“If you know the enemy and know yourself, you need not fear the result of a hundred battles. If you know yourself but not the enemy, for every victory gained you will also suffer a defeat. If you know neither the enemy nor yourself, you will succumb in every battle.”
—Sun Tzu, The Art of War
extremely Bret Devereaux voice This is the first part of a planned three part series called “Further Ironies” that will discuss generalization, specialization, and how we keep getting it all wrong. We’re actually going to take that Sun Tzu quote in reverse:
Part I: If you know neither the enemy nor yourself, you will succumb in every battle. The current state of most deployments, because generally speaking (hah), we don’t know the kinds of actors we’re designing for.
Part II: If you know yourself but not the enemy, for every victory gained you will also suffer a defeat. Sometimes you win, and sometimes you lose, and you’re not sure exactly why some of your products beat the brakes off of the problem and the rest have become speed bumps because you’ve misclassified the task at hand.
Part III: If you know the enemy and know yourself, you need not fear the result of a hundred battles. When you win consistently because you know what the true problem at hand is and how to face it1.
So, let us start as the Vorlons do: Who are you?
My fellow homo sapiens sapiens, we are generalists2. And sweet heavens are we intensively expensive bitches for it physically, socially, and temporally.
Physically: we’re a mess. We need 2,000 kcal per day at a fairly continuous intake rate, and while we can source those calories from a wide variety of inputs we need a balance of the kinds of inputs. We don’t carry much fat but don’t bulk too much muscle either, so we’re neither resilient to long lean periods nor can we lift much compared to our size. We can stand, walk, and run on our back two limbs for much of the day; this is useful for getting our sensors up at height to look around but it’s terribly energy expensive. We can swim, run, climb, crouch, dig, pick, carry, and manipulate, but aren’t the best at any of these things. We’re awake most of the day like ruminants but are efficient enough at calorie acquisition that we have broad swaths of time free3 in which to hallucinate stories, potential futures, and new ways of interacting with the world that don’t exist yet.
And oh, that brain! 1.5kg of fat and goo hogging 20% of your energy production. Like using an RTX 5070 to play web games. And you can’t let that brain get too big while in utero, or else you’ll never make it out alive. So unlike most other living things on this planet, we’re useless when we’re born (and for the next couple of decades). Which is why we need to invest in each other…
Socially: the cause of, and solution to, all of humanity’s problems. We couldn’t take care of ourselves, much less anyone else, for 20+ years. Societies are how we teach each other how to survive and thrive by learning all sorts of skills. We expect that the same human being who can prepare meals can also swim in a pool, change a tire, read a book, fix the wifi, and figure out whatever the hell happened in American health insurance this year. These are all skills developed by one human being and shared to others.
Now, that’s not unique to humans; there’s orca grandmothers out there right now teaching their pod’s smallest ones a unique method that they personally developed of turning over ice floes to get the tasty snacks on top. Our differentiator from the rest of the mammals is that we developed methods of outsourcing our memories and skills at scale. No other species on this planet is collectively still angry at a guy who sold sketchy copper four thousand years ago4.
Temporally: it takes a long time to bake a fully functional human being. Basic common skill acquisition aside, our brains aren’t even fully formed until our mid-to-late 20’s, which is why schizophrenia and similar disorders tend to show up around then. An ant knows how to forage and find her way home minutes after birth. But the ant will never be able to understand calculus.
Domain expertise takes years beyond that. Bruce Lee feared not the man who has practiced ten thousand different kicks once, but the man who practiced one kick ten thousand times. However, he likely never feared a man who can only kick. Skilled, expert human beings have broad experience and skills but focus them deeply on a few5.
Both my father and I can execute a full Thanksgiving dinner from scratch; however, he typically specializes in the soup and rolls and I specialize in turkey, stuffing, and gravy6. But no matter how many times my cats witness Thanksgiving, they’ll never be able to do the job because they are murderhobo specialists rather than virtuoso kitchen generalists.
Human beings are like this everywhere. We’ve built a common set of knowledge and skills, and then adapt them and learn more about how to apply them in new settings. We can laterally apply lessons learned in one context to other contexts that seem similar. We come from knowing nothing about an environment and how things work in it, to becoming expert operators in the space of weeks or months.
All of those costs we incur as generalists across physical, social, and temporal realms pay off with interest when we enter a new environment with no experience, then can learn and adapt from others and from the things around us to become experts ourselves.
And until you step back and really look for it, it’s all invisible because that’s just what we do. It is so normal that you cannot see it.
So to help us see it, let’s use a specialist as a mirror.
Physically, a CNC machine is utterly superior to human beings in production speed, precision, repeatability, and durability7. Once dialed in with a drawing and a spec, the CNC machine will outperform a human machinist every time in those dimensions. It will not tire, it will not get bored, and its focus will not wander.
Socially, a CNC machine derives no value from working in concert and communication with other CNC machines8. It arrived complete with the knowledge of how to be a CNC machine. Firmware updates may provide small tweaks but the CNC machine stays a CNC machine. It will never be more.
Temporally, a CNC machine takes a few weeks to install and 1-2 hours to change between programs. It was born an expert in CNC machining.
Within its operating envelope, the CNC machine is far better than a human being. Outside of that envelope, it does nothing.
So, bringing it back from behind the mirror: there are no generalists that can be forced into true specialization because specialists will always outperform them at the specialized task. And there are no specialists that can be forced into true generalization because specialists have not paid the tax to acquire the breadth with which a generalist can operate.
And should you also be unable or unwilling to tell the difference between a specialist problem and a generalist problem, you shall indeed succumb in every battle.
Next time in the series, we’ll talk about what happens when you try to force a round peg generalist into a square hole specialist problem9, and vice versa.
- Probably doesn’t require a novel ability to discern psychology through art, but it wouldn’t hurt. ↩︎
- Note “generalist” versus “general purpose”. “General purpose” is a non-term that’s become so unbounded and undefined that it’s a meaningless grab bag phrase for whatever unfocused nonsense people want it to be. You know, like roadmaps. ↩︎
- God: “I have made mankind!”
Angels: “You fucked up a perfectly good monkey is what you did. Look at it. It’s got anxiety.” ↩︎ - Ea-nasir, my dude, wtf. You lovingly collected and filed your own hate mail, ya sicko. ↩︎
- The MBAs call these people “T-shaped resources”. ↩︎
- The gravy is done when I’ve burned myself on the side of the roaster. ↩︎
- Gwenn explores this at length in The Myth of Collaborative Robotics. ↩︎
- Save your pseudo-social griping about LLMs and the Hugging Face incident, we’ll get there sometime later. ↩︎
- Mostly, you make UX researchers cry. ↩︎
