Possible end of the great stagnation: Childhood's End

How we might drive ourselves into a technological cul-de-sac, and then drive ourselves out of it as the recombinations of knowledge take hold

[Part of an ongoing series - one, two, three, four, five, six]


There are as many ways to argue how the great stagnation might end as there are people getting excited by new technologies. Excitement is a decent proxy for progress even if it's not even close to being predictive. But still, excitement is hard to deny, because it's ... exciting ... so we should probably have a look and see how happy we should be.

In a similar trajectory, we need to see that it's the application of hard sciences that's hard, not just a discovery. The great stagnation says that both are in productivity declines, but the efforts needed for both are different.

At least one of them is difficult to justify beyond identifying new areas to research and breaking through the bureaucratisation barrier where senior scientists influence research arenas and implicitly reduce potential dispersion.

Elon Musk says that it's rocket engineers that got us to the moon, not rocket scientists.

While the stagnation in any particular domain is an indication of maturity of the field-paradigm combination, it doesn't indicate the emergence of new fields or new paradigms.

I am not talking about the emerging science of social networks, which seems to be an ugly stepchild trying to mix psychology and network theory with the bare minimum of empirical analysis. But perhaps something like a new emergence of genetics based medicine. Or quantum computing. Or biotechnology.

However the emergence of a new paradigm doesn't immediately tell us whether it's a step change, an intermediate boost, or a flash in the pan curio. In fact almost by definition it's likely to be underestimated, since otherwise it wouldn't be a novel idea in the first place.

Backwards looking explanatory indicator variables would include:

  • Will the TFP increase with all the supposed progress going on now?

  • Will the speed of new innovations, or applications thereof, per capita speed up in specific sectors?

  • Will there be simultaneous emergence of progress in multiple sectors?

Since a large percentage of our conviction in any conclusion will come from eventual backwards looking analyses, we also need something forward looking to figure out if we're on the right track. So before assessing whether we're now in an age of new productivity, a few pointers on what would I want to see:

  1. A possible world where the great polarisation has room to end: where the intra firm labour cost premia isn't as high as it is today. This can be done by simultaneous growth in multiple sectors of the economy, for example

  2. A burgeoning solution to the cost disease: where lower skilled jobs are still enough for basic consumption

  3. Clear knowledge growth trajectory: A true surfeit of awesome avenues of research that's opening up, enough that B+ students start looking like A+ students.

  4. Creation of knowledge clusters: Most historic great discovery domains have come with the creation of rather small clusters of people. It's less an indication of a small world effect than a view that smart people when put together with other smart people, when they're at the ground floor of a new discovery chain

What would bring us this new wind beneath our sails? Answer from a technologist is always technology, and it's no different in this case. So, a quick analysis of a few of the top emergent technology trends that have all gotten press on how they'll change the world forever:

  • CRISPR and its brethren

    • On the plus side it could create a step change in medicine - actual improvement wise and cosmetic; for improvements in existing processes and the creation of whole new options

    • Will that be enough as a step change? Presumably yes at the same range as penicillin (towards the optimistic end) and keyhole surgery (at the low end)

    • Will it affect inequality? Like everything on a cost curve I'd presume it'll worsen it at first, from an access POV, and then flow through to the rest of the populace later on

    • Will it improve general productivity? Most likely. There's one benefit of getting healthier and living longer!

  • Modular construction

    • This falls into reducing inefficiencies part of the economy

    • This could help increase the amount of construction done and reduce the externalities, which would be useful

    • Mostly it could help reduce the great polarisation by negatively impacting the cost disease, at least for shelter

  • Material science

    • Too early to tell but it could have impact across multiple sectors and create whole new possibility spectrums

    • It'll probably create new industries - cost curves are kind of magic in that when things get cheap enough everything changes!

  • Space race

    • While opening up the technological frontier will help push some possibilities along, it's unclear where the impact actually will be felt

    • There's potential for manufacturing, tourism, and better communications tech

    • This will undoubtedly be beneficial overall, especially since it brings forth the primacy of manufacturing, but impact on the rest has to filter through the cost disease lens

  • AI

    • The ultimate general purpose technology. If it can work ... Logically i can't say why it wouldn't, but history hasn't been kind to hubris in this area

    • Specialised applications will of course help improve aspects of existing economic activity but going above that will require some version of a new economic reality that doesn't revolve around amazing programmers dreaming up new tech and plebs generally getting replaced

    • The efficiency increase part of the tech will be helpful if there's also an impact on the cost disease aspect

When the dust is settled there's likely to be increased productivity amongst the scientists for some parts of material science and biology, and continuing in AI within computer science. All applications focused. There's also likely to be a continuation of the trend towards more robotics research and success with it especially for vertical specific and entertainment applications.

