Short version (courtesy of Nanashi)

Our brains' pattern recognition capabilities are far stronger than our ability to reason explicitly. Most people can recognize cats across contexts with little mental exertion. By way of contrast, explicitly constructing a formal algorithm that can consistently cats across contexts requires great scientific ability and cognitive exertion.

Very high level epistemic rationality is about retraining one's brain to be able to see patterns in the evidence in the same way that we can see patterns when we observe the world with our eyes. Reasoning plays a role, but a relatively small one. Sufficiently high quality mathematicians don't make their discoveries through reasoning. The mathematical proof is the very last step: you do it to check that your eyes weren't deceiving you, but you know ahead of time that your eyes probably weren't deceiving you.

I have a lot of evidence that this way of thinking is how the most effective people think about the world. I would like to share what I learned. I think that what I've learned is something that lots of people are capable of learning, and that learning it would greatly improve people's effectiveness. But communicating the information is very difficult.

It took me 10,000+ hours to learn how to "see" patterns in evidence in the way that I can now. Right now, I don't know how to communicate how to do it succinctly. In order to succeed, I need collaborators who are open to spend a lot of time thinking carefully about the material, to get to the point of being able to teach others. I'd welcome any suggestions for how to find collaborators.

Long version

For most of my life, I believed that epistemic rationality was largely about reasoning carefully about the world. I frequently observed people's intuitions leading them astray. I thought that what differentiated people with high epistemic rationality is Cartesian skepticism: the practice of carefully scrutinizing all of one's beliefs using deductive-style reasoning. 

When I met Holden Karnofsky, co-founder of GiveWell, I came to recognize that Holden's general epistemic rationality was much higher than my own. Over the course of years of interaction, I discovered that Holden was not using my style of reasoning. Instead, his beliefs were backed by lots of independent small pieces of evidence, which in aggregate sufficed to instill confidence, even if no individual piece of evidence was compelling by itself. I finally understood this in 2013, and it was a major epiphany for me. I wrote about it in two posts [1], [2].

After learning data science, I realized that my "many weak arguments" paradigm was also flawed: I had greatly overestimated the role that reasoning of any sort plays in arriving at true beliefs about the world. 

In hindsight, it makes sense. Our brains' pattern recognition capabilities are far stronger than our ability to reason explicitly. Most people can recognize cats across contexts with little mental exertion. By way of contrast, explicitly constructing a formal algorithm that can consistently cats across contexts requires great scientific ability and cognitive exertion. And the best algorithms that people have been constructed (within the paradigm of deep learning) are highly nontransparent: nobody's been able to interpret their behavior in intelligible terms.

Very high level epistemic rationality is about retraining one's brain to be able to see patterns in the evidence in the same way that we can see patterns when we observe the world with our eyes. Reasoning plays a role, but a relatively small one. If one has developed the capacity to see in this way, one can construct post hoc explicit arguments for why one believes something, but these arguments aren't how one arrived at the belief.  

The great mathematician Henri Poincare hinted at what I finally learned, over 100 years ago. He described his experience discovering a concrete model of hyperbolic geometry as follows:

I left Caen, where I was living, to go on a geological excursion under the auspices of the School of Mines. The incidents of the travel made me forget my mathematical work. Having reached Coutances, we entered an omnibus to go to some place or other. At the moment when I put my foot on the step, the idea came to me, without anything in my former thoughts seeming to have paved the way for it, that the transformations I had used to define the Fuchsian functions were identical with those of non-Euclidean geometry. I did not verify the idea; I should not have had time, as upon taking my seat in the omnibus, I went on with a conversation already commenced, but I felt a perfect certainty. On my return to Caen, for convenience sake, I verified the result at my leisure.

Sufficiently high quality mathematicians don't make their discoveries through reasoning. The mathematical proof is the very last step: you do it to check that your eyes weren't deceiving you, but you know ahead of time that your eyes probably weren't deceiving you. Given that this is true even in math, which is thought of as the most logically rigorous subject, it shouldn't be surprising that the same is true of epistemic rationality across the board.

Learning data science gave me a deep understanding of how to implicitly model the world in statistical terms. I've crossed over into a zone of no longer know why I hold my beliefs, in the same way that I don't know how I perceive that a cat is a cat. But I know that it works. It's radically changed my life over a span of mere months. Amongst other things, I finally identified a major blindspot that had underpinned my near total failure to achieve my goals between ages 18 and 28. 

I have a lot of evidence that this way of thinking is how the most effective people think about the world. Here I'll give two examples. Holden worked under Greg Jensen, the co-CEO of Bridgewater Associates, which is the largest hedge fund in the world. Carl Shulman is one of the most epistemically rational members of the LW and EA communities. I've had a number of very illuminating conversations with him, and in hindsight, I see that he probably thinks about the world in this way. See Luke Muehlhauser's post Just the facts, ma'am! for hints of this. If I understand correctly, Carl correctly estimated Mark Zuckerberg's future net worth as being $100+ million upon meeting him as a freshman at Harvard, before Facebook. 

