Why Generalists Triumph in a Specialized World

By David Epstein



Originally published

May 28, 2018




Mar 26, 2020


Apr 2, 2020

PurchaseExternal link
Starting with sports
We start with a story that holds as a core theme throughout the book:
Tiger Woods was effectively raised with a putter in his hands. He was always working on his golf fame. "At eight, [Tiger] beat his father for the first time. The father didn't mind, because he was convinced that his boy was singularly talented and that he [the father] was uniquely equipped to help him." (2)
"He was engaging in 'deliberate practice,' the only kind that counts in the now-ubiquitous ten-thousand-hours rule to expertise." (5) "Reams of work on expertise development shows that elite athletes spend more time in highly technical, deliberate practice each week than those who plateau at lower levels" (5)
This is the classic story. Tiger has the Mamba Mentality.
Takeaway: "The response, in every field, to a ballooning library of human knowledge and an interconnected world has been to exalt increasingly narrow focus." (6)
Roger Federer, on the other hand, hardly played tennis at all early in his life because he was so busy playing other sports. His parents were more "pully" than "pushy". Obviously, he eventually settled on tennis, and obviously this "late start" didn't hamper his ability to succeed professionally.
Powerful similar example: Great Britain significantly improved its performance in the Summer Olympics by creating pipelines for athletes to try new sports they've never played via sampling period. Simple.
"They play a variety of sports, usually in an unstructured or lightly structured environment; they gain a range of physical proficiencies from which they can draw; they learn about their own abilities and proclivities; and only later do they focus in and ramp up technical practice in one area." (7)
This is in deep contrast to highly specialized coding bootcamps and technical degree options universities are moving towards. Sampling periods seem to be getting shorter and shorter...
At the same time, on lower education: "the school system was frustratingly one-size-fits-all, made for producing 'the gray average mass.'" (16)
Takeaway: this story is far more common; "later specializers make up for the head start by finding work that better fit their skills and personalities" (9)
People often encode previous experience and career pivots as "liabilities" rather than "unique advantages" (10)
A polish man, Laszlo takes the Tiger path for his girls, aiming to morph them into chess masters via intensive focus. He takes them out of the school system. He architects databases of games. It works. For him, "chess was just an arbitrary medium for his universal point." (18) However, the point is not universal and chess is not arbitrary.
"It relies on one very important, and very unspoken assumption: that chess and golf are representative examples of the activities that matter to you." (18)
"In those domains, which involved human behavior and where patterns did not clearly repeat, repetition did not cause learning. Chess, golf, and firefighting are exceptions not the rule." (20) He drills into this a little further: "in wicked domains, the rules of the game are often unclear or incomplete, there may or may not be repetitive patterns and they may or may not be obvious, and feedback is often delayed, inaccurate or both." (21)
Interestingly, this echoes tasks which Reinforcement Learning agents struggle to learn and properly compete with humans. Even in some game environments the human performance threshold has proved insurmountable to date.
Humans who are hardwired, effectively, to succeed in tasks like golf and chess become incredibly capable pattern recognizers
Chess computers can beat humans when they are better at recognizing familiar patterns. Humans relaying strategy on top of these computers ("centaurs") proved very effective in the earlier days of chess computers. (23)
AlphaGo succeeded at Go by first training on moves by chess experts then fine tuning strategy via self-play
AlphaZero came to learn to play chess (and several other games) with incredibly few hardcoded biases via self-play
If you show a very experienced chess player the state of a chess board in game for ~ a second, they can recreate it from memory significantly better than less experienced players. This does not hold for a random arrangement of pieces on a board.
Chunking: "grouped pieces into a smaller number of meaningful chunks based on familiar patterns" (25)
"Chunking helps explain instances of apparently miraculous, domain-specific memory" (25)
"Your restaurant server doesn't just happen to have a miraculous memory; like musicians and quarterbacks, they've learned to group recurring information into chunks." (26)
The same goes for developers who are very used to operating within their own programming environment (extreme example being vim)
We effectively build our own auto encoders
Building "islands of genius" (27)
Tiger's parents aimed to skip the "sampling period" entirely (65)
Transitioning to music
In research on children learning to play music, "The less skilled students tended to spend their time on the first instrument they picked up, as if they could not give up a perceived head start." (67)
"Learning to play classical music is a narrative linchpin for the cult of the head start; as music goes, it is a relatively golflike endeavor...And yet even classical music defies a simple Tiger story." (67)
Many successful classical musicians grew up in contexts where they didn't have the privilege to devoutly learn one instrument from experts but rather had to be scrappy: learning anything musical when they had the chance.
Another example was a successful pianist who was born cross-eyed making it hard for him to learn musical notation → learned via imitation (71). This brings a more creating and experimental approach to learning.
Musician and surgeon Charles Limb "saw brain areas associated with focused attention, inhibition, and self-censoring turned down when the musicians were creating. 'It's almost as if the brain turned off its own ability to criticize itself." (75)
"The jazz musician is a creative artist, the classical musician is a re-creative artist." "...breadth of training predicts breadth of transfer." (76) "The more contexts in which something is learned, the more the learner creates abstract models, and the less they rely on any particular example." (77)