But will it be a step change in the rate going forward? It's possible. Of course it is. But without falling deeper into the incredulity problem, is it likely?


A slightly different tack. The way a company grows makes a good case study to explore how growth actually works. Make stuff, sell stuff, get money. Even other economic indicators like GDP measure up roughly the same way. And since we briefly talked about physics here's a model that has physics envy at its heart in terms of simplicity.

Let's play with a model.

The purpose isn't to make something realistic, but to hone an intuition on how progress could actually work. At least for the first part I feel we've gotten there too.

The first clue here is that it shows that any steady improvement in the top line comes from increased productivity or increased numbers. Those are the only ways to bump this rate up.

And yet, it's clear that there's a clear problem with this mathematics. If the rate could have been sustained, pretty soon everything would be subsumed by it. That's just how exponentials work.

For example, let's keep the productivity the same and see what happens when the population is made to grow exponentially.

Looks like so does the overall figure. We've just arrived at the mystery of why GDP seems to grow so damn steadily, at least in a Malthusian era sense. Not as a full explanation of course since the real world is messy, but at least as an explanatory variable.

But what about the GDP per capita growth after? Doesn't that break the narrative?

Yes it does. But it's also true that at the early stages exponential curves are bloody hard to tell apart. So while it might've looked like productivity was standing still, it might've just been slowly accumulating and growing, to the point where a bunch of stuff that was discovered could then be used. So you could combine the calculus from 1700s with thermodynamics from 1700s and 1800s and build an engine!

There's another factor that's been omitted from the model. That's capital.

Capital is funny in that it can be accumulated. If you build something, it's been built. You don't need to rebuild it. It's like buying a perpetual license vs saas, if you indulge the venture capitalist for a moment.

While it does need maintenance to function properly, it allows one to continue creating things by using things that've been created before.

Unlike ideas though, which are assumed to drive productivity increases, capital depreciates. Which means the effort to keep the lights on (KTLO in annoying, but true, industry jargon) is large and continuous. There's a reason that comprises two thirds of all large company IT spends. Gregory Clark writes in his 2005 paper

The early economy did not lack technological advances, it was just that most of these were in goods that did not appear in the consumption bundle of the average consumer: imported spices, sugar, books, gunpowder, paints, silk textiles, glass, and paper. And there are many goods or services which were improved where we do not even have a price: clocks, music, theater, art, eyeglasses, and newspapers for example.

Does this sound like any particular time in history? Maybe reminiscent of our griping that while we have satellites and slick cars and keyhole surgery and mRNA vaccines made in a jiffy and more entertainment than we can shake a stick at and supercomputers in our pockets, we still seem to live in a declining age?

The answer is that we might be. Or we just might be living in a polarised age where the things group A wants is available in plenty and at the right prices, and the things group B wants is not. The problem is that the things group B wants include healthcare and education and shelter. But they're getting the things group A cares more about, which is cellphones and cars and rockets that shoot up into space.

Luckily group A also wants gene editing and medicines and to live forever. So maybe we should be trying to cushion group B while technology growth runs its course, rather than restrict group A?


In the seminal novel by Arthur C Clarke, the Overlords help the Overmind help other, evolutionary advanced, species come and join them. The generation they shepherd to this end is the last generation of human children, and they themselves cannot do it because they've hit an "evolutionary cul-de-sac".

Despite all their powers and their brilliance, the Overlords were trapped in some evolutionary cul-de-sac. Here was a great and noble race, in almost every way superior to mankind; yet it had no future, and it was aware of it. In the face of this, George’s own problems seemed suddenly trivial.

They would never know how lucky they had been. For a lifetime, mankind had achieved as much happiness as any race can ever know. It had been the Golden Age. But gold was also the color of sunset, of autumn: and only Karellen’s ears could catch the first wailings of the winter storms.

While the metaphor might fail if you think about it for more than a second, the picture it paints feels all too real. Despite the incredible advances we've been seeing in select industries, in others we've been stuck in neutral. We find ourselves in an evolutionary cul-de-sac for technologies like airliners.