I would like to share what I learned. I think that what I've learned is something that lots of people are capable of learning, and that learning it would greatly improve people's effectiveness. But communicating the information is very difficult. Abel Prize winner Mikhail Gromov wrote

We are all fascinated with structural patterns: periodicity of a musical tune, a symmetry of an ornament, self-similarity of computer images of fractals. And the structures already prepared within ourselves are the most fascinating of all. Alas, most of them are hidden from ourselves. When we can put these structures-within-structures into words, they become mathematics. They are abominably difficult to express and to make others understand. 

It took me 10,000+ hours to learn how to "see" patterns in evidence in the way that I can now. Right now, I don't know how to communicate how to do it succinctly. It's too much for me to do as an individual: as far as I know, nobody has ever been able to convey the relevant information to a sizable audience!

In order to succeed, I need collaborators who are open to spend a lot of time thinking carefully about the material, to get to the point of being able to teach others. I'd welcome any suggestions for how to find collaborators.

New Comment
85 comments, sorted by Click to highlight new comments since: Today at 11:37 AM
Some comments are truncated due to high volume. (⌘F to expand all)Change truncation settings

An interesting post. You started with the assumption that formal reasoning is the right way to go and found out that it's not necessarily so. Let me start from the opposite end: the observation that the great majority of people reason all the time by pattern-matching, this is the normal, default, bog-standard way of figuring things out.

You do not need to "retrain" people to think in patterns -- they do so naturally.

Looking at myself, I certainly do think in terms of patterns -- internal maps and structures. Typically I carry a more-or-less coherent map of the subject in my head (which certain areas being fuzzy or incomplete, that's fine) and the map is kinda-spatial. When a new piece of data comes in, I try to fit it into the existing (in my head) structure and see if it's a good fit. If it's not a good fit, it's like a pebble in a shoe -- an irritant and an obvious problem. The problem is fixed either by reinterpreting the data and its implications, or by bending and adjusting the structure so there is a proper place for the new data nugget. Sometimes both happen.

Formal reasoning is atypical for me, that's why I'm not that good at math. I find situations where you have ... (read more)

3JonahS9y
The reference class that I've implicitly had in mind in writing my post is mathematicians / LWers / EAs, who do seem to think in the way that I had been. See my post Many weak arguments and the typical mind. People outside of this reference generally use implicit statistical models that are not so great. For such people, the potential gains come from learning how to build much better implicit statistical models (as I did as a result of my exposure to data science.) I don't know whether learning more advanced statistics would work for you personally - but for me, it was what I needed. Historically, most people who have very good implicit statistical models seem to have learned by observing others who do. But it can be hard to get access to them (e.g. I would not have been able to connect with Greg Jensen, Holden's former boss, during my early 20's, as Holden did.)
0Lumifer9y
Mathematicians, yes, but that's kinda natural because people become good mathematicians precisely by the virtue of being very good at formal reasoning. But I don't know about LW/EA in general -- I doubt most of them have "mathematical minds". Really? Math geeks/LW/EA are the creme de la creme, the ultimate intellectual elite? I haven't noticed. "Normal" people certainly don't have great thinking skills. But there is a very large number of smart and highly successful people who are outside of your reference class. They greatly outnumber the math/LW/EA crowd.
7JonahS9y
Within the LW cluster I've seen a lot of focus on precision. It's not uncommon for people in the community to miss the main points that I'm trying to make in favor of focusing on a single sentence that I wrote that seems wrong. I have seldom had this experience in conversation with people outside of the LW cluster: my conversation partners outside of the LW cluster generally hold my view: that it's inevitably the case that one will say some things things that are wrong, and that it's best to focus on the main points that someone is trying to make. By "generally" I meant "most people," not "for a fixed person" – i.e. I don't necessarily disagree with you. Separately, I believe that a large fraction of transferable human capital is in fact in elite math and physics, but that's a long conversation. My impression is that good physicists do use the style of thinking that I just learned. In the case of elite mathematicians, I think it would take like 5 years of getting up to speed with real world stuff before their strength as thinkers started to come out vividly.
-2Good_Burning_Plastic9y
Of people worldwide, or of people reading this post? Considering the former leads to this failure mode.
2Lumifer9y
Both. Mathematicians are weird people, they think differently :-) I don't think most of LW is mathematicians.
0VoiceOfRa9y
As a mathematician I can testify that even most mathematicians think in maps.

Very high level epistemic rationality is about retraining one's brain to be able to see patterns in the evidence in the same way that we can see patterns when we observe the world with our eyes.

Can you explain a bit more why you think that the way people with very high level epistemic rationality process evidence is analogous to how we recognize visual patterns? Do you think these two mental processes are fundamentally using the same algorithms, or just that both are subconscious computations that we don't understand very well?

