense in which many valuable things can’t be easily deliberately practiced. There’s a spectrum of difficulty, and a spectrum of “how close to the gold standard of practice you’re getting”.
Now, when people can deliberately practice something for many hours, they get extremely, insanely good. To where their performance looks like magic. The magic of mastery. Now, I’m very interested in how I can better approximate deliberate practice in domains outside, e.g. piano or tennis, where it’s typically much harder to do. Understanding the core of what makes deliberate practice so insanely effective is important. I think it’s something about “every rep improving performance” and “rep” being fairly atomic, on the order of 1s-1min. And then doing a shitload of reps. Something like that? Not sure.
Becoming a magician through deliberate practice in domains where most people can’t or don’t, seems really cool. Bonus if it’s in domains most people don’t improve in much at all. One interesting example of this is empathetic attunement. Through deeply understanding yourself and others, I think you can develop a level of empathy that borders on mind-reading. People far further than me on the path of understanding themselves report this. (See: Joe Hudson discussing it on charisma on command) and I totally buy it’s possible.
Prior, slop post below. May delete later:
Since the degree to which I understand and have practical facility with how my mind works determines basically everything about my life, understanding myself better is a very high priority. Classical self-improvement versions of this (e.g. learning the basics of how dopamine works) are great, but tend to asymptote fairly early, at least the way they’re classically implemented (not well).
One way I like to think about implementation in self-improvement is in terms of principles and particulars. There are often general principles like “how dopamine works” and then many particular applications of this principle. Actually applying the principle requires learning a little for each particular, or at minimum recognizing the principle applies. The abstract to concrete mapping does not happen for free.
Anyway. Classical self-improvement is… okay. Fairly good even, when practiced as a skill (rather than as passive information consumption). But I’m not just interested in getting a little better. I want to go superhuman. And that’s going to require scaling out further than existing methods do, and in some new ways. Before discussing what novel R&D looks like here, I’ll discuss a few narrow superpowers we know how to develop, some commonalities, and speculation about general factors that can help with “everything”.
- Jhanas: States of extreme bliss that can be entered on demand with sufficient skill. Popularized by Nick and Jhourney1.
- Memory techniques: Records of ~400 images in 5 minutes, decks of cards in 15s, etc. Images are particularly interesting as they’re a very general category.
- Calculation: Records involve multiplication of 8 digits numbers mentally in seconds, and adding extremely quickly
- More broadly, any domain where deliberate practice works for thousands of hours produces superpower-like expertise.
What’s particularly interesting about the examples I’ve given is that they’re general skills. If you master the Jhanas, you can have incredible quantities of energy, happiness and ‘updatability’ (of a certain kind) on tap. If you were to master memory techniques you could (in theory, if you used them) have a huge quantity of “mental whiteboard” on tap. Similarly, for calculation, it’s a general skill that could theoretically be applied to any situation involving quantities.
Now, typically masters of memory techniques don’t see large general gains that help them in other domains. There’s no reason they shouldn’t be able to get general gains though. So what’s going wrong? I think two things:
- You have to actually learn to use it where it’s useful. Transfer mostly won’t happen by itself.
- Non-working memory isn’t that useful. We have computers. Classical techniques focus on the wrong things.
Now, I haven’t trained myself to have a flexible mental whiteboard yet, but I’m fairly confident it would work. My roommate Ethan has been training for this and seeing success so far. (Ethan is also where I got most of the ideas and clear thinking about the stuff mentioned above.)
(I’m also more focused on solving executive function before attempting to increasing my effective intelligence. Marginal executive function gains pay for themselves very quickly, and I’ve always wanted superhuman executive function. I’ll add writing about this soon. I’m actively updating this page whenever I have time haha.)
Some brief stuff about how understanding the mind through experience can work. I’ll organize this later. So, how do you begin to understand things better? How do you start to build out a causal model for our experience? The kind of causal model that would say “given the subtle unpleasant heaviness I feel, doing [untranslatable-15] will lead to the unpleasantness diminishing by 50% over the next 5s with 80% probability” - as well as a thousand other things. Imagine having an intuitive causal model like this. How could you build towards it?
You might start by sitting with your eyes closed, paying attention to all the subtleties of my experience you can detect. Thoughts, emotions, movement of attention, what you’re “doing”, etc. And then thinking hard about causality and implied models of the mind. When you have a theory, test it, and while testing it continue to build out your causal model for state transitions and broader causality.
An important thing that makes this avenue of research workable is that you can train yourself to have “print statements” in the mind for debugging. If you note a class, say, “hearing” or “heaviness” every time that percept occurs in the center of attention, over time the process of these notes will become automated. Similar to typing, you become able to intend for logging and classification to occur, and it does.
Other random notes. Will weave in later. Running out of time editing rn:
Some fairly core questions w.r.t. understanding my own mind and becoming more capable:
- How does updating work? How do I actually change my mind, not at a logical level but a more intuitive one that changes behavior?
- How does reinforcement work in my brain? Credit assignment? What does it feel like? How can I predict if I’ll do something more/less?
- How does generalization work? How can I create more positive manifold and transfer for myself?
- How does “wanting” work?
- What’s valence (feeling good/bad) and how does it relate to reinforcement and when?
I’ll write more thoughts and results later. I have partial answers to some of these, but a lot more research and training needs to be done. I’m still a mostly normal human, with very inconsistent and limited superpowers. Sometimes I’ll feel bad and be able to turn that to a state of intense uniform bliss. Often I won’t be able to - more practice is needed. As for the real superpowers, they haven’t been invented yet.
I’m excited to do some research on the intersection of AI and sane psychology/psychophysics. Unfortunately (fortunately?), the field of sane psychology consists almost exclusively of my roommate, Ethan Kuntz.