Joining these insights with those of Adam Grant, "...creativity may be difficult to nurture, but it is easy to thwart...The parents with creative children made their opinions known after their kids did something they didn't like, they just did not proscribe it beforehand. Their households were low on prior restraint." (77)
On evolution
When asked abstract questions, "remote villagers reverted to practical narratives based on their direct experience" (42)
Remote villagers are fare less likely to be deceived by the Ebbinghaus illusion (deducing the two orange shapes are actually the same size); "it may reflect the fact that premodern people are not as drawn to the holistic context" (43)
"The more they had moved toward modernity, the more powerful their abstract thinking, and the less they had to rely on their concrete experience of the world as a reference point" (44)
"Computational thinking" (50) in a modern, flexible world where we use different levels of abstraction to attack different kinds of problems
"Like chess masters and firefighters, premodern villagers relied on things being the same tomorrow as they were yesterday...They were perfectly capable of learning from experience, but failed at learning without experience. And that is what a rapidly changing, wicked world demands" (53)
"Relations are really hard for other species...allows humans to reason through problems that they have never seen in unfamiliar contexts." (103) Analogies are different from instincts. The further removed from the current problem an analogy is, the wider our scope of potential solutions is.
On education
"Learning itself is best done slowly accumulate lasting knowledge, even when that means performing poorly on tests of immediate progress. That is, the most effective learning looks inefficient; it looks like falling behind." (11)
Cognitive entrenchment: "experts had a far more difficult time adapting to new rules than did nonexperts" (32)
Similar to overfitting in ML; "pretending the world is like golf and chess is comforting" (35)
"scientists inducted into the highest national academies are much more likely to have avocations outside their vocation. And those who have won the Nobel Prize are more likely still." (33)
Humans are very good at "trying to do the least amount of work that we have to in order to accomplish a task." (83); students are often in environments where they can change conceptual problems into procedural (cookie cutter) ones → can't apply learnings outside of rigid academic structures
"...for learning that is both durable (it sticks) and flexible (it can be applied broadly), fast and easy is precisely the problem." (85)
"Excessive hint-giving...bolsters immediate performance, but undermines progress in the long run...Socrates was apparently on to something when he forced his pupils to generate answers rather than bestowing them. It requires the learner to intentionally sacrifice current performance for future benefit." (85)
This is particularly true when our generated answers are wrong
"Struggling to retrieve information primes the brain for subsequent learning, even when the retrieval itself is unsuccessful. The struggle is real, and really useful." (88)
In school "like a lot of professional development efforts, each particular concept or skill gets a short period of intense focus, and then on to the net thing, never to return. That structure makes intuitive sense, but it forgoes another important desirable difficulty: 'spacing,' or distributed practice." (88)
"Short-term rehearsal gave purely short-term benefits...Repetition, it turned out, was less important than struggle." (89)
"It is difficult to accept that the best learning road is slow, and that doing poorly now is essential for better performance later." (90)
In a survey of students about their Calculus I teachers, the teachers who were better reviewed tended to give better grades, though those students tended to do worse in later iterations of Calculus classes. More "difficult and frustrating" Calculus I classes led to better long term performance. (91)
"teachers and students must avoid interpreting current performance as learning." (92)
Deep dive into this notion of desirable difficulties: things which are hard but are objectively better for us.
"students who learned in blocks—all examples of a particular type of problem at once—performed a lot worse come test time than students who studied the exact same problems but all mixed up." (95) This means the brain can't just go into procedural mode.
"Whether the task is mental or physical, interleaving improves the ability to match the right strategy to a problem." (96)
"Kind learning environments experts choose a strategy and then evaluate; experts in less repetitive environments evaluate and then choose." (96)
Similar to dropout in machine learning, "desirable difficulties like making connections and interleaving make knowledge flexible, useful for problems that never appeared in training." (96)
On organizations
"But the drive to specialize goes beyond that. It infects not just individuals, but entire systems. as each specialized group sees a smaller and smaller part of a large puzzle." (12)
"Overspecialization can lead to collective tragedy even when every individual separately takes the most reasonable course of action." (12)
"stable structures and narrow worlds" (30) → people will just operate like AI systems
"'How Not to Teach People to Discover Rules'—that is, by providing rewards for repetitive short-term success with a narrow range of solutions" (31)
"increasing specialization actually creates new opportunities for outsiders." (178)
Certain pathways, organizationally and mentally become hardwired. "Knowledge is a double-edged sword. It allows you to do some things, but also makes you blind to other things that you could do." (179)
Can result in "intellectual archipelagos" (278)
Yokoi, a major innovator for Nintendo, "had been fortunate to come to a playing card company rather an established electronic toymaker with entrenched solutions, so his ideas were not thwarted because of his technical limitations." (199)
The majority of their successes leveraged lateral thinking, combining old tech in new ways
Having total congruence across a business can be a challenge because people make decisions just to fit in with norms. Instead, we can expand "organization's range by identifying the dominant culture and then diversifying it by pushing in the opposite direction" (257) → "the power of cross-pressures in promoting flexible, ambidextrous thought" (258)