6JonahS9y
I'm going on the one learning algorithm hypothesis that's become popular in parts of the neuroscience community, together with my subjective impressions. Intuitively, it seems parsimonious to suppose that evolution hijacked the algorithms used for sensory processing for general intelligence. I don't think that the algorithms are the same, but I would guess that they're relatives.
9Wei Dai9y
Interesting. I found one paper that explains the one learning algorithm hypothesis and gives evidence for it. Quoting from it: Is there anything more up to date or comprehensive than this paper? This tangent aside, I agree that it would be really valuable to improve the way we process evidence subconsciously. I'm a bit skeptical that you've actually found such a method, but I hope that you succeed in writing it down and that it really works.
4Kaj_Sotala9y
Our Coalescing Minds paper had the one learning algorithm hypothesis as one of its assumptions; I wasn't the neuroscience expert, but my co-author was, and here's what he wrote about that premise (note that the paper was intended for a relatively popular audience, so the neuroscience detail was kept light):
2jacob_cannell9y
The paper you linked to about the one learning algorithm hypothesis is from 2012. Since that time the theory has gained significant strength from the advances in DL, and in particular the work on deep reinforcement learning. Proving that an ANN with a relatively simple initial/prior architecture and about 1 million neurons can reach human-level performance on a set of 100 games when trained end to end with RL is pretty strong (albeit indirect) evidence for the one learning hypothesis. One key remaining question is then: how does the brain actually implement approximate optimization/learning that is at least as good as back-prop? We know that back-prop is not biologically realistic. On that front, Bengio's group has made significant recent progress with a new technique/theory called target propagation 1, which originated in part as an explanation for how the brain could implement credit assignment, but it also shows promise as a potential replacement for backprop 2 - which further increases the biological plausibility. In terms of more direct evidence, the hippocampus in particular appears to have a simple explanation in terms of reinforcement learning 3. In terms of the prefrontal cortex in particular, there are working theories that explain much of the PFC as a set of modules specialized for working memory buffers that are controlled by gating units in the basal ganglia. That whole system in particular is also driven/learned through dopamine based RL.

I'd be glad to offer what help I can. Based on other posts of yours, I would definitely practice brevity. This post is over 1000 words long and easily could be condensed to 250 or less.

Per our email exchange, here is the condensed version that uses only your original writing:

"Our brains' pattern recognition capabilities are far stronger than our ability to reason explicitly. Most people can recognize cats across contexts with little mental exertion. By way of contrast, explicitly constructing a formal algorithm that can consistently cats across contexts requires great scientific ability and cognitive exertion.

Very high level epistemic rationality is about retraining one's brain to be able to see patterns in the evidence in the same way that we can see patterns when we observe the world with our eyes. Reasoning plays a role, but a relatively small one. Sufficiently high quality mathematicians don't make their discoveries through reasoning. The mathematical proof is the very last step: you do it to check that your eyes weren't deceiving you, but you know ahead of time that it's your eyes probably weren't deceiving you.

I have a lot of evidence that this way of thinking is how the most effective people think about the world. I would like to share what I learned. I think that what I've learned is something that lots of people are capable of learning, and that

... (read more)

While I agree that there's value to being able to state the summary of the viewpoint, I can't help but feel that brevity is the wrong approach to take to this subject in particular. If the point is that effective people reason by examples and seeing patterns rather than by manipulating logical objects and functions, then removing the examples and patterns to just leave logical objects and functions is betraying the point!

Somewhat more generally, yes, there is value in telling people things, but they need to be explained if you want to communicate with people that don't already understand them.

3Nanashi9y
I definitely agree that you shouldn't be so brief as to not get your point across, I think the level of brevity depends on what your goal is. In this case, he's asking for help. It isn't until 1,500 words in that the two most important questions: "What does he want?" and "Why should I help him?" are answered. (Besides, he specifically wanted help in communicating things succinctly.)
1[anonymous]9y
The post reminded me of The creative mind by Margaret Bowden; her examples, in particular Kekule seeing the benzene ring, seem relevant here. (Although the book definitely could be shorter:)
9Nanashi9y
Here is the even-further edited version, condensed to 150 words. You'll note it very quickly gets to the three main points: * What are you talking about? * Why should we listen to you? * What do you want? Let me know if I summarized any part of your thoughts incorrectly.
7JonahS9y
Thanks very much, both for the shorted version and for the notes. I added the shorted version at the top of my post.
5Nanashi9y
Not a problem at all. What you're talking about is something I believe in, so I'm glad to help.
3Richard Korzekwa 9y
I do not think the entire post was too long, but I do think reading the short version first was helpful. It's sort of like reading an abstract before diving into a journal article. If nothing else, it helps people who are uninterested save some time. I'm not convinced this is true, but regardless, what about people who neither agree nor disagree? To a large extent, explaining why your viewpoint is right is exactly the same thing as explaining in detail what your viewpoint is.

I agree that a picture of many weak arguments supporting or undermining explicit claims does not capture whet humans do---the inferences themselves are much more complex than logical deductions, such that we don't yet know any way of representing the actual objects that are being manipulated. I think this is the mainstream view, certainly in AI now.

