Having taught thousands of students and watched hundreds of graduate students, friends, and colleagues tackle advanced material, I have come to realize that there is a common path people follow when mastering a technology. There are three distinct stages, or levels of enlightenment, that individuals proceed through.
I want to chronicle this seemingly universal process, partly to seek feedback from those of you with deeper insights than mine, partly to provide reassurance to young graduate students, partly to help people who are entering my field, but mostly to serve as a framework for how to evaluate communities and improve discourse. For any community will consist of various individuals, scattered throughout this continuum between the various stages of enlightenment. And it's useful to be able to categorize and guide appropriately.
One caveat: I am not claiming to speak from a position of authority on this topic. All of this is "meta" and outside my area of expertise. I can claim to have reached Level 3 only in a limited number of narrow fields, and I'm constantly trying to explore new fields myself. So for all I know, maybe there are 16 stages of enlightenment and I've only seen 3! But I do think it's useful to discuss how people respond to intellectual challenges, what this means for those around them, and perhaps how we can all engage in better discussions.
Almost all newcomers to an area are greeted by a barrage of ideas developed by people who came before them. Even if an area seems "brand new," say, computing in the 40s, it buds off of existing fields, like electrical engineering or math, and entails the use of techniques drawn from both. Mastering these techniques is often a difficult process, for the road ahead seems windy and the end is not clear.
For one, learning a new area requires adopting the terminology and frameworks used within that area, some of which might be confusing and even off-putting. For instance, what we call "slow start" in networking refers to exponential growth, one of the fastest possible ramp up functions possible. Every new system to master comes with idiosyncratic complexities that reflect its history. The bigger the gap is between how a clever person would design something from scratch versus how it is right now due to its history, the more offputting and inaccessible it is.
And the ties between the disparate lines of work aren't crystal clear. There are few people, for instance, who can relate how the work on failure detectors, in distributed systems, is related to consensus protocols. The relationship between a new consensus protocol and others that came before it may not be clear, even to the authors of the work themselves.
And there is seemingly a ton of material to master. This is especially true when one is learning on their own and there isn't someone to guide them through the enormous data bank of published work. And it can be confusing, if not incredibly unproductive, because there is often a giant gap between what is said publicly and what is known among the experts. For instance, there are some well-publicized results that are known by experts to be misguided or misleading, such as the CAP Theorem. A simple reading path will often guide one into the underbush, where even the trees, let alone the forests, are not visible. Self-study in such areas will lead one into ratholes.
The standard reaction most people have at this stage is to feel overwhelmed. People universally feel like they have far too much reading to do.
There are some common pitfalls at this stage.
Lack of integration is probably the biggest problem that people encounter as they try to make sense of the various papers they are reading. Instead of fitting the new material into a cohesive whole, they learn and memorize results. Instead of applying uniform standards and criteria to different ideas, they evaluate each idea separately, within its own framework. For instance, it is quite common for people who have done some but not enough reading in databases to use the acronyms ACID and CAP, without realizing that the C in these acronyms is completely different. Or, for another instance, for people who hold day jobs building distributed systems to claim that Paxos is a Byzantine Fault Tolerance algorithm (yes, there are Byzantine Paxos variants, but no, Paxos does not handle Byzantine failures).
Lack of time is an issue that plagues the part-timers. In most areas of study, it is impossible to perform the necessary synthesis in one's mind except by reading a large number of papers in a relatively short span of time (say, 4-8 papers per week for 10 weeks). The mental cache needs to contain the material to be cross-connected, and that can't happen at a rate of 1 paper per week.
Finally, lack of breadth is a common problem. You might say "hey, wait a minute, in this age of ADHD, you're telling me that it's lack of breadth, and not lack of focus, that is a problem?" Indeed, in my experience, focus isn't critical in the early stages. There is so much to learn that it doesn't matter if you attack left or right or center -- as long as you do the reading voraciously, you will master it (and if you're not doing the reading, we go into the "lack of time" category above). The problem cases I've seen are invariably people who are so focused on one topic that they try to go deep and skip the foundation-building that comes from reading a broad base of papers.
Most people never proceed past Level 0. It is possible to always remain behind the reading, just with a slightly expanding vocabulary and a constantly foggy understanding. Being in a structured degree program helps yank one out of this zone, but is not a guarantee.
As people gain more expertise in an area, the Level 0 feeling of "inferiority in front of the collective works of mankind," gives way to its opposite. They have now mastered a sub-piece of human knowledge. They speak the lingo. And most importantly, they have acquired the ability to critique it. They now possesses destructive powers.
Specifically, people at this level can read a paper and either figure out its critical weaknesses, or they have built a catalog of quick critiques they have memorized. For instance, every time someone says proof of stake, a half expert will cite the "nothing at stake" problem, to knowing nods from other Level 1s and 0s. Or someone will introduce a new database and a Level 1 will ask if a database is "CP or AP." These cringeworthy discussions are signaling mechanisms that Level 1s have adopted to distinguish themselves from Level 0s.
Most Level 1s are incredibly good at identifying problems with past work. In fact, they can't help but do this; they've been trained to become weakness-finding-machines. This skill is a critical enabler for what they need to do at the next level up. But leveling up is hard.
So they see it as their hard-earned privilege and, now, birthright to attack every idea that came before them, as well as every new idea that they encounter. Snark and negativity flows with copious abandon from half-experts.
And as we know from the Lord of the Rings and the entire "power-corrupts" canon, it's hard not to wield destructive powers. This is where we begin to see damaging behaviors. No work is good enough. Everything that has ever come before is crap. The greats on whose shoulders they rise clearly didn't know anything about the demands of today's systems.
There are three behavioral patterns that are common among half-experts.
