Here's a roundup of top developments in the biotech space over the last 24 hours.Scaling The Peaks(Biotech stocks hitting 52-week highs on Oct. 29)Cabaletta Bio Inc(NASDAQ:CABA)Edwards Lifesciences Corp(NYSE:EW)ESSA Pharma Inc(NASDAQ:EPIX)GENMAB
For the past several months, I’ve been immersed in nineteenth century history. Specifically, the history of interchangeability in technology between 1765, when the Système Gribeauval,the first modern technology doctrine based on the potential of interchangeable parts, was articulated, and 1919, when Frederick Taylor wroteThe Principles of Scientific Management.
Here is the story represented as aDouble Freytag diagram, which should be particularly useful for those of you who have readTempo. For those of you who haven’t, think of the 1825 Hall Carbine peak as the “Aha!” moment when interchangeability was first figured out, and the 1919 peak as the conclusion of the technology part of the story, with the focus shifting to management innovation, thanks in part to Taylor.
The unsung and rather tragic hero of the story of interchangeability wasJohn Harris Hall(1781 – 1841), inventor of the Hall carbine. So I am naming my analog to Moore’s Law for the 19th centuryHall’s Lawin his honor.
The story of Hall’s Law is in a sense a prequel to the unfinished story of Moore’s Law. The two stories are almost eerily similar, even to believers in the “history repeats itself” maxim.
Why does the story matter? For me, it is enough that it is a fantastically interesting story. But if you must have a mercenary reason for reading this post, here it is: understanding it is your best guide to the Moore’s Law endgame.
So here is my telling of this tale. Settle in, it’s going to be another long one.
InA Brief History of the Corporation,I argued that there were two distinct phases — an early mercantile-industrial phase that was primarily European in character, extending from about 1600 to 1800, and a later Schumpeterian-industrial phase, extending from about 1800-2000, that was primarily American and Russian in character.
Each phase was enabled by a distinct technological culture. In the early, British phase, ascientific sensibilitywas the exception rather than the rule. The default was the craftsman sensibility. In the later, American-Russian phase, the scientific sensibility was the rule and the craftsman sensibility the exception (it is notable that the American-Russian phase was inspired by French thought rather than British; call it Napoleon’s revenge).
What was this (much romanticized today) craftsman sensibility?
Consider this passage about the state of steel-making in Sheffield, the leading early nineteenth century technology center for the industry, before the rise of American steel. The quote is from Charles Morris’ excellent book The Tycoons, my primary reference for this post (it is nominally about the lives of Rockefeller, Carnegie, J. P. Morgan and Jay Gould, but is actually a much richer story about the broad sweep of 19th century technology history; I am not done with it yet, but it has been such a stimulating read that I had to stop and write this post):
Making a modest batch of steel could take a week or more, and traditional techniques were carefully passed down from father to son; one Sheffield recipe started by adding “the juice of four white onions.”
Morris attributes the onion story to Thomas Misa’sNation of Steel,which is now on my reading list.
American steel displaced British steel not because it was based on the Bessemer and open hearth processes (Bessemer was English), but because the industry was built from the ground up along scientific lines, with no craftsman-baggage slowing it down.
The interesting thing about this recipe for onion steel is that it illustrates both the strengths and the weaknesses of the craftsman sensibility. You can only imagine the tedious sort of uninformed experimentation it took to consider adding onions to a steel recipe. There is something beautiful about the absence of preconceived notions in this sensibility. No modern metallurgist would even think to add onions to a metal recipe.
On the other hand, if a modern metallurgist were faced with data showing that onions improved the properties of steel, he or she would not rest until they’d either disproved the effect, or explained it in less bizarre terms. The recipe would certainly not get passed down from “father to son” (“mentor to mentee” today) unexplained.
What America brought to manufacturing was a wholesale shift from craftsman-and-merchant thinking about technology and business to engineer-and-manager thinking. The shift affected every important 19th century business sector: armaments, railroads, oil, steel, textile equipment. And it created a whole new sector: the consumer market.
But this was not the result of an abstract, ideological quest for scientific engineering and manufacturing, or a deliberate effort to replace high-skill/high-wage craftsmen with low-skill/low-wage/interchangeable machine operators.
It was a consequence of a relentless pursuit of interchangeability of parts, which in turn was a consequence of a pursuit of greater scale, profits and competition for market share (which drove greater complexity in offerings) on the vast geographic canvas that was America. Craft was merely a casualty along the way.
