If you prefer to listen rather than read, this blog is available as a podcast here. Or if you want to listen to just this post:
I’ve been playing this game called Hexcells Infinite. It’s kind of like minesweeper, but with at least a half dozen ways of indicating how many “mines” there are in a given cell. The way play generally progresses is with long pauses of thinking interspersed with fairly rapid clicking once you figure something out because that initial insight cascades to reveal a bunch of nearby “mines”. It’s quite an enjoyable game and if you’re looking for something casual with a playtime of between 5 and 20 minutes I’d recommend it (there are actually three editions: normal, plus and infinite). But that’s not why I bring the game up. I bring it up because the manner in which it plays is an example of a very, very tiny S-curve.
What’s an S-curve? It’s a curve that looks kind of like a flattened “S”, it starts out nearly horizontal, turns up into something that looks exponential and then flattens out again at the top. Just as my hexcells game play starts with not much activity as I think, before going through a burst, then gradually tapering off as I run out of obvious moves. Anywhere positive feedback loops battle with constraints you’ll see S-curves, and they’ve been a topic of frequent discussion recently (at least in my corner of the internet).
As far as that discussion goes, I’m probably somewhat late to the game, but I think thus far people have mostly been focused on smaller S-curves, perhaps not as tiny as the one I experience when playing Hexcells, but fairly small nonetheless. I want to go in the exact opposite direction and focus on the possible existence of very large S-curves. And, in particular, whether we’re near the top of any of those curves.
Energy
One of the points which has been made in other spaces is that if you combine a series of S-curves that combination looks very much like exponential growth. For example, take something like Moore’s Law, which is the exponential growth in the number of transistors that can fit in a given space. At first glance this may seem like one curve, but in reality it’s a bunch of S-curves stacked on top of each other. You might have an S-curve associated with transistors and then another S-curve around advances with integrated circuits. Farther along there’s the S-curve related to various methods of lithography, and cpu architecture. But as each advance followed immediately on the heels of the last one, there was never a time for the Moore’s Law graph to reach the top of any given S-curve and flatten out. Though perhaps that’s finally about to happen.
My point in bringing this up is not to talk about computer chips, but to point out that something similar happened with energy. If you look at a graph of worldwide energy use, you’ll see a similar vaguely exponential curve, but you’ll notice that within that curve you have various smaller S-curves, sources of energy which start off small, grow really fast and then level off. First there was wood, then coal, then oil. And for a long time there was a lot of attention being paid to the inevitable leveling off of oil, or peak oil as it’s commonly known. Though, of course, just like with processors there was every reason to suspect that another S-curve would come along and keep the overall energy curve pointing up. Initially nuclear power seemed very promising as a candidate for this next innovation, but then it mostly stalled. Fortunately or unfortunately, depending on who you ask, something else came along, fracking, and a new curve started. There’s also, of course, renewables, which could easily be a blog post on its own.\
As I mentioned previously in this space energy production has been growing at somewhere north of 2%/year for centuries, basically through the stacking of the S-curves I’ve been talking about. This growth has been fundamental to the world we now live in, and it’s unclear what happens if that growth stops, but it’s probably bad. And when we tie all of the above together there are many reasons to think that we may be facing exactly that possibility. That we have reached some sort of inflection point. For example here are some of the questions I’m pondering:
1- Do we still have to worry about the S-curve of peak oil. Or is it now an S-curve of peak natural gas?
2- As I pointed out, much of progress seems based on maintaining a certain rate of energy growth. What happens if the technology is there, but the political will isn’t? For example with nuclear power, and possibly fracking.
3- Related, if fracking is problematic even without its contribution to greenhouse gas emissions, and nuclear is problematic despite its lack of the same. How does climate change factor into the continued use of certain sources of energy? So far it doesn’t seem to have had much of an impact either way.
4- In the past new technology was implemented as soon as it was feasible, with little regard to public opinion or politics. This is no longer the case. How does this new reality interact with our reliance on continual progress? Or with the diffuse harm that comes from technological innovation? (i.e. it’s one thing to demand 100% renewable energy, it’s quite another thing to actually make that switch.)