We could go faster than Mach 1, but then we'd have to redesign the wings and the whole body to reduce drag, which means we almost need a discontinuous Transformers style design where an aircraft has a particular shape before Mach 1, and one after.

Clearly needs a bit of work to get there! One can see why this isn't the highest priority for Boeing and Airbus.

But then there's Boom Supersonic, a recent unicorn, trying to actually create a new design and take it to market. So a discontinuous paradigm is maybe exactly what's needed here too!

But maybe that's fine. Maybe there's a punctuated equilibrium theory that can help here that demonstrates that steady growth is the norm and at some point the exponentials just fall apart into a version of collapse. And if we're really smart and really prepared, that collapse becomes a reprieve rather than a death knell. It's not that it needs to be steady, just that there needs to be regular upwards momentum.

To solve the issues of the great polarisation require a society that needs to allow all participants to be able to have good lives. Technology can help here, as it has in the past. But there needs to be a direct and incontrovertible impact of that technology in either directly hitting the cost disease, to not make the basics of life unachievable for a vast tract if society, or the great stagnation through creation of new industries that employ lots of people across socioeconomic and educational strata, and which impacts the other sectors too in the broader economy.

In fact the latter was the promise of the information technology revolution. Unfortunately we've been stuck in the efficiency maximisation paradigm through increasing hyperspecialisation within jobs making almost all of them eminently replaceable. But that could change if there were truly a way to break through and reach the benefits of the internet as originally advertised!

Tyler seems to believe this is clearly the way we get out of the great stagnation.

It might very well be, but it would need a redefinition that assures us how this actually increases our aggregate productivity rather than for a small group of infovores who survived childhood's end. Currently the benefits that it accrues to one segment of the population also makes a vast swath of the population rather replaceable. Even that Venn diagram has actual overlapping segments. Just because your programming has improved with the net doesn't mean you won't be replaced even easier today, either by another programmer, by an outsourced programmer, or by automation.


A detour back into venture capital. When we look to fund a company there are a large number of variables that matter. To get to success, you need a company that can demonstrate the following, in order:

  1. Technological feasibility - can the damn thing be built?

  2. Customer adoption feasibility - will the damn thing be actually bought by anyone?

  3. Financial feasibility - can the damn thing sell?

  4. Competitive feasibility - will the competitors crush you?

  5. Regulatory feasibility - will you be allowed to sell the damn thing?

Let's take a look at an awesome startup and see how we can think about this:

  1. Easy enough to figure out and test, within a few million you can know if an idea is reasonably viable. For biotech, it could take a billion dollars! For hardware, tens if not hundreds of millions.

  2. You could argue about the positioning here, but you can ask the customers in software or hardware. In biotech you kind of know. You don't need to ask "Would you like your cancer cured?"

  3. Easiest in biotech, again. People who need medicines will buy medicines. People who don't need medicines will buy vitamins.

  4. This is where speed to market is helpful, as is some heavyweight media behind you.

  5. Regulatory stuff tends to be the biggest barriers for companies in medicine, pharma, aerospace, finance, insurance, <insert your favourite sector to bash here>, social media now ... the list goes on.! Nuclear research hit a lead wall (see what I did there?) in this department.

It's worth noting that one reason why VCs like software is because you can test and fail pretty fast there. It makes sense that we leap after the same companies like well trained lemmings. You generally only need to throw sales and marketing dollars at the problem once you know it can be solved! And that seems like an easier problem than funding lab-coats for a decade and hoping they find something. Seems is an operative word there, since power laws are unforgiving.

So when does the innovation inverse of Baomol’s disease come to play?

At what point does sufficient innovation in one area help push the frontier forward in others?

For instance, as we got better in the hardware side of the chip business, that clearly made us much better in the information wrangling side.

As we got better at software, i.e., the collaboration and communication and analysis parts, we started to let that bleed into other parts of the world.

Science became even more analytical because empiricism was now even more possible.

Medicine followed a similar path too.

So did architecture. And engineering. And design itself.

Once software seeped in, you could outsource your pre hoc thinking to a disparate group including customers and experts and other decision makers. You could experiment more freely, which changed the paradigm.

But as it kept getting better we could also make a credible argument for there being an even better cognitive layer. Not only could we outsource our hypothesis generation, we could outsource even some entire classes of problems.