So why was interchangeability of parts a holy grail in this pursuit?
Interchangeability, Complexity and Scaling
The problem is that even the highest-quality craft does not scale. When something like a rifle is mass-produced using interchangeable parts, breakdowns can be fixed using parts cannibalized from other broken-down rifles (so two broken rifles can be mashed-up to make at least one that works) or with spare parts shipped from an warehouse. Manufacturing can be centralized or distributed in optimal ways, and constantly improved. Production schedules can be decoupled from demand schedules.
A craftsman-made rifle on the other hand, requires a custom-made/fitted replacement part. The problem is especially severe for an object like a rifle: small, widely-dispersed geographically, and liable to break down in the unfriendliest of conditions. Conditions where minimizing repair time is of the essence, and skilled craftsmen are rather thin on the ground. It is no surprise that the problem was first solved for guns.
Let’s do some pidgin math to get a sense of what a true mathematical model might look like.
Roughly speaking, scaling production for any mechanical widget involves three key dimensions: production volumeV, structural complexityS(the number of interconnections in an assembly is a good proxy measure forS,just like the number of transistors on a chip is a good proxy for its complexity) and operating tempo of the machine in use,T(since the speed of operation of a machine determines the stress and wear patterns, which in turn determines breakdown frequency; clock-rate is a similar measure for Moore’s Law).
For complex widgets, scaling production isn’t just (or even primarily) about making more new widgets; it is about keeping the widgets in existence in the field functioning for their design lifetime through post-sales repair and maintenance. The greater the complexity and cost, the more the game shifts to post-sales.
You can combine the three variables to get a rough sense of manufacturing complexity and how it relates to scaling limits. Something likeC=SxTprovides a measure of the complexity of the artifact itself. Breakdown rateBis some function of complexity and production volumes,B=f(C, V).At some point, as you increaseV,you get a corresponding increase inBthat overwhelms your manufacturing capability. To complete this pidgin math model, you can think in terms of someB_max=f(C, V_max)above whichVcannot increase without interchangeability.
Modern engineers use much more sophisticated measures (this crude model does not capture the tradeoff between part complexity and interconnection complexity for example, or the fact that different parts of a machine may experience different stress/wear patterns), but for our purposes, this is enough.
To scale production volume aboveV_max without introducing interchangeability, you have to either lower complexity and/or tempo or increase the number of skilled craftsmen. The first two are not options when you are trying to out-do the competition in an expanding market. That would be unilateral disarmament in a land-grab race. The last method is simply not feasible, since education in a craft-driven industrial landscape means long, slow and inefficient (in the sense that it teaches things like onion recipes) 1:1 apprenticeship relationships.
There is one additional method that does not involve interchangeability: moving towards disposability for thewholeartifact, which finesses the parts-replacement problem entirely. But in practice, things get cheap enough for disposability to be a workable strategy only after mass production is achieved. Disposability is rarely a cost-effective strategy for craft-driven manufacturing, though I can think of a few examples.
These facts of life severely limited the scale of early nineteenth century technology. The more machines there are in existence, the greater the proportion of craftsmen whose time must be devoted to repair and maintenance rather than new production. Since breakdowns are unpredictable and parts unique, there is no way to stockpile an inventory of spare parts cheaply. There is little room for cannibalization of parts in the field to temporarily mitigate parts shortages.
What was needed in the 19th century was a decoupling of scaling problems from manufacturing limitations.
Interchangeability and the Rise of Supply Chains
Interchangeability of parts breaks the coupling between scaling and manufacturing capacity by substituting supply-chain limits for manufacturing limits. For a rifle, you can build up a stockpile of spare parts in peace time, and deliver an uninterrupted supply of parts to match the breakdown rate. There is no need to predict which part might break down in order to meaningfully anticipate and prepare. You can also distribute production optimally (close to raw material sources or low-cost talent for instance), since there is no need to locate craftsmen near the point-of-use.
So when interchangeability was finally achieved and had diffused through the economy as standard practice (a process that took about 65 years), demand-management complexity moved to the supply chain, and most problems could be solved by distributing inventories appropriately.
These happy conditions lasted for nearly a century after widespread interchangeability was achieved, from about 1880 to 1980, when supply chains met their own nemesis, demand variability (thatproblem was partially solved using lean supply chains, which relied in turn on the idea of interchangeability applied to transportation logistics: container shipping. But I won’t get into that story here, since it is conceptually part of the unfinished Moore’s Law story).