Antibiotics
I would offer up antibiotics and another example of a big S-curve. One that appears to definitely be plateauing out recently. I would also argue that unlike previous examples it’s less obviously a composite of lots of smaller S-curves. Yes, new antibiotics have been developed (though that process is getting harder and harder) but my impression is that most of the upward slope is entirely due to just having antibiotics available in the first place (i.e. penicillin) and that subsequent classes of antibiotics allowed us to hold our ground, but didn’t bring any big jump in effectiveness. All of which is to say that there is not some metaphorical nuclear power equivalent waiting to save us once antibiotics are no longer effective. We have one tool and we’ve already extracted most of the benefit.
Obviously I am not the first person to point this out, but my broader contention is that we may be reaching the top of a lot of our big S-curves and our effectiveness at dealing with the diminishing effectiveness of antibiotics could be indicative of how we deal with the other S-curves as they plateau out. So far the signs are not encouraging.
Manned Space Exploration
Manned space exploration has been in the news a lot lately. Not only is the 50th anniversary of Apollo 11 coming up next month, but both SpaceX and Blue Origins have announced plans to send humans to Mars. And then of course there’s Trump’s very… interesting(?) tweet from a few days ago:
For all of the money we are spending, NASA should NOT be talking about going to the Moon – We did that 50 years ago. They should be focused on the much bigger things we are doing, including Mars (of which the Moon is a part), Defense and Science!
Where does all of this put us as far as an S-curve for the manned exploration of space? I would argue that we’ve already experienced an S-curve, one which plateaued awhile ago. Remember the description of S-curves we started with. It begins with a positive feedback loop. When you’re talking about the Apollo missions this is a little vague, but obviously competition with the Soviets was a big part of it. After an initial burst things taper off as you run into constraints. On that end things are not vague at all, the constraints of manned space exploration are legion, particularly when you’re trying to do it at the government level.
That last bit is key, I would argue that we have run through the governmental S-curve already and that we’re at the beginning of a new S-curve, the manned exploration of space by private entities. In this new stage we’ll see some more innovations (like reusable rockets) but eventually even Musk and Bezos will run out of places where they can economize and improve, and things will reach another plateau. We’ve seen S-curves which stack one after the other and give the impression of continuous exponential growth. This, on the other hand, is an example of two curves with a long gap in between. Also once the current private entity fueled curve plateaus it’s unclear when or in what domain another one will start. And what’s even more uncertain is whether that will happen before or after we have a long-term sustainable presence somewhere other than Earth. My bet would be that it will definitely be before, and that there is no smooth path to the stars, or even Mars.
Scientific method
At last we finally arrive at the S-curve that worries me most of all, the S-curve of scientific discovery. For decades if not centuries it has been more or less an article of faith that scientific progress would continue to increase in essentially an exponential fashion, and indeed by some measures it still is, for example scientific output, measured in terms of scientific papers, doubles every nine years. But are all of those papers just as impactful as Einstein’s On the Electrodynamics of Moving Bodies? Definitely not, meaning that at best the number of scientific papers is a very rough proxy for scientific progress, not a direct measure of it. But even if you disagree, and argue that the ever doubling number of papers means that scientific progress hasn’t slowed down, there is absolutely no law that says that it never will. And many reasons to think that it’s already happening.
Not too long ago I read The Making of the Atomic Bomb by Richard Rhodes. It was just before I started reviewing everything I read, but maybe I’ll go back and pick that one up, because it was truly a great book. One of the things that was striking is how amazingly fruitful the pre-war years were for physics. Everywhere you turned people were uncovering new things, the structure of the atom, the existence of neutrons, the discovery of fission. (George Gamow also noticed this leading him to write the book Thirty Years that Shook Physics). All of this is a classic description of the bottom of the S-curve. As discoveries and scientists feed off one another it produces a positive feedback loop of understanding.
