Gradatim Ferociter
One of my favorite phrases is gradatim ferociter, the motto of Blue Origin. It means "step by step, ferociously." The phrase helped me figure out why, despite really liking Nassim Taleb's idea of antifragility, I've always found it kinda depressing. The idea seems to miss a major point: growth and innovation are processes that involve necessary fragility in proportion to how fast you want to move, and how radical of a move you want to make. Babies are fragile, babies grow fast, and I think most of us would agree, babies are a good thing. Applied with ideological fervor (as most of its fans apply it), the notion of antifragility seems to retard radical leaps of innovation, encourage Luddism, and celebrate trivial increments. All while prioritizing zero-sum winning/crowing over "fragilistas" by exploiting fragility in what others are building, rather than non-zero-sum pursuit of building interesting things.
There is no good reason to be ideological about fragility. How much fragility you tolerate can and should be a function of the calculated risks you want to take on (including risk on behalf of others, which is a question of ethics, not strategy). That in turn is a function of how well you know what you know, how fast you want to grow, how willing you are to break things, and most importantly, how ambitious you are. Gradatim ferociter is one point in a space of possibilities defined by a 2x2 with fragility on the x axis, and risk (specifically knowledge risk) on the y axis. There is no notion of "antifragility" in this model, but something like it emerges out of the interaction of regular (positive-only) fragility, knowledge risk, and ambition.
The fragility versus knowledge risk 2x2
1/ While the idea of "things that gain from uncertainty" has some utility, lately, I've found myself returning to a pre-Taleb understanding of the phenomenon,
2/ Applied to knowledge, fragility is best understood as the quality with which something is known: represented by grades in traditional schooling or other meaningful controlled test results.
3/ So while both A and C are passing grades in a calculus class, the student who gets an A doesn't just understand more of calculus, but understands it with less fragility.
4/ The notion of fragility doesn't apply directly to things that are wholly or partially known unknowns. If you want to know the fragility of what you're doing ahead of time you're out of luck.
5/ Tentative, untested, or weakly tested knowledge: Hypotheses, guesses, conjectures, speculations, and so forth have unknown fragility.
6/ Knowledge risk can be simply defined as the cost of uncertainty in knowledge fragility. It's like having taken a course and acquired some knowledge, but never having been tested or graded.
7/ When you try to do something new, like build a reusable rocket, or run companies on blockchains, you need to mix and match different grades of knowledge and ignorance.
8/ You need to use known ideas at different levels of tested fragility, and untested known-unknown ideas at different levels of fragility uncertainty.
9/ Think of knowns as graded bricks corresponding to credentialed skills. In the picture above, the A (green), B (yellow), C (orange), and D (red) bricks are knowns of different levels of fragility.
10/ These bricks all fit together will, and have rigid shapes that hold, but have different levels of structural quality. This is not a bad thing. Not every brick needs to withstand every sort of shock.
11/ Think of the fragility in terms of the tale of the Three Little Pigs and the Big Bad Wolf. Green bricks are made of stone, yellow bricks of clay, orange bricks of wood, and red bricks of straw.
12/ Think of known unknowns as ambiguously shaped lumps of unknown material composition. They might be as robust as grade-A stone, or as weak as grade-D straw. You just don't know.
13/ These are the gray blobs with question marks in the picture above. They have ambiguous shapes, and their grades are unknown. Testing will reveal their true colors, shapes, and local "fit."
14/ Now we can ask and answer the question: given an initial tower of graded bricks and ungraded blobs, and a steady stream supplying new graded bricks and ungraded blobs, how should you build?
15/ The initial tower represents the state of the thing when you encounter it, like say the state of rocket technology when you decide to attempt reusable rocket landings.
16/ (Unknown-unknowns, which are out of scope for this newsletter, can be thought of as hidden sub-surface layers of the tower structure you don't know exist at all).
17/ You can choose to build more or less robustly. The more picky you are, the slower things will go. If you insist on all grade-A green bricks for example, you will be really slow.
18/ Chances are, you won't just be slow, you'll actually stagnate, since the world changes at a certain minimum pace, driving yours. Hence the label "stagnation" for the lower left in the 2x2.
19/ You'll have to wait for appropriate green bricks, swap out any grey blobs and build slowly. This way, once you've removed legacy fragility uncertainties, your chances of even small failures are minimized.
20/ This is gradatim timidis, not gradatim ferociter. It is It is step-by-step, but zero-risk, and you pay a high cost for being sure that there won't be even the smallest of avalanches in the tower you're building.
21/ At the other extreme, you have move-fast-and-break-things. You want to build something as tall as possible, as fast as possible, and you throw on bricks and blobs in any order to do so.
22/ This is of course, the House of Cards approach to architecture. Tested and untested knowledge, low-grade and high-grade knowledge, all slapped together flimsily.