I don't know what it means to say that our pattern recognition capabilities are stronger than our logical reasoning; they are two different kinds of cognitive tasks. It seems like saying that we are much better ... (read more)

[-][anonymous]9y70

This is interesting. I have found that when you are like 16, you often want everything to be super logical and everything that is not feels stupid. And growing up largely means accepting "common sense", which at the end of the day means relying more on pattern recognition. (This is also politically relevant - young radicalism is often about matching everything with a logical sounding ideology, while people when they grow and become more moderate simply care about what typical patterns tend to result in human flourishing more than about ideology.)... (read more)

1JonahS9y
Much of what you say resonates with me. I think that a major problem that very smart young people often have is not meeting older counterparts of themselves. The great mathematician Don Zagier was an extreme prodigy, progressing so rapidly that he earned his PhD at age 20. But despite the fact that he possessed immense innate ability, he needed to learn from a great mathematician in order to become one. He wrote There's some overlap between what you write and Nick Beckstead's post Common sense as a prior, which I recommend if you haven't read it before. I went through something similar to what you went through, but for me it has a happy ending – it's not that my ideas were wrong all along, it's that I hadn't yet learned how to integrate them with the wisdom of people who were older than me. I suspect that something similar is true of you to some degree as well.
0Gondolinian9y
For a counterexample, I am 16 and almost all my decisions/perceptions are based on implicit pattern recognition more than explicit reasoning. ETA: I think I missed your point.
2[anonymous]9y
My point is that I was like this guy you probably aren't.
[-][anonymous]9y60

If I understand correctly, Carl correctly estimated Mark Zuckerberg's future net worth as being $100+ million upon meeting him as a freshman at Harvard, before Facebook.

Well, if I understand the post correctly, even as a freshman, Mark apparently had previous experience with owning/running a business, and was deliberately trying to become a tech entrepreneur. Now, given that someone is from a privileged family, is attending school at (almost) the maximally privileged and well-connected institution (at least on the East Coast) for wannabe rich guys, ha... (read more)

You have not understood correctly regarding Carl. He claimed, in hindsight, that Zuckerberg's potential could've been distinguished in foresight, but he did not do so.

4JonahS9y
I'm puzzled, is there a way to read his comment other than as him doing it at the time?
1Eliezer Yudkowsky9y
Yes, as his post facto argument.

Thanks for the post Jonah.

In medical school, I was taught that when you're a novice doctor, you'll make diagnoses and plans using deliberative reasoning, but that experts eventually pattern-match everything.

If that's true, then pattern-matching might arise naturally with experience, or it might be something that's difficult to achieve in many domains at once.

When I read your article, the reasons that I might doubt that you deserve collaborators are:

1) that enthusiastic self-reports of special perceptual-cognitive abilities have a low prior probability 2) ... (read more)

What you are describing is my native way of thinking. My mind fits large amounts of information together into an aesthetic whole. I took me a while to figure out that other people don't think this way, and they can't easily just absorb patterns from evidence.

This mode of thinking has been described as Introverted Thinking in Ben Kovitz's obscure psychology wiki about Lenore Thomson's obscure take on Jungian psychology. Some of you are familiar with Jungian functions through MBTI, the Myers-Briggs Type Indicator. Introverted Thinking (abbreviated Ti) is the... (read more)

4Burgundy9y
Continuing a bit… It’s truly strange seeing you say something like “Very high level epistemic rationality is about retraining one's brain to be able to see patterns in the evidence in the same way that we can see patterns when we observe the world with our eyes.” I already compulsively do the thing you talking about training yourself to do! I can’t stop seeing patterns. I don’t claim that the patterns I see are always true, just that’s it’s really easy for me to see them. For me, thinking is like a gale wind carrying puzzle pieces that dance in the air and assemble themselves in front of me in gigantic structures, without any intervention by me. I do not experience this as an “ability” that I could “train”, because it doesn’t feel like there is any sort of “me” that is doing it: I am merely the passive observer. “Training” pattern recognition sounds as strange to me training vision itself: all I have to do is open my eyes, and it happens. Apparently it isn’t that way for everyone? The only ways I’ve discovered to train my pattern recognition is to feed myself more information of higher quality (because garbage-in, garbage out), and to train my attention. Once I can learn to notice something, I will start to compulsively see patterns in it. For someone who isn’t compulsively maxing out their pattern recognition already, maybe it’s trainable. Another example: my brain is often lining people in rows of 3 or 4 according to some collection of traits. There might “something” where Alice has more of it than Bob, and Bob has more of it than Carol. I see them standing next to each other, kind of like pieces on a chessboard. Basically, I think what my brain is doing is some kind of factor analysis where it is identifying unnamed dimensions behind people’s personalities and using them to make predictions. I’m pretty sure that not everyone is constantly doing this, but I could be wrong. Perhaps someone smarter than me might be able to visualize a larger number of people in

I'd welcome any suggestions for how to find collaborators.