These days, we’ve got far more scientists working on things, publishing, as I mentioned far more papers, but the discoveries of the last 30 years have been much less consequential. All of the laws of physics where things are unchanging and easy to replicate, have largely been uncovered, or will require spending billions of dollars on a new particle collider. It’s pretty clear that all of the places where the scientific method was easy to apply have been mined out. That we have picked all of the low hanging fruit. The S-curve is starting to plateau as we bump up against various constraints
In part this is because much of science has moved on to experiments about human physiology and behavior, where there are numerous constraints. It’s difficult to establish control groups, things aren’t unchanging, and there are vast differences between individuals, meaning that instead of groundbreaking discoveries that shake our understanding of the universe we get small discoveries about how we just have to assume a “power pose” and it will immediately make us more confident. Worse than the smallness of these discoveries is the fact that 50% of the time they fail to replicate (like the research about the power pose). That sounds a lot like a plateau to me.
Tying all of this together, we have this idea that progress is a smooth curve moving ever upward towards a better and better future, and indeed this has been the case for the last few decades and in some cases for the last few centuries, but as I pointed out a couple of times, the bottom of an S-curve is indistinguishable from exponential growth. It’s only as you get farther along that the difference is apparent. And I would argue that we’re finally reaching the stage where it’s clear that most of the things we’ve come to expect from progress aren’t exponential, that they won’t grow forever, and that in fact we’re nearing the top of a lot of S-curves which have been powering civilization for a long time. And as they start to plateau it’s unclear what will happen
This is not to say that progress is over, even if most things should be viewed as an S-curve instead of something that grows exponentially, there are lots of S-curves remaining, and we’re still at the bottom/high growth part of many of them. But it’s unclear how much comfort this should give us. Saying that while we may be close to peak antibiotics, we’re nowhere near peak Facebook, is not particularly reassuring.
Undoubtedly lumping all trends under the heading of an S-curve will turn out to be too crude, some trends will end up being more complicated, and some really will turn out to essentially grow forever. But just as undoubtedly some of the trends that have powered the modern world over the last few hundred years are S-curves, and they will plateau if they haven’t already. How we will deal with these plateaus? These changes in direction? Will the process be smooth and uneventful or catastrophic? For a long time we’ve essentially been able to innovate our way out of the problems we’ve created, but we’re coming to a time when we’ll no longer be able to count on that.
I know that at least some of my readers love nothing more than proving me wrong. Well if you were to look at donations, they also resemble an S-curve. This is a chance to prove me wrong, make it grow exponentially!
1. As I pointed out previously, you’ll see an S-curve in any finite system. Most of the previous predictions of peak oil were sub-predictions about peak oil production (usually from some previously-known source). Fundamentally, all fossil fuels are high-density energy stored in the form of carbon bonds underground. Once those stores are gone, you don’t get them back except through the same time-consuming process that put them there. We just don’t really know how large those stores are, and therefore how long we have in exponential growth before we hit the inflection of the S-curve.
2. What do you imagine would cause decreased appetite for total energy expenditure? I could see competition among different energy sources decreasing the palatability of certain sources, but I don’t see a situation where global energy demand decreases. I could be wrong, but I’m curious how.
3. Every source of energy production is going to produce some negative externality, whether it’s whales going extinct, pollution, nuclear waste, or just using finite energy stores (like the finite mass of uranium on the planet). I suspect solar/wind/etc. will have negative externalities the next generation will struggle against.
4. I think regulation tends to follow innovation, not lead it. Thus, drones with video on them were sold before low airspace rules were clearly delineated. The same with Uber, AirBNB, and a dozen other ‘tech’ startups that are really about changing the way we do things using technology. Perhaps large-scale, expensive projects (like fusion or self-driving cars) secure favorable regulatory environments prior to production.
2- I don’t think we are going to decrease our appetite for energy expenditure. But I think we may end up with a decreased supply of certain energy. I guess an interesting question is whether people’s appetite for energy will overcome their distaste for nuclear should it become the only option.
3- I agree though people are not particularly rational about evaluating the true cost of various externalities. Many people worry about the CO2 from coal, but very few people give any weight to the people killed by air pollution. Nuclear is amazingly safe but it’s externality sticks around for a really long time. As far as the externalities of wind and solar, I think the most likely thing is that people are going to start getting sick of seeing windmills and solar panels every where they look because that’s about what it’s going to take, because of the very low energy density.
I’m not sure why our energy consumption wouldn’t be subject to an s-curve. We only have 5 senses and we can only accept so much ‘input’. I suspect we are not really capable of infinite energy consumption per person.