This is about as general purpose as it gets. It's just not reality yet. The world is too messy, and dealing with it needs people who are also messy.

But throughout this journey of one industry getting better, it's pushed all others too, primarily in efficiency increasing ways that provide changes in scale rather than changes in scope:

  • Medical devices - they've gotten substantially better, and cheaper, over the years

  • Consumer electronics - enough said

  • Cars and general transport - safer and more reliable than ever before. They're practically consumer electronics devices themselves.

  • Manufacturing - we produce more than ever before while employing far less people and consumers buy things for less than ever before

But with the advent of some levels of cognition in there, the potential for actual change in industries is even more drastic.

Medicine for one could change dramatically. It's no longer about more efficient bookkeeping and patient charts, it's about the actual diagnosis. In science too, it's about hypothesis generation, rather than later stage better statistical testing.

But those are implications of the application of a tech that we believe is forthcoming.

There's also the potential to do things that have never been done before. New material designs that could transform entire industries from commerce to construction. New molecule based drug discovery and manufacturing changes the way we create medicines.

Could this have happened without the revolution in information technology? Perhaps. But it's hard to tell how or indeed when. Just like Von Neumann played with AI and had great ideas but didn't get anywhere until someone thought to repurpose GPUs, sometimes you need the rest of the world to catch up.

There are dual forces pulling each industry. If you computerise, you get more efficient. But so do your (usually larger) competitors who can afford it too. So the larger guys who reap the benefits get larger while the others don't.

Meanwhile those who make the efficiency come about quite naturally do much better. Everyone wants them!

At the same time as this efficiency is passed on to consumer surplus, the same producers who are getting squeezed are also consumers. They're the ones getting their purchasing power stripped in this scenario.

So the question is then less about whether the constant rates of productivity growths we saw now coming to a (natural) halt, and more about when we can see an uptick in innovations across multiple sectors.

Let's try a small mental model to see if this makes any sense.

First we assume a set of technologies, and also assume a base "progress level" which you can think of as a growth trajectory. Second, the growth in each tech increases as a function of previous period + RAND()*RANDBETWEEN(0,1) - this is to help introduce some punctuated growth for the sectors. So in effect there's some random chance of it being pushed forward by a little bit, or stall for a period. And thirdly, if cumulative growth in all technologies raises the average of the technology progress to be > average of the period where everything started, a new technology magically emerges! This is a way of saying once we know enough about a bunch of technologies on average, or when we know a hell of a lot about one technology, we figure out a way to create something new.

Do bear in mind that the word technology here is a bit of a placeholder. It could also be a fundamentally new science that we've figure out. Is quantum computing a new technology? I think it could very well be. What about CRISPR? Same. What about AI as we know it today, in the sense that it can help us search things better? Probably not, unless it breaks through some barrier that allows it to be more broadly deployed in the real world. Once autonomous vehicles enter the equation, this absolutely becomes one.

The implications of this are that we can assess the average technological progress in each age - by definition indexed to 1st period.

Second, the growth rate according to this model declines over time - mostly because I've made the growth rate in tech linear. This isn't a philosophical point, I just didn't care to play formulae with google sheets that much.

And once we do that, we see a couple of things. One is that the number of technologies keep increasing over a period of time. And as each new technology emerges, the number of potential combinations of all existing knowledge bases or technologies increases rather rapidly.

This to me is one reason why we don't really have polymaths much anymore compared to the olden days. There just are way too many things one needs to be an expert in to become a polymath, and since the fields themselves are getting deeper, it's just bloody hard!

But if we follow the logic, reaching Childhood's End is inevitable. If a new technology emerges when one reaches saturation, or the combined technological might of the rest reaches a certain threshold, there will be new emergences of technology. Without enough ability to recombine though, the combinations to explore to find the new technology and its applications shoot right past us. And recombinations require someone to get on top of [X] number of fields, where [X] can vary according to the current state of technology

Every sector needs to look at new ways of pushing the boundary forward. To innovate. If you're a company your job is easier when there's a ready made value enhancer sitting in the efficiency bucket from IT. Even if that might not be long term effective necessarily, it becomes a default. And when you employ your small amount of "innovation resources", the 10% of Transform from the Run-Grow-Transform budget, you can't help but end up in a cul-de-sac.

The way out requires either we start searching for harder answers and remove the organisational friction that exists today, or we wait for the technologies to reach a point that the externalities can't help but bubble over.!