The price that had to be paid for this solution was that the American economy had to lose the craftsmen and work with engineers, technicians and unskilled workers instead. This creates a very different technology culture, with different strengths and weaknesses. For example the scope of innovation is narrowed by such codification and scientific systematization of crafts (prima facienutty ideas like onion steel are less likely to be tried), but within the narrower scope, specific patterns of innovation are greatly amplified (serendipitous discoveries like penicillin or x-rays are immediately leveraged to the hilt).
Why must craft be given up? Even the best craftsmen cannot produce interchangeable parts. In fact, thecraftis practically defined by skill at dealing with unique parts through carefully fitted assemblies. (“Interchangeability” is of course a loose notion that can range from functional replaceability to indistinguishability, but craft cannot achieve even the coarsest kind of interchangeability at any meaningful sort of scale).
Put another way, craft is about relative precision between unlike parts. Engineering based on interchangeability is about objective precision between like parts. One requires human judgment. The other requires refined metrology.
From Armory Practice to the American System
It was the sheer scale of America, the abundance of its natural resources (and the scarcity of its human resources), that provided the impetus for automation and the interchangeable parts approach to engineering.
As agriculture moved westward through New York, Pennsylvania and Michigan, the older settled regions began to turn to manufacturing for economic sustenance. The process began with the textile industry, born of stolen British designs around what is now Lowell, Massachusetts. But American engineering in the Connecticut river valley soon took on a distinct character.
Like the OSD/DARPA/NASA driven technology boom after World War II, the revolution was driven by the (at the time, fledgling) American military, which had begun to acquire a mature and professional character after the war of 1812 (especially during the John Quincy Adams administration).
The epicenter of the action was the Springfield Armory, the PARC of its day, and outposts of the technology scene extended as far south as Harper’s Ferry, West Virginia.
John Hall was among the hundreds of pioneers who swarmed all over the Connecticut valley region, dreaming up mechanical innovations and chasing local venture capitalists, much like software engineers in Silicon Valley today.
There were plenty of other extraordinary people, including other mechanical engineering geniuses like Thomas Blanchard, inventor of the Blanchard gun-stock lathe (which was actually a general solution for turning any kind of irregular shape using what is known today as a pattern lathe). By the time he was done with gun stocks, a bottleneck part in gun-making, with all sorts of “subtle curves along multiple axes” he had created a system of 16 separate machines at the Springfield Armory that pretty much automated the whole process, squeezing out all craft of what had been the single most demanding component in gun-making.
British gun-making was like British steel-making before people like Blanchard and Hall blew up the scene. Here is Morris again:
The workings of the British gun industry were reasonably typical of the mid-nineteenth-century manufacturing. It was craft-based and included at least forty trades, each with its own apprenticeship system and organizations. The gun-lock, the key firing mechanism, was the most complicated, while the most skilled men were the lock-filers…[who]… spent years as apprentices learning to painstakingly hand-file the forty or so separate lock pieces to create a unified assembly… When the Americans breezily described machine-made stocks, and locks that required no hand fitting, they sounded as if there were smoking opium.
Among the opium-smoking geniuses, Blanchard at least enjoyed a good deal of success. Hall did not.
He put together almost the entire “American System” through his single-minded drive, in the technology-hostile Harper’s Ferry location far from the Connecticut Valley hub. When he was done, he had created an integrated manufacturing system of dozens of machines that produced interchangeable parts for every component of his carbine. Even parts from production runs from different years could be interchanged, a standard some manufacturing operations struggle to reach even today.
The achievement was based on relentless automation to eliminate human sources of error, increasingly specialized machines, and rigorous and precise measurements (there were three of every measurement instrument, one for production use, one for calibration, and a master instrument to measure wear on the other two).
It was a massive systems-engineering accomplishment. The Hall carbine was the starter pistol for the American industrial revolution.
Overtake, Pause, Overdrive
Hall did not reap much of the rewards. Thanks to unfortunate exploitative relationships (in particular with a shameless patent troll,William Thornton, a complete jerk by Morris’ account), he was banished to Harper’s Ferry rather than being allowed to work in Springfield. And his work, when completed, was acknowledged grudgingly, and with poor grace. The Hall carbine itself was obsolete by the time his system was mature, and others who applied it to newer products reaped the benefits.