Yes good point. Although we technically do have more than five senses, it’s objectively true that energy consumption is a finite resource and therefore we expect to see an S-curve somewhere down the line. My point was I don’t expect that to happen anytime soon. Flying cars and trips to Mars are demands for accelerating energy consumption. Whether we can provide that is another matter, but I don’t see us reaching an inflection in energy demand soon.
I am going to have to strongly disagree here, particularly if you’re saying that voluntary demand for energy is going to plateau. I have seen no evidence for that. First if we’re talking about energy use to augment our senses. As lighting got cheaper people just used far more of it:
https://www.npr.org/sections/money/2014/04/25/306862378/episode-534-the-history-of-light
Second outside of our senses there’s all manner of technology that is incredibly energy intensive:
https://www.technologyreview.com/s/613630/training-a-single-ai-model-can-emit-as-much-carbon-as-five-cars-in-their-lifetimes/
https://arstechnica.com/tech-policy/2018/05/new-study-quantifies-bitcoins-ludicrous-energy-consumption/
I can see why you might think that demand would plateau, but there’s no evidence to support that hypothesis.
When was the last time you added more light bulbs to your home? When did you last say I want this room or outdoor area to have even more lightening? Probably has been a while and if tomorrow you won a Home Depot shopping spree, add more light to your home is probably not going to happen. We have probably already hit peak lighting. If you wanted the lights on in every room in your home 24 hours a day, you can do that. Do you want too? Probably not. Even places that are lighted up 24 hours a day (24hr snack store), do we want more light added there? Probably not.
Can you stimulate your taste buds with sugar, salt, and other flavors all the time? Sure, but you probably don’t want too and we don’t have the capacity to give your taste buds more than that. Flying in circles in a flying car and hoping aboard ships to take you to different planets? That would consume a lot of energy but the online VR Starwars game I’m sure they will try to release in a decade or so will take a lot less energy and will probably get quite a few people to spend *less* time in whatever passes for the flying car of 2030.
If we achieve Star Trek Holodeck, how much more energy per capita would we need? At least for those who are content to stay on earth and not demand lots of long term space trips done in a short time.
Recently built a new home and noticed how many more lights they use in new construction. I remember when I was young we had one fixture per room with maybe 2-5 lights. From where I’m sitting now I count 25+ lights. That’s probably as many as the house I grew up in had total.
Meanwhile I added under the cabinet lighting (activated by a touch sensitive sensor along the cabinet. Cool system I designed myself.)
Some people add lighting inside their drawers and cabinets that comes on when one is opened. Honestly, I feel like lighting is a very bad example to use if your case is “we’re going to see its demand plateau”, since as of now its growth is still strong. Not only are we adding more lights than ever, we’re inventing ever more ways to add them.
My neighbor just bought one of those solar powered motion activated lights for the side of his house he never uses because … he never uses it. If the light goes on he’ll want to know why. New houses now have exterior ‘display’ lights, not meant to light up your front porch, but just to show off the house better. Why? Because when it comes to whether we can see something with our eyes or not we’ll buy another light to see it/show it off.
I suspect you are still on the flat part of the S-curve for lighting, even if you’re not, you admit there is one. While you could theoretically pack an infinite number of photons in a finite space, the human eye can only process so much light. Suppose your entire interior is lighted up like an operating room and your exterior is roughly like a football stadium for a night game. Will you still want more lights?
I already agreed with the idea of an S-curve for demand for lighting. It’s a finite system, so all exponential growth will be limited to an S-curve at some point. All I’m saying is that the specific example of lighting is still in exponential growth phase, as I see it. Partly due to cheaper, more long-lasting lighting technologies. Partly due to better batteries.
Also, note that not all the lights are on all the time. Sometimes you want light under the cabinet. Sometimes you don’t. I suspect much future growth will also come in the form of IR augmentation of photography. Maybe you don’t think of that as lighting, but the lighting people do. We’ve barely begun to scratch the spectrum and the possibilities.
And I think this is subtle complication hidden under the surface. All finite systems will eventually become S-curves, but you can be in exponential growth phase for a long time given the scales involved (astronomical, global, technological). Thus, one person might cry doom because of runaway growth and be wrong, because of course the system will be constrained by S-curve dynamics; it’s finite! Meanwhile, another person might cry doom because of S-curve growth and be wrong long enough to be irrelevant, because the inflection point is so far into the future we’ll long since have innovated around the problem.