23/ Two other regimes are worth noting. Junk is when you are knowledge-risk averse but have low standards: there are no question-mark blobs in your tower, but you freely use low-grade crap.
24/ Junk is common in lemon marketplaces. Because you are dealing in knowns, you know ahead of time that what you're building is shoddy and have a clear sense of how it might fail.
25/ Junk building is a calculated risk that you won't have to hang around long enough to face the consequences. You'll take the profit and run, or you'll find a new sucker to exploit.
26/ And finally we have gradatim ferociter: where you mix varied fragility knowledge blocks and question-mark blobs in strategic ways so that the system is more stable than the parts.
27/ For example, you might choose to save your grade-A green bricks for the structurally critical load-bearing lines of construction, but use C-grade orange bricks elsewhere.
28/ You might set a minimum threshold for knowns and unknowns: no more than 10% orange C bricks no bricks below C. No more than 30% gray blobs, with no two next to each other.
29/ You might interleave knowns (graded, colored bricks) and known unknowns (question-mark gray blobs) carefully to manage risk rather than avoid it.
30/ In such a structure (top left in the 2x2), you don't preclude the possibility of failure, but you bound and contain it. If a yellow brick fails, it'll cause a small avalanche.
31/ If a grey blob turns out, upon testing, to be made of weak grade-D material, you contain the fallout and allow yourself the redundancy in the system to swap it out for something better.
32/ The point here is that when you're building something big and ambitious, you have degrees of design freedom that don't exist when you are only fearfully building small, tiny things.
33/ In large-scale systems, there is a combinatorial explosion in the number of ways things can be put together, and these ways vary widely in systemic risk and fragility.
34/ The sum can even be less fragile and risky than the parts. The tower you build is essentially a risk portfolio of interacting elements with designable exposures.
35/ To make it concrete, if you have a budget of 1 yellow brick and 1 grey blob, there's really only 3 tower configurations you can build (count them).
36/ But if you have (say) 7 green A-bricks, 12 yellow B-bricks, 15 orange C-bricks, 50 red D-bricks, and 30 grey ?-blobs, the number of tower designs you can build is HUGE.
37/ Even if you make it an apples-to-apples comparison, by specifying a tower height and durability (maybe "best" is the tallest tower left standing after 1 year), the space is huge.
38/ When you are willing to accept contained setbacks, the "ferociously" in "step-by-step, ferociously" makes sense. You take 1 step at a time, but nature may take 0 to -n steps at a time.
39/ Ferociousness is a measure of how capable you are of recovering from the worst case n step setback and continuing your step-by-step building to make up for lost height and continued net growth.
40/ In this picture, we can also understand "antifragility" as an emergent concept. If there are a lot of grey ?-blobs in your tower design, then uncertainty might reveal many of them to be green A-grade.
41/ So there is a potential that you can "gain from uncertainty" as testing reveals what you're building to be surprisingly stronger than you thought.
42/ What is "gained" is not actually the property of greater robustness, but your knowledge of that robustness, which can be translated into more material "gains."
43/ How might this happen? For instance, you might design your tower with a mix of bricks and an overall shape so you don't expect avalanches that go more than 8 bricks/blobs deep.
44/ BUT you find, in practice, that avalanches never go more than 3 bricks/blobs deep. You have a robustness margin now. Your design turned out to be too safe relative to your risk appetite.
45/ You can choose to enjoy the unexpectedly high margin of safety, or you can dial up the risk, changing your building design to now only expect 5-deep avalanches.
46/ Gradatim ferociter isn't necessarily at odds with "move fast and break things." It's an additional commitment to rebuilding what breaks, and accepting non-minimum mortality risk.
47/ While gradatim ferociter is generally a slower mode than move-fast-and-break-things, especially in the early stages, this is not out of fear of breaking things,
48/ Rather it is out of low initial knowledge of how things might break. This changes as you persist with the strategy and fragility uncertainty is lowered.
49/ Bezos made this point in his description of Blue Origin strategy. While the motif for Blue Origin is the tortoise (as in the fable of tortoise and hare) it isn't necessarily a motif for slowness.
50/ Bezos explained it as "slow is smooth, smooth is fast.@ Now we can understand it in the context of our risk-and-fragility model.
51/ If you're in gradatim ferociter mode, every step reveals more information about your robustness margin in a controlled way. This is the slow-is-smooth part.
52/ But then you parlay that information into greater speed by shrinking your robustness margin to build faster. That's the smooth-is-fast part. You don't just accelerate, you accelerate robustly.
53/ Gradatim ferociter isn't about going slow. It's about knowing how fast you can go while still knowing how hard you can fall and still get back on track.
54/ In epistemological terms, gradatim ferociter is a sort of network effect in the structure of your knowledge. Slow at first, but capable of gathering extreme momentum.
