Multiple LW and EA voices (1, 2, 3) have written or spoken about the idea of becoming world-class generalists. More specifically, about combining "six specializations," or "the effort equivalent of two to three PhDs, plus enough work in the relevant fields in order to master them," or building a "cluster of aptitudes."
The number of specializations and amount of effort in each required to attain world-class status would depend on where we draw the boundaries around skills, and on other assumptions. Often, commenters object that you can't expect arbitrary skills to mix-and-match. However, we don't need to assume arbitrariness just because these posts are framed that way. Skills can be chosen for synergy; the core argument is just that there's a bias toward depth rather than breadth that causes us to leave value on the table.
Understanding specialization is important, because it's the engine of economic growth. A large market permits increasing specialization, which allows problems to be solved, or solved more efficiently. Enough farmers can grow food to support a class of specialist farm tool manufacturers, enhancing agricultural output further. This allows engineers to improve the materials on which farm tools are based, reinforcing the cycle.
When we think of specialization, we sometimes think about narrowness and depth of skill. The stereotype of a programmer who knows a huge number of libraries and every twist and turn of a single language's syntax, but who neither knows any other languages, nor has any job-relevant non-coding skills. This person always prefers to learn one more library in the language they already know rather than to learn a new language, and they exclusively choose to work on software that allows them to rely entirely on what they know. They may be unusually efficient at solving these problems, but they may find that the open problems become increasingly trivial with time.
Alternatively, we can think about narrowness and depth of problem. This is a slightly less common frame, and my goal here is to promote it. To increase our specialization in a problem means that we select a certain broad challenge about which we are not very familiar, and break it down into its component sub-problems. It also means understanding prior work on these sub-problems, and acquiring a broad range of skills in order to continue work on them, or try new approaches.
Innovation is a back-and-forth between discovery of new problems and acquisition of new skills. Learning new skills allows you to discover problems you didn't know about before, and improves your ability to distinguish good ideas from bad as a side effect. Discovery of problems gives you a basis for choosing new skills to acquire in order to solve it. Over time, you're likely to acquire a suite of skills that's uncommon, or even unique. This will allow you to identify unique problems, spaces with little or no competition.
This natural process of skill-building and problem discovery can be inhibited by a bias for the familiar. Succumbing to that bias produces workers who are scary-good at solving a very narrow set of problems, and useless for anything else. They probably can't innovate very well, but they'll impress other people, and themselves, with they're unquestionable competence in this narrow skillset.
They'll also probably never want for work. They create unambiguous, substantial, legible value. If they found enough work to incentivize them to become this good at their skill, it's unlikely (though not impossible) that such work will suddenly vanish. Almost every doctor, lawyer, mechanic, musician, cook, teacher, and cab driver you'll ever meet is using this strategy. They found a career niche that provided them not only with a job, but with an endless opportunity to practice, and practice, and practice that job. These jobs are respectable ways to spend one's life. They keep the world turning. They lead to specialization. Just not novel forms of specialization.
The point of these "world-class generalist" arguments is that if you want to be an innovator, the easiest way is by mastering several plausibly-related but practically orthogonal skills. This process will allow you to discover a valuable combination problem for only you to solve.
An associated idea is that problem discovery doesn't just happen magically by learning a skill. Learning a skill gives you information about just one person's problems in that space: your own. But it also offers you a basis for understanding the problems that other people have in that new skill-space. It makes you one of them. This is when you start to discover good problems to work on.
This all suggests a general formula for boostrapping your own personal innovation capacity. Remember, the point here is that we are trying to overcome our bias for the familiar and convenient practice opportunity, the known known and the known unknown. Instead, we are motivating a choice to prioritize an exploration of unknown unknowns, by acquiring new skills and relationships.
My main point of uncertainty is in what level of competence is required to reap the benefits of learning that new skill. The timeframes recommended at the start of this post (mastery of 6 specializations, or 2-3 PhDs of effort + work experience) suggest that the recommendations in this post are relevant only at approximately 5-year intervals. When you are considering your next major career move, or a major new academic endeavor, choose something deeply unfamiliar and trust that your previous skillset will turn out to make you uniquely valuable, even though you don't know how. But in between, just focus on building those skills in your new specialty. That's still where 90% of your effort will go.
On the other hand, it could be that shorter timeframes make this rule of thumb a guide for decisions on much more frequent timescales. There could be brand-new skills that give the biggest payoff-per-unit-effort in timescales on the order of months, weeks, or days, and still present significant synergies. If true, then this idea of prioritizing unknown unknowns becomes an even more useful guide to action, a go-to heuristic even on a daily basis.
My guess is that the latter is true, and that "prioritize the unknown unknowns" heuristic is something that applies even on a daily basis. If you're in your first years of learning engineering, and have time to do things outside class, it seems potentially very useful to get a shallow but broad overview of key concepts from a wide variety of engineering areas. Rather than spending lots of time solving every problem in the textbook, read quickly, wrap your head around the key concepts, and move on until you hit a wall that requires retracing your steps and deepening in a particular area.
My main goal with this post is to identify and synthesize an argument I've seen made roughly the same way by two different prolific LW authors, along with a persistent counterargument. It's clearly a topic of interest to many, and is probably relevant to the Effective Altruism community as well, as many young people are trying to figure out how to make a difference.
This is in a bit of tension with other aspects of EA. The movement has a bit of a thing for quantifiable cost/benefit analyses of organizations performing concrete interventions. I think that this can tend to orient participants around the idea that they somehow need to go into their career knowing in advance that it's going to result in the same extremely legible benefits. For a movement that's heavily about addressing risk, it sometimes seems to promote a drive in its adherents for certainty about the ultimate fruits of their future labor.
What I think many fail to grasp is that the legible success of EA, in terms of funding directed, analysis produced, members acquired, compelling arguments generated, and so on, is the endpoint of a long and uncertain process that probably has its origins in the dorm rooms of a cohort that's about 10 years older than the newcomers. At that time, none of those people knew what they'd be able to accomplish, and they had to figure things out in a state of great uncertainty throughout that time in order to create what they've created.
This is always how it is in innovation. Right now, I'm trying to invent a little gadget to solve a surgical problem in our spinal cord injury research lab. I don't know if it'll work out, provide benefit, or create any ultimate value for me in my career, or for patients in the future. I didn't know this problem existed a few months ago, and when I started this project, I had only a fraction of the skills required to execute the solutions I'm now trying. It might turn out to be a stupid project that I'll have forgotten about in six months, or it might turn into my first original contribution to the field of SCI repair.
All I can do is learn new skills, search for and prioritize new problems, build new relationships, ask for advice, and iterate.
That's all you can do, too.
Premature optimization is the root of all evil.
Premature optimization is the root of all evil.