“I am going to have to strongly disagree here, particularly if you’re saying that voluntary demand for energy is going to plateau. I have seen no evidence for that.”
I think some pretty powerful evidence is the makeup of products versus services in the economy. Here’s India as an example (https://commons.wikimedia.org/wiki/File:1951_to_2013_Trend_Chart_of_Sector_Share_of_Total_GDP_for_each_year,_India.png)
In other words, there’s more money writing video games than there is selling a video game system. Given that is the case, it seems pretty hard to see how energy consumption per person isn’t on an S curve. To make a video game system you have to move around atoms, which is energy intensive. To write a video game you just have to feed calories to people with brains. Services are also cumulative. Today a programmer is writing a much more sophisticated program than someone writing for the Commodore 64 35 or so years ago. Yet he doesn’t eat more food than his counterpart from the past.
AI’s don’t eat food so they may require more energy but it only makes sense to deploy an AI if the cost of providing it energy is less than the cost of feeding people to do the job instead. To put a new upward bend in the energy S curve, then, AI would need to meet the power of a human brain AND use as little or less energy as a human brain does. Huge leap there.
This goes back to my previous assertion that consumption is limited by our senses. We cannot demand infinite food because we cannot eat it. There’s no point demanding more video gaming than 24-hours a day worth, even less.
Our ability to spend is less limited but remember not to confuse spending with actual consumption. Humans bid against each other for rivalrous goods. Billionaires may compete with each other to own a Picasso, but just because spending goes into the hundreds of millions there doesn’t change the fact that all the energy used to produce it was spent decades ago and if anything such rivalry probably cuts energy consumption since overpaying for a status lowers your ability to do real consumption.
I don’t think transportation is going to save the S curve of energy consumption anytime soon. Transportation as a share of production I suspect has also fallen to services. What you’re really arguing for is humanity changing course and pulling away from the idea world and back towards the world of physicality. Where’s the evidence of that?
“What you’re really arguing for”
I call strawman.
What I have argued for is that the acceleration is due to NEW growth. I think our host has made a similar argument. Without new growth, you’d be right. An AI would do the job of a hundred thousand (or more) workers for the same price. But the AI we have today does work nobody would think to hire 10 workers for 10 years ago. That’s because of new growth. And in that sense new growth isn’t the zero-sum game you laid out where there’s a fixed amount of work out there waiting to be done.
An example: TIL a machine learning algorithm picks the cover art that’s displayed on Netflix for each option I see. If I’ve been watching a bunch of comedy, it shows me a picture of Robin Williams for Good Will Hunting; if I’ve been watching more romance, it’ll show Matt Damon and Minnie Driver smiling at each other. In the past, nobody would pay someone to do that kind of mundane work because the benefits are so minuscule. Today, AI is taking jobs nobody thought to invent ten years ago.
And this should be intuitive, given we know two centuries ago nearly everyone worked on farms, and nobody worked as a crane operator unloading container ships. We invent new ways to consume, not just better ways to consume what we have. Yes, transportation is finite, but we travel faster and farther than ever before. Why assume in the future we won’t want to travel faster and farther (off Earth, under Earth)? Read any old-time novel from the early 1900’s and they’ll make the same argument about horses versus motor-carriages. How far could a person want to go in a day? And why, when everything you really need is in town?
Again, I’m not arguing that an S-curve doesn’t exist for a finite system. I’m arguing that we don’t know where we are on that curve, and it’s easily possible we haven’t scratched the surface of what our senses can consume. Or not consume, as in the case of food where we often find it easier to just throw it away. Yet still we pay for it, and continue to demand more.
This brings me to an abstract question: Is consumer demand a finite resource? Before you say yes, consider what demands a consumer in 1850’s Cincinnati might have that are NOT met by what the average worker could earn. There are probably a few, but those could either be exchanged for other demands that worker didn’t have in 1850 (save money on a cell phone, TV, refrigerator, washer/dryer, AC, and fancy car; they’ll need transportation, but a jalopy will easily beat out their original expectations anyway so long as it has a windshield and goes 30 mph). Maybe they don’t want to work and live in the lap of luxury their whole life, which we still can’t do, but we’re not far from what someone in 1850 would consider ‘lap of luxury’ that can be achieved on very little work today. The point is we have exponentially more demands as consumers today – and not unreasonably so.