Between 1825 and the 1910s, the methods pioneered by Hall spread through the region and beyond, and were refined and generalized. In the process, first America, and then the world, experienced a Moore’s Law type shock: rapidly increasing standards of living provided by an increasing variety of goods whose costs kept dropping.
Culturally, the period can be divided into three partially overlapping phases: an overtake phase (1851 – 1876) when America clearly pulled ahead of Britain as the first nation in the technology world, a “pause” represented by the recession of the 1870s, and finally an over-drive phase beginning in the 1880s and continuing to the beginning of World War I, when the American model became the global model (and in particular, the Russian model, as Taylorism morphed into state doctrine).
Overtake: 1851 — 1876
The overtake phase has a pair of useful bookend events marking it. It began with the 1851 Crystal Palace Exhibition, the first of the great 19th century world fairs, when the world began to suspect that America was up to something (McCormick’s harvester and Colt’s revolver were among the items on display), and ended with the 1876 Centennial World Fair in Philadelphia, when all remaining doubt was erased and it became obvious that America had now comprehensively overtaken Britain in technology.
When Britain finally caught on and hastily began copying American practices following the Philadelphia fair, the result was a revitalization of British industry that produced, among other things, the legendaryEnfield rifle (the rifle subplot in the story of interchangeability has an interesting coda that is shaping the world to this day, theRussian AK-47, as pure an example of the power of interchangeability-based mass manufacturing as has ever existed).
It wasn’t just guns. In every industry America began to show up Britain. Much of the credit went to showboating hustlers who claimed credit for interchangeability and the American System/Armory Practice, and made a lot of money without actually contributing very much to core technological developments. These included Eli Whitney of cotton gin fame, the McCormicks of the harvester, Samuel Colt (revolvers) and Isaac Singer (sewing machines). While they certainly contributed to the development of individual products, the invention of the American model itself was due to technologists like Blanchard and John Hall.
In the initial decades of the overtake, fueled in part by opportunity (and profiteering) associated with the Civil War and government subsidized building out of the railroad system, much of the impact was invisible. But by the 1890s, as the infrastructure phase was completed, the same methods were unleashed on everyday life, creating modern consumer culture and the middle class within the short space of a single generation.
The Pause: the 1870s
The Civil War looms large as the major political-economic event in this history (1861 – 1865), but the bulk of the impact was felt in the decade that followed, once the dust had settled and interrupted infrastructure projects were completed.
This impact took the form of the rather strangelong recession of the 1870s, which was very culturally very similar to the one we are currently experiencing (increased economic uncertainty and fall in nominal incomes, hidden technology-driven increases in standard of living, foundational shifts in the nature of money — back then it was a greenbacks vs. gold thing).
One way to understand this process is that the infrastructure phase had created both tycoons and an extremely over-leveraged economy. It was the uncertain gap between “build it” and “they will come.” It was a huge, collective pause, a national decade of breath-holding as people wondered whether the chaos unleashed by the new infrastructure would create a better social order or destroy everything without creating something new in its place.
Starting in the 1880s, the bet began paying off in spades. The recession ended and the over-drive boom began, as people figured out what to do with the newfound capabilities in their environment.
Overdrive: 1880s — 1913
A good early marker here is probably the first Montgomery Ward catalog in 1872, the first major sign that the new infrastructure allowed old businesses to be rethought, leading to the creation of the modern consumer economy.
The mail-order catalog was by itself a simple idea (the first catalog was just a single page), but the reason it disrupted old-school merchants was that it relied on all the infrastructure complexity that now existed.
Trains that ran on reliable schedules, to deliver mail, telegraph lines that brought instant price updates on western grain to the East Coast, steel to build everything, oil and electricity to light up (and later, fuel) everything, new financial systems to move money around, and of course, the application of interchangeability technology to everything in sight.
It took Sears, starting in 1888, to scale the idea and truly take down the merchant elites who had defined the old business culture, but by World War I, middle-class consumer culture had emerged and had come to define America. In another 50 years, it would come to define the world.
It was such a powerful boom that globally, it lasted a century, with two world wars and a Great Depression failing to arrest its momentum (as an aside, I wonder why people pay so much attention to the 1930s depression to make sense of the current recession; the 1870s recession makes for a far more appropriate comparison).