Keep posting the material here. Post to Main. Don't worry about it not being polished enough: you'll get plenty of feedback. Ignore feedback that isn't useful to you.

Some contrary evidence about usefulness of explicit models: http://www.businessinsider.com/elon-musk-first-principles-2015-1

My take is that you need both, some things are understood better "from first principles" (engineering) others are more suitable for pattern matching (politics).

2JonahS9y
Yes, as I say in another comment, my sense had been that what works best is 50% intuition and 50% explicit reasoning, and now I think it's more like 95% vs 5%. If you're spending all of your time thinking, that still leaves roughly an hour a day for explicit reasoning, which is substantially more than usually.
0btrettel9y
I think there might be some confusion over terms here. I don't think "pattern matching" is the best way to phrase this. Musk seems to be arguing for "rule learning" (figuring out the underlying rule) as opposed to "example learning" (interpolating to the nearest example in your collection). In the book Make it Stick, the authors mention that rule learners tend to be better learners. (These terms come from the psychological literature.) I don't think this observation is incompatible with the importance of recognizing patterns. You need to "pattern match" which rule to invoke. You also need to recognize the pattern that is the rule in the first place. Recognizing which examples to use also could be pattern matching, too, so this is why I don't think the term is right. In the same book mentioned previously, the authors write about Kahneman's systems 1 and 2, and I got the impression that mastery often is moving things from system 2 (more careful reasoning) to system 1 (automatic pattern matching, which might simply be precomputed). Here's an example: Vaniver suggested to me before that (if I recall correctly) when playing chess, someone might not explicitly consider a certain number of moves; their brain just has a map that goes from the current state of the board and other information to their next move. Developing this ability requires recognizing the right patterns in the game, which could come from simply having a large library of examples to interpolate from, or whatnot. This is precisely what I thought of when I read that it took (the famous) 10,000 hours for JonahSinick to see the patterns. (To be fair, you do need both, but it seems that if you can develop good rules, you should use them. Also, developing accurate intuition is useful, whether it uses explicit rules or not.)
0ChristianKl9y
Musk is very interesting in his regard. He didn't start SpaceX and Tesla because he reasoned himself into those projects having a high chance of commercial success. He choose them because he believed in those goals. He's driven by passion towards those goals.
0Dr_Manhattan9y
Even if I agree with you on the goals (I can claim he used meta-rationality here, in the sense that someone should try to make humans interplanetary species, even if he thought his chance of success was less than 50%) a lot the thinking that made him arrive at SpaceX seemed to be "one can actually do this way cheaper than the currently accepted standards, based on cost of materials etc"
0ChristianKl9y
I don't think Jonah or I argues that you should never make calculations. Musks did make many decisions on that path and from the outside it's hard to get an overview of what drives which decision.

I believe that what you are describing is known as internalizing.

A new paper may give some support to arguments in this post:

The smart intuitor: Cognitive capacity predicts intuitive rather than deliberate thinking
Cognitive capacity is commonly assumed to predict performance in classic reasoning tasks because people higher in cognitive capacity are believed to be better at deliberately correcting biasing erroneous intuitions. However, recent findings suggest that there can also be a positive correlation between cognitive capacity and correct intuitive thinking. Here we present results from 2 studies that directly con
... (read more)
[-][anonymous]9y20

Do you really think that this is something that can be taught through writing?

Most intuitive pattern recognition comes through repeated practice, and I think that it might make more sense to create some sort of training regimen/coaching that allows others to have that practice, instead of writing a post about it.

If you did create this training, I'd be incredibly interested in taking it (probably up to about $300 or so, which is admittedly small for this type of thing).

algorithms that people have been constructed (within the paradigm of deep learning) are highly nontransparent: nobody's been able to interpret their behavior in intelligible terms.

Not quite true Jonah: http://arxiv.org/pdf/1311.2901.pdf

0JonahS9y
Even if what I said isn't literally true, it's still true that the cutting edge research in pattern recognition is in deep learning, where the algorithms that are in some sense highly nontransparent.
2jacob_cannell9y
Upon reading your comment about non-transparency in DL I thought of the exact same paper on visualizing ANN features that Dr_Manhattan posted. There was a recent post on the machine learning subreddit about using similar techniques to investigate the neural representations automatically learned in language model RNNs. There is no intrinsic connection between transparency and automatic feature learning techniques. Consider the case of a large research team where the work in creating a visual system is divided amongst dozens of researchers, who each create specific features for certain layers/modules. The resulting features are not intrinsically opaque just because the lead researcher doesn't necessarily understand the details of each feature each engineer came up with. The lead researcher simply needs to invest the time in understanding those features (if so desired). Deep learning simply automates the tedious feature engineering process. You can always investigate the features or specific circuits the machine came up with - if so desired. It is true that ML and DL optimization tools in particular are often used as black boxes where the researcher doesnt know or care about the details of the solution - but that does not imply that the algorithms themselves are intrinsically opaque.