Perhaps there’s something in human social psychology that is never content with the status quo. Maybe this is adaptive, where we’re always looking to maximize our resources, no matter how many we have. Thus, even if we’re surrounded by an abundance of food and plenty of transportation options and lots of light and heat and energy, we’ll find ways to waste what we have so we can continue to demand more. If I can write a computer program that is instructed to “make paperclip” forever and ever amen, why can’t humans have a built-in program to “demand more stuff” forever and ever amen, with no ‘off switch’ telling us we have enough? In that case, we’re not dealing with a finite system, but an infinite one and should not expect an S-curve for consumer demand.
OK but how does that increase energy per capita? You can only watch, say, 5 or 6 movies in a day. If the advance is a better suggestion system run by AI, well that isn’t going to take as much energy as the creation of twice as many movies. Even the production of twice as many movies seems to follow an energy s curve. Since movies are just lights on a rectangle, a kid with a computer today can toss up on youtube what a studio in the 80’s would have needed a million or so to do.
I’m not sure I’d call antibiotics an S-curve. It’s more a new technology that requires research to maintain its effectiveness. Thus, you start with penicillin, then you move on to penicillinase-resistant versions, etc. It’s not so much an S-curve as it is, “okay this doesn’t work anymore, let’s move on to the next one”. Sure, it’s a bit of an arms race, and we’ve been ahead in that regard since penicillin was first discovered, but there’s no guarantee we’ll stay ahead forever. We’ll likely see a future trading off of leads in this space between ourselves and pathogens.
Same with the Space Race in the 60’s. I don’t see this as an S-curve, especially since we went up there an didn’t go back. If this were an S-curve, I’d expect it to taper off, not to see us actually lose ground we once held. Therefore, this can be seen as more of a publicity stunt that had many ancillary benefits. The space S-curve didn’t really begin to grow until SpaceX started getting stuff into orbit on the cheap. Once we can get up there, even if progress tapers off, we’re still on the moon and going back over and over again (from an S-curve perspective; if you want to argue we might lose innovation once discovered, that’s another post). There are lots of resources and reasons to keep pushing out into space. And although you’re right the challenges are legion, the biggest challenge is the immense gravity/atmosphere of Earth, which we’ve passed on the S-curve at this point.
I’ll defer to your greater expertise in terms of antibiotics, but if you drew a graphs of deaths preventable by antibiotics would it not look like a reverse S-curve?
You may have a point with the space race, insofar as going to the moon was an outlier, but you could imagine that there was a s-curve which peaked at basically satellites (maybe the plateau is geosynchronous satellites) and we’ve been stuck at that plateau for awhile. And now there’s a new S-curve starting with private companies.
On space, I’d say that’s about right. We were sending up satellites at a slow clip for a while before SpaceX started the price war to drive down payload prices for getting cargo into orbit. Once we broke through that log-jam costs started dropping.
As to antibiotics, I’d say the ‘deaths preventable by antibiotics’ really isn’t an S-curve as you’re imagining it. Once penicillin is invented, the deaths preventable is only less than 100% (of non-resistant infections) because of availability of health care resources. Spread of resistant strains might be considered similar to an S-curve, depending on how much you use each antibiotic. Still, I really think trying to squish the S-curve concept here brings less understanding of the subject, not more. It’s not a good description of the dynamic. Unless you want to talk about distribution of health care resources, at which point you have one S-curve through the developed world, and another in the developing world.
I think the analogy breaks down a bit because if you don’t use an antibiotic for a while it starts to become effective again. Going back to vacuum tubes, on the other hand, will not make for faster computers even if we max out silicone.
Also the ‘arms race’, I suspect, is somewhat limited. There are only so many tricks bacteria can pull out and with HIV what ended up happening is we backed it into a corner with a cocktail….the drug couldn’t evolve immunity to one drug without leaving itself open to the others. Once the strategy was figured out, the process of refining the cocktail brings us into a cure.