First is sheer laziness. When asked to critique an idea, Level 1s will simply google related words, parse what they see (for they can now do this properly, as they are not Level 0s), and deliver a non-original critique that, in essence, was provided to them by someone else. Some might hang out on IRC channels to pick up others' ideas and regurgitate them. Many Level 1s have their pet issues that they feel defines them. For instance, some might harbor a special love or special hatred for certain techniques (e.g. certain algorithms, approaches and so forth), ready to go into a canned rant every time these are mentioned.
Second, they exhibit an obsession with their own ideas. At this stage, ideas are rare and hard to come by for the Level 1. Typically, their solutions are complex, under-thought and under-justified. It is quite common to see solutions that are essentially kitchen sinks that "solve" multiple problems at the same time. Achieving novelty with a simple and elegant solution is hard: you really have to invent something new. Creating something novel by gluing together disparate components is almost trivial. For instance, in a world where the only rain protection consists of hats, inventing an umbrella would be the next best invention, but requires intellectual prowess that a Level 1 simply lacks. It's much easier to invent a novel hat, incorporating a propeller, a wide brim, and a feather on top. Because no one has done that before, it's automatically novel. Crucially, its weaknesses cannot be readily discovered by other lazy Level 1s by simple googling. Too wet? It has a wide brim. Too hot? Propeller fixes that. What's the weird feather for and why didn't you design something simpler? Ah, but you see, the wide brimmed hats for sale never anticipated the needs of my particular hat usage, and you're a dummy if you don't see why the feather is absolutely essential.
Finally, in my experience, Level 1s have few ideas, but what few ideas they have, they repeat over and over again. People say that when you have a hammer, everything looks like a nail. I don't know if that's universally true, but what I have seen is that, to a blacksmith's apprentice who worked so hard to forge that first hammer, everyone else's hammers look totally inferior.
Level 1 half-experts often develop narratives to justify their destructive behavior. One narrative is that whatever they are working on is too important, too exceptional, too precious. This is wrong, of course, unless they are working on nuclear launch systems, but then again, Level 1's are never allowed to work on nuclear launch systems. I've met people who worked on (non-nuclear) launch systems, as well as many people who worked on systems that actually serve about a billion people. None of these experts had 1/100th the hubris of a half-expert. Another narrative, commonly used by Level 1s in industry, is to justify destructive behavior by citing their business concerns. Somehow, especially in the US, bad businesspeople firmly believe that "all is fair in business" or "it's just business," as if those words could ever provide an adequate basis for destructive behavior.
Of course, any given community will have some Level 1 members in transition. In unhealthy or immature communities, where there are large groups of people at Level 0, these people might even be revered. After all, they have a better understanding of issues than Level 0 muggles who don't even have the right vocabulary. And they might seem to have an insightful comment or two every now and then. And if they can't produce an insightful comment themselves, they can always acquire and resell someone else's ideas.
Quite a few people get stuck at Level 1 and never proceed beyond. Interestingly, such people are quite rare in PhD programs: the number of people who dropped out of our PhD program because they were stuck in Level 1 is incredibly small. What is much more common is the self-taught techie who managed to reach Level 1 on his or her own, but is unable to make further progress. They often form or join a tribe of like-minded people, for much needed validation and for protection against criticism from people who know better. And they frequently make much of the noise. People at higher levels are actually busy with productive activities, while the Level 1 typically spends more time jockeying for social recognition instead of technical competence.
There is no characterizable Level 2. Between Level 1 and the ultimate Level 3 lies some chaotic times, different for every individual, marked by much hard work.
At some point in the process of trying to master, and then advance, a new area, something magical happens. The researcher realizes that, indeed, all past work has weaknesses, and that this is normal, for it is quite difficult to achieve perfection. Their destructiveness gives way to a positive attitude that can extract valuable contributions from other people's work. They realize that we are all in this together. They become able to function in a cohesive environment of like-minded peers. And their productivity skyrockets.
Those of you who are familiar with successful software houses (like Google on the large company end of the spectrum, and Chain on the startup side, and everything in between) will immediately recognize that these are the qualities that most happy science and tech communities encourage. Hacker News, for all its many faults, works very hard to avoid snark in an effort to keep out Level 1 discourse, and it succeeds somewhat. Academia, of course, is all about principled behavior, from design to critique.
I'm not going to say anything about the value of reaching Level 3, or whether it makes a community more healthy. The causality could run the other way -- instead of happy communities being a result of people who have reached Level 3, it could be that happy communities allow people reach the final stage of enlightenment.
And are there communities that are devoid of Level 3s, yet able to function? Undoubtedly so.
But in my experience, the most stable and productive environments tend to have a large number of people who have reached this final stage. Such people are able to provide and receive criticism without making it personal. They can see the value in half-baked ideas and foster them, instead of shooting them down on sight. New people fearlessly join such groups because they know that their ideas will get fair consideration.
I don't want to paint too rosy or idealistic a picture. We know that even academia, which tries very hard to foster a Level 3 environment, has personal rivalries and petty infighting. But the discourse, at least in the communities I have been a part of, is almost always civil, the problem cases are confined, and there exist common values to help steer the group out of ruts of negativity.
Overall, the bottom line is that this is the exact progression I have seen hundreds of times in graduate programs at several different schools. Those who reach Level 3 seem to be far more productive. Of course, it's certainly possible to be productive without fully going through the stages -- my own institution has awarded PhD degrees to people who never progressed beyond Level 1. But somehow, there seems to be a strong correlation between the Level 3 mindset and the ability to advance human knowledge.
Perhaps it's worth talking about the nebulous Level 2, the all too critical transition between Level 1s and Level 3s. I'd like to hear from the readers on what they think enables people to cross over the chasm. Is it a hard challenge? Mentorship? Recognition? Something else? If so, how do we help more researchers achieve higher levels of enlightenment?