Does this capture any of what you're talking about? This is my intuitive take away from the post so I want to check if it's not what is intended. An analogy: we know that the lens has flaws and we can learn specific moves to shift the lens a bit so that we can see the flaws more easily. For those with high levels of epistemic rationality, bumping the lens around in just the right ways is, or has become, an automatic process such that they seem to have a magic ability to always catch the flaws right away. We ask them for an algorithm to do that and they poi... (read more)

0JonahS9y
Yes :-). I wouldn't say that it perfectly encapsulates what I was trying to say, but I myself don't yet know how to give a perfect encapsulation either. Some of the comments that other commenters have made are very much on point as well.

How many bad ideas or ambiguously true ideas do mathematicians have for every good idea they produce? How many people feel "deep certainties" about hypotheses that never pan out? Even when sometimes correct, do their hunches generally do better than chance alone would suggest? I agree with the idea that pattern recognition is important, but think your claims are going too far. My opinion is that successful pattern recognition, even in the hands of the best human experts, relies heavily on explicit reasoning that takes control over the recognition... (read more)

5dxu9y
That may reflect more of a lack of sufficient practice on your part than anything else. It takes a long time to become familiar enough with a topic that your brain can start intuitively and spontaneously generating good ideas on that topic. As an example, despite having spent several years playing chess, I still have to consider every position carefully and with deliberation; although there have been cases in which the move which immediately springs to mind is correct, I've found that in general the opposite is true. However, there is evidence that top grandmasters do not view chess positions this way; their play is based a lot more on "feeling" than "thinking". (I don't have the source for it, but I definitely remember reading something about it in both GEB and Thinking, Fast and Slow.) Clearly, this means that despite having played chess for so long, I have still not yet reached the level at which intuition can play a significant role in my calculations. Based on what you've written here, I would judge it likely that you are in a similar situation with respect to calculus. (Also see this. I think that the "post-rigorous" stage described in this post matches nicely with what Jonah said above.)
3JonahS9y
Thanks :-). I was going to respond along these lines before seeing that you had spoken my mind.
3satt9y
I feel a link to an old comment of mine belongs somewhere under this top-level post, and this subthread might be the best place for it, so I'm putting it here...
127chaos9y
If you're right, in chess it requires years and years of domain specific practice to get pattern recognition skills adequately prepared so that scrupulous thought is not required when evaluating moves. That doesn't seem like an argument against the importance of scrupulous thought to me, it seems like the opposite. Scrupulous thought is very hard to avoid relying on. I think you're wrong however. I think once you reach a certain level of familiarity with a subject, the distinction between pattern recognition and scrupulous reasoning itself breaks down. I don't think chess experts only use the raw processing power of their subconscious minds when evaluating the board, I think they alternate between making bottom-up assessments and top-down judgements. The accounts given in the neurology books are reactions to the popular perception that reasoning abilities are all that matters in chess, but if they've given you the impression that reasoning isn't important in chess then I feel like they may have gone too far in emphasizing their point. Expert chess players certainly feel like they're doing something important with their conscious minds. They give narrative descriptions of their rounds regularly. I acknowledge that explicit thought is not all there is to playing chess, but I'm not prepared to say experts' accounts of their thoughts are just egoist delusions, or anything like that. I suppose one point I'm trying to make here is that biased stupid thought and genius insightful thought feel the same from the inside. And I think even geniuses have biased stupid thoughts often, even within their fields of expertise, and so the importance of rigor should not be downplayed even for them. Genius isn't a quality for avoiding bad thoughts, it's quality that makes someone capable of having a few good thoughts in addition to all their other bad ones. When genius is paired with good filters, then it produces excellence regularly. Without good filters, it's much less reliable. F
3dxu9y
To use the chess analogy once more: this seems to conflict with the fact that in chess, top grandmasters' intuitions are almost always correct (and the rare exceptions almost always involve some absurd-looking move that only gets found after the fact through post-game computer analysis). Quite often, you'll see a chess author touting the importance of "quiet judgment" instead of "brute calculation"; that suggests extremely strongly to me that most grandmasters don't calculate out every move--and for good reason: it would be exhausting! Likewise, I'm given to understand many mathematicians also have this sort of intuitive judgment; of course, it takes a long time to build up the necessary background knowledge and brain connections for such judgment, but then, Jonah never claimed otherwise. From the post itself: If we could find a way to quickly build up the type of judgment described above, it could very well change the way people do things forever, but alas, we're not quite there. That's the whole point of Jonah's request for collaboration. (In an ideal world, I'd participate, but as a 17-year-old I doubt I'd have much to contribute and a lot of my time is used up preparing for college at this stage anyway, so... yeah. Unfortunate.)
127chaos9y
I was not aware most grandmasters' first instincts ended up being correct usually, interesting. I've been changing my position somewhat thoughout this conversation, just so it's clear. At this point, I guess what I think is that a hard distinction between "reasoning" and "pattern recognition" doesn't make much sense. It seems like successful pattern recognition is to a significant extent comprised of scrupulously reasoned ideas that have been internalized. If someone hypothetically refused to use explicit reasoning while being taught to recognize certain patterns, I'd expect that person to have a more difficult time learning. Reasoning about ideas in the way that is slow and deliberative eventually makes patterns easier to recognize in the way that is fast and intuitive. If someone doesn't incorporate slow thought originated restrictions into their fast pattern matching capabilities, then they will start believing in faces that appear in the clouds, assuming that they ever learn to pattern match at all.
0Lumifer9y
That is true which is why most people are not great thinkers. However high skill might not come from explicit reasoning, but from refining the pattern matching to prune away false branches. Mastery of a skill comes not from the ability to do a lot of Bayesian updates correctly and really fast, it comes from practicing till your intuition (=pattern-recognition engine) starts to reliably lead you towards good solutions and away from bad ones.