What ultimately killed it was its own success. Semiconductor manufacturing probably represents the crowning achievement of the Armory Practice/American System that began with a lonely John Hall pushing ahead against all odds at Harper’s Ferry.
Moore’s Law was born as the last and greatest achievement of the parent it ultimately devoured: Hall’s Law.
When you step back and ponder the developments between 1825 and 1919, it can be hard to make sense of all the action.
There is the pioneering work in manufacturing technology. There is the explosion of different product types as the American System diffused through the industrial landscape. There is the story of the rise of the first tycoons. There is the rise of consumerism and the gradual emergence of the middle class. There is the connectivity by steam and telegraph.
Then there is the increasingly confident and strident American presence on the global scene (especially through the World Fairs, two of which I already talked about). And of course, you have the Civil War, the California Gold Rush, the cowboy culture that existed briefly (and permanently reshaped the American identity) before Jay Gould killed it by finishing the railroad system.
There was the rise of factory farming and the meatpacking and refrigerator-car industries together killing the urban butcher trade and suddenly turning Americans into the greatest meat eaters in history. Paycheck economics took over as the tycoon economy killed the free agent.
In fact, there was a lot going on, to put it mildly. And that was just America. The rest of the world wasn’t exactly enjoying peace and stability either. Perry had kicked down the doors of Japan, Opium wars had ravaged China, the East India Company (the star of myHistory of Corporationspost) had been quietly put out to pasture and the Mughal empire had collapsed. The Ottomans were starting on a terminal decline. Continental Europe had begun its century-long post-Napoleon march towards World War I (the US Civil War served as a beta test for the post-Bismarck model of total war, just as the Spanish Civil war served as a beta test for World War II).
But just as Moore’s Law provides something of a satisfying explanatory framework for almost everything that has happened in the last 50 years, the drive towards the holy grail of interchangeability provides a satisfying explanatory framework for much of this action. Here’s my attempt at capturing what happened (someone enlighten me if something like this has already been proposed under a different name) :
Hall’s Law: the maximum complexity of artifacts that can be manufactured at scales limited only by resource availability doubles every 10 years.
I believe this law held between 1825 and 1960, at which point the law hit its natural limits.
Here, I mean complexity in the loose sense I defined before: some function of mechanical complexity and operating tempo of the machine, analogous to the transistor count and clock-rate of chips.
I don’t have empirical data to accurately estimate the doubling period, but 10 years is my initial guess, based on the anecdotal descriptions from Morris’ book and the descriptions of the increasing presence of technology in the world fairs.
Along the complexity dimension, mass-produced goods increased rapidly got more complex, from guns with a few dozen parts to late-model steam engines with thousands. The progress on the consumer front was no less impressive, with the Montogmery Ward catalog offering mass-produced pianos within a few years of its introduction for instance. By the turn of the century, you could buy entire houses in mail-order kit form. The cost of everything was collapsing.
Along the tempo dimension, everything got relentlessly faster as well. Somewhere along the way, things got so fast thanks to trains and the telegraph, that time zones had to be invented and people had to start paying attention the second hand on clocks.
There is a ton of historical research on all aspects of this boom, but I suspect nobody has yet compiled the data in a form that can be used to fit a complexity-limit growth model and figure out the parameters of my proposed Hall’s Law, since it is the sort of engineering-plus-history analysis that probably has no hope of getting any sort of research funding (it would take some serious archaeology to discover the part-count, operating speed and production volumes for a sufficient number of sample products through the period to fit even my simple model, let alone a model that includes things like breakdown rates and actual, as opposed to theoretical, interchangeability).
But even without the necessary empirical grounding, I am fairly sure the model would turn out to be an exponential, just like Moore’s Law. Nothing else could have achieved that kind of transformation in that short a period, or created the kind of staggering inequality that emerged by the Gilded Age.
Break Boundaries and Tycoon Games
Both Moore’s Law and Hall’s Law in the speculative form that I have proposed, are exponential trajectories. These trajectories generally emerge when some sort of runaway positive-feedback process is unleashed, through the breaking of some boundary constraint (the termbreak boundaryis due to Marshall McLuhan).
The positive-feedback part is critical (if you know some math, you can guess why: a “doubling” law in a difference/differential equation form has to be at least a first-order process; something like compound interest, if you don’t know what the math terms mean).
Loosely speaking, this implies a technological process that can b