55/ This sheds light on another Bezos mantra: to get big fast. The gradatim ferociter principle adds nuance to that: there is no point to getting big fast if you cannot stay big for long.
56/ So ultimately gradatim ferociter is about sustainably high speed and acceleration in innovation, and enduring scale in what gets built.
57/ We learn our fragility/risk tradeoff habits very early. Each of the 4 quadrants represents a particular approach to education at school.
58/ In the bottom left we have the straight-As student who has no real curiosities outside the curriculum (hidebound traditionalists are the same, the grading authorities are just different). They stagnate early.
59/ In the bottom right we have the coaster/slacker who just wants to graduate with a credential, mostly with Bs and Cs, a few easy As chosen for being easy, and no ungraded extracurricular interests.
60/ At the top right we have the hobbyist dilettante building houses of cards. Excited and curious about many things both in and out of class, and always up to some madcap scheme or the other.
61/ And finally at the top left we have the kind of person who is building a philosophically robust base of knowledge, ignoring artificial distinctions between formal and non-formal education.
62/ This student is also not artificially attached to the cultural meanings of grades. An A is only worth earning if it will play a structurally critical role in the tower. Otherwise a B or C might do.
63/ For much of my life, through most of my formal schooling, I was in the house of cards quadrant, getting good grades but mostly being guided by outside "grey ?-blob" madcap schemes and interests.
64/ For a critical part (the infamous IIT entrance tests which I took in 1993) I went all green for a couple of years (11th and 12th grades), then I went back to my dilettante house-of-cards ways.
65/ It was only after around 2000 or so, at age 26, that I began to appreciate the value of gradatim ferociter philosophy, though I didn't know the phrase then.
66/ I went from mostly dilettantish to somewhat more of a calculated systematic learner, while remaining agnostic to whether what I was learning had formal grade-value or not.
67/ Not because I wanted to build something ambitious, or "be somebody" but because the things I was interested in simply could not be reached via house-of-cards, junk, or all-green strategies.
68/ Perhaps the best experience of gradatim ferociter in my life has been building my blog, ribbonfarm, which just celebrated its 10th anniversary. It's the most satisfying thing I've ever done.
69/ That, perhaps, is the best reason to adopt gradatim ferociter, rather than antifragility as your operating philosophy. You get to do deeply satisfying things instead of merely staying alive longer.
70/ The all-green-bricks stagnation strategies only keeps you safe, and allows you to gloat when you survive catastrophes that others succumb to. That's an impoverished life to my mind.
71/ The junk building strategy is a way to exploit the weaknesses of others, but ultimately misses out on the richness of exploring unknowns and working with ungraded, uncertain-fragility knowledge.
72/ The house-of-cards approach is the definition of mind-candy. You are on a track of a continuous stream of small sugar highs from precariously reaching dizzying heights of insight knowledge for brief periods.
73/ But gradatim ferociter doesn't just get you to the frontier. It allows you to stay there as long as you like, and chasing the horizon for as long as you want, without things collapsing under your feet.
74/ We can add details to my model here. You could model "use it or lose it" as green bricks going yellow or orange from disuse unless they're in a working tower.
75/ You could model acquisition of knowledge as grey ?-blobs turning into colored bricks. Fragility uncertainty being squeezed out, and modifying next building steps.
76/ Six grey blobs turn into red D-grade materials in a row? Probably time to add more green bricks to the tower.
77/ It is clueless and naive to treat experimental success and failure as a cross you must nobly bear, outside of your control. Fatal experimental failure risk should be no higher than it needs to be.
78/ Gradatim ferociter allows you to control your success rate and choose how much to win. And no, winning (or even not-losing) all the time is not the optimum to shoot for. That's a recipe for stagnation.
79/ You could make a fun tetris-like game where you get a starter tower and a supply of new bricks and blobs that you must then build with, with avalanches happening all the time.
80/ A good game to train kids to think this way would be to give them a starter tower of say 10 bricks and blobs, and a supply of 100 bricks and blobs that arrive one by one.
81/ Here's the kicker: after the 100 bricks/blobs have been incorporated, the final score isn't the height reached. The game then runs an "aging" simulation of another 1000 rounds. The terminal height is your score.
82/ While Jenga trains house-of-cards and tower-of-junk instincts, and Tetris trains grade-A stagnation instincts, this Gradatim Ferociter game would train your Great Works instincts (anyone want to build it?).
83/ But ultimately, this tower metaphor, games based on it, and even the 2x2 of fragility versus knowledge risk are just scaffolding.
84/ The real question is: do you want to do satisfying things with your life, or simply gloat about others failures, cheat them, or live off faddish stimulation-seeking?
85/ If you want satisfaction, then you don't have to look to space programs or other grand things to understand gradatim ferociter. The best example is watching a baby learning to walk.