Is this what you were referring to in "Is Scott Alexander bad at math?" when you said that being good at math is largely about "aesthetic discernment" rather than "intelligence"? Because if so that seems like an unusual notion of "intelligence", to use it to mean explicit reasoning only and exclude pattern recognition. Like it would seem very odd to say "MIT Mystery Hunt doesn't require much intelligence," even if frequently domain knowledge is more important to spotting its patterns.

Or did you mean somet... (read more)

7JonahS9y
The distinction that I'm drawing is that intelligence is about the capacity to recognize patterns whereas aesthetic discernment is about selectively being drawn toward patterns that are important. I believe that intelligence explains a large fraction of the variance in mathematicians' productivity. See my post Innate Mathematical Ability. But I think that the percent of variance that intelligence explains is less than 50%.
0Sniffnoy9y
Ah, I see. I forgot about that, thanks!

On my return to Caen, for convenience sake, I verified the result at my leisure.

Why? Because...proofs are needed to persuade other people, I suppose. You don't need proofs and arguments if you reasoning solipsistically.

Your observation that

the most effective people in the world have a very specific way of thinking. They use their brain's pattern-matching abilities to process the world, rather than using explicit reasoning

is the subject of Malcolm Gladwell's book Blink.

I don't remember that Gladwell gave any tips for actually developing one's skills for this type of thinking, but he does have a number of interesting stories and analysis about this type of thinking. It also makes the observation that this type of non-explicit reasoning can lead us astray.

I suspect that... (read more)

0JonahS9y
Thanks for your interest :-) There's certainly overlap, but I'm making a more precise claim: that one can develop powerful intuition not only in particular domains but that one can also develop powerful general predictive models to get very high epistemic rationality across the board. Yes, my realization is about relative effect sizes: I used to think that the right balance is 50% intuition and 50% explicit reasoning or something, whereas now I think that it's more like 95% intuition and 5% explicit reasoning. (I'm speaking very vaguely here.) Ah, but explicit reasoning isn't the only antidote: you can also use intuition to correct for emotional and cognitive biases :-). I know that it's highly nonobvious how one would go about doing this. Somewhat tangentially, you might be interested by my post Reason is not the only means of overcoming bias. (The post is 4.5 years old..I've been thinking about these things for a long time :P.)
0Lumifer9y
Why do you think so? Basically, what evidence do you have that you can build strong "intuitions" which will work across diverse domains? My off-the-top-of-my-head reaction is that in dissimilar domains your intuition will mislead you.
2JonahS9y
It's really hard, that's why almost nobody knows how to do it :P. Roughly speaking, the solution for me was to develop deep intuition in a lot of different domains, observe the features common to the intuitions in different domains, and abstract the common features out. Finding the common features was very difficult, as there are a huge number of confounding factors that mask over the underlying commonalities. But it makes sense in hindsight - we wouldn't be able to develop deep intuitions in so many different domains if not for there being subtle underlying commonalities - there weren't evolutionary selective pressures specifically for the ability to develop general relativity and quantum field theory - the fact that it's possible for us means that the relevant pattern recognition abilities are closely related to the ones used in social contexts, etc.
1pwno9y
Have you explicitly factored these out? If so, what are some examples?
0Lumifer9y
The question is why do you think it is even possible? So, do you feel that your intuition will work successfully in the fields of, say, post-modernist literary critique, agriculture, and human biochemistry?
2JonahS9y
Because I've seen other people do it, I've observed a strong correlation between the ability to do it and overall functionality, and I've recently discovered how to do it myself and have seen huge gains to both my epistemic and instrumental rationality. I know that I'm not providing enough information for you to find what I'm saying very compelling. Again, it took me 10,000+ hours before I myself started to get it. I might well have been skeptical before doing so. I don't know – it depends on the relative roles of skill and luck in these fields. If you're talking about those major discoveries from the past that required integrating a diverse collection of sources of information, I believe that the people who made the discoveries were using this style of thinking. For example, I believe that this was probably true of Norman Borlaug.
0g_pepper9y
Yes, I was just about to edit my post to say "an antidote" rather than "the antidote". As a practical matter, no one is going to explicitly reason through every situation. A more practical antidote is to recognize biases and learn rules of thumb for avoiding them. A classic example is the conjunction fallacy. Explicitly calculating conditional probabilities will obviously correct this fallacy, but most of us are not going to do that most of the time. However, if one is aware of the fallacy, one can develop a rule of thumb that states that less specific hypotheticals are usually more probable than more specific hypotheticals; this rule is sufficient for avoiding the conjunction fallacy most of the time. However, even here, explicit reasoning played a role in avoiding the bias; explicit reasoning was used to learn about and understand the bias, and to develop the rule of thumb. Is using this sort of rule of thumb what you mean by using intuition to correct for emotional and cognitive biases?