For example, there can be a difference between "chain of command" and "chain of communication" → dynamism in incentives → act in the "best" way not necessarily the "expected" way
In the 2008 financial crisis, each lending market was considered independently by regulators since there were no cultural cross-pressures → failed to look at systemic issues (279)
"the innovation ecosystem should intentionally preserve range and inefficiency." (284) "Scientists are increasingly required to provide evidence of immediate and tangible applications of their work. That is head start fervor come full circle; explorers have to pursue such narrowly specialized goals with such hyperefficiency that they can say what they will find before they look for it." (285)
Cures for many diseases (like HIV) have come from radically different domains of research
On thinking and perspective
"the more internal details an individual can be made to consider, the more extreme their judgement becomes." "Focusing narrowly on many fine details specific to a problem at hand feels like the exact right thing to do, even it is often exactly wrong." (110)
"Around 90 percent of major infrastructure projects worldwide go over budget (by an average of 28 percent) in part because mangers focus on the details of their project and become overly optimistic." (110)
Understanding causal and deeper structures requires often more distant analogies than understanding superficial links
End of history illusion: "we recognize that our desires and motivations sure changed a lot in the past...but believe they will not change much in the future." (156)
Biologically, we are bound to change as we age: "Adults tend to become more agreeable, more conscientious, more emotionally stable, and less neurotic with age, but less open to experience. In middle age, adults grow more consistent and cautious and less curious, open-minded, and inventive. The changes have well-known impacts, like the fact that adults generally become less likely to commit violent crimes with age, and more able to create stable relationships." (157)
Context principle: our nature influences how we respond to situations, but different situations can alter our very natures
"Be a flirt with your possible selves." "Test-and-learn...not plan-and-implement." (163)
"The disparity between the total quantity of recorded knowledge...and the limited human capacity to assimilate it, is not only enormous now but grows unremittingly...how can frontiers be pushed...if one day it will take a lifetime just to reach them in each specialized domain?" (179)
One response: dilettantes "can push forward by looking back; they can excavate old knowledge but wield it in a new way" (189)
Another: there is a difference between producers and experts. Producers link things and build on top of very deep but abstracted knowledge sources. Experts create API's for building off of their work. This ability to build on abstracted knowledge bases should improve over time.
Communications technology has made this knowledge sharing global and easier
Hedgehogs burrow into their ideas. Foxes "draw from an eclectic array of traditions" (221)
The more time experts spend on upping scientific literacy in one domain, the more likely they hold dogmatic views around it
"Beneath complexity, hedgehogs tend to see simple, deterministic rules of cause and effect framed by their area of expertise, like repeating patterns on a chessboard." However "most cause-and-effect relationships are probabilistic, not deterministic." (229)
Overlearning: firefighters struggle in cases where they have to run—"when a firefighter is told to drop is firefighting tools, he is told to forget he is a firefighter." (247) Navy seamen often forget to remove steel-toed shoes when abandoning ship. "Normally reliable organizations that clung to trusty methods, even when they are led to bewildering decisions." (247) "Behavior has become so automatic that they no long even recognize it as a situation-specific tool." (248)
On professional life
"If we treated careers more like dating, nobody would settle down so quickly." (131) Higher education could be an enormous sampling period, but more and more it is a fast-track to specialization.
This trend has an interesting alignment with Angela Duckworth's notion of grit: "work ethic and resilience...'consistency of interests'—direction, knowing exactly what one wants" (133). Placing an emphasis on direction and even resilience can actually make us end up in local minima without properly considering other options
As an example: "Some people might start memorizing root words for the National Spelling Bee and then realize it is not how they want to spend their learning time. That could be a problem of grit, or it could be a decision made in response to match quality information that could not have been gleaned without giving it a try." (135)
"We fail...when we stick with tasks we don't have the guts to quit" (136)
"I worked in labs during and after college and realized that I was not the type of person who wanted to spend my entire life learning one or two things new to the world, but rather the type who wanted constantly to learn things new to me and share them." (142)
"finding a goal with a high match quality in the first place is the greater challenge, and persistence for the sake of persistence can get in the way" (143)
"humans are bedeviled by the 'sunk cost fallacy'" (143)
Taking a short-term greedy path jumping from task to task and focus to focus can set us up for long term fit; "you have to carry a big basket to bring something home" (153)
For Phil Knight of Nike, "his main goal for his nascent shoe company was the fail fast enough that he could apply what he was learning to his next venture. He made one short-term pivot after another, applying the lessons as he went." (155)
Michelangelo "constantly changed his mind and altered his sculptural plans as he worked. He left three-fifths of his sculptures unfinished, each time moving on to something more promising." (164)
In a case of comic book creators, "where length of experience did not differentiate creators, breadth of experience did. Broad genre experience made creators better on average and more likely to innovate." (209)

← Back to all books