An exploration of the unknown through known first-principles seems to be a good balance between order and chaos.

Oh this is nice. I've also come to realise this over time, ,in different words, and my mind is extremely tickled by how your formulation puts it on an equal footing with other non-explicit-rationality avenues of thought.

I would love to help you. I am very interested in a passion project right now. And we seem to be classifying similar things as hard-won realisations, though we have very different timelines for different things; talking to you might be all-round interesting for me.

Hi Jonah, this article is very intriguing since I might be going through a similar phase as you. Please add me to any list of collaborators you're drawing up.

0JonahS9y
Great :-). Send me an email at jsinick@gmail.com.

This seems valuable—I'm interested in helping (will email).

I want to highlight that "communicating how to do it" might not make sense as a frame. Pattern-matching is closely related to chunking. Ctrl+F yields other people who've mentioned chess, so I'll just point at that and then note that we actually know exactly how to communicate the skill of chunking chessboards: you get the person to practice chess in a certain way. There are of course better and worse ways to do this, but it seems like rather than looking for an insight to communicate you want to look for a learning process and how to make it more efficient by (e.g.) tightening feedback loops.

I have a lot of evidence that this way of thinking is how the most effective people think about the world. Here I'll give two examples. Holden worked under Greg Jensen, the co-CEO of Bridgewater Associates, which is the largest hedge fund in the world.

BW also uses a lot of explicit models, https://www.youtube.com/watch?v=PHe0bXAIuk0

Holden working under Greg is also generally weak evidence about how Greg thinks.

0JonahS9y
I wasn't making an argument, I was stating my position. I have far more evidence than I can convey a single blog post.

I personally agree with your core thesis that pattern matching is central. I invested a lot of effort into Quantified Self community building and gave press interviews praising the promise of QS. I think at the time I overrated straight data over pattern matching. Today I consider pattern matching much more important. I'm happy to collaborate on developing this line of thought.

I'm weary of whether using the word 'rationality' in this context is useful. Webster defines the word as: 'the quality or state of being agreeable to reason'. Wikipedia says: 'Ratio... (read more)

0JonahS9y
No, I haven't. Thanks for pointing it out. :-)

The coolest possible output of a collaboration like this would be some kind of browser-based game you could play that would level up your rationality.

Also, what characteristics/skills does your ideal collaborator have? Maybe what you want to do is find an effective altruist whose work could benefit very strongly from the skills you describe, tutor them in the skills, and having taught 1 person, see if you can replicate the most effective bits of teaching bits and scale them to a larger audience.

This sounds like an explanation for the old adage: "Go with your gut". If your brain is a lot better at recognizing patterns than it is at drawing conclusions through a chain of reasoning, it seems advisable to trust that which your brain excels at. Something similar is brought up in The Gift of Fear, where the author cites examples where the pattern-recognition signaled danger, but people ignored them because they could not come up with a chain of reasoning to support that conclusion.

Sufficiently high quality mathematicians don't make their di

... (read more)
1JonahS9y
I read Hadamard's book 8 years ago and liked it a lot. What I missed is that I mistakenly thought that Poincare's style of thinking was reserved for supergeniuses, and that all that someone like me could do was to clumsily use explicit reasoning. I found out otherwise when I worked on my speed dating project. Something very primal in me came out, and I worked on it almost involuntarily for ~90 hours a week for 12 weeks. I finally had the experience of becoming sufficiently deeply involved so that the problems that I was trying to solve percolated into my subconscious and my intuition took over. I rediscovered a large fraction of standard machine learning algorithms (it was faster than learning from books for me personally because of my learning disability). Before this, I had no idea how capable I was. It made me realize that being a great scientist might be within the reach of a much larger fraction of the population than I had thought.

-

[This comment is no longer endorsed by its author]Reply
0JonahS9y
I don't have a concrete plan yet. I have draft posts that I've written that are insufficiently polished for publication that I can share with you. You can get in touch with me at jsinick@gmail.com.

What would be the goal of any such collaboration: LessWrong posts, a book, a podcast series? Knowing what you will produce will help you sell yourself to potential collaborators.

0JonahS9y
I don't know what the best route would be. Again, as far as I know, nobody has succeeded in doing what I want to do, so there's a prior against conventional approachs working. If I can interest Scott Alexander, I think that he might be able to do it, though if I recall correctly, he's also blogged about how the more important a topic is to his mind, the less views his posts on it get.