Computers, People and Sofas
What will the world look like once computers go superintelligent (SI)? That is: if we assume that in the not-so-distant future, computers will think and act much better than humans, and do so autonomously, and in self-improving fashion, what then?
What will happen to all the places, products, and services around us, the companies and institutions that provide them, and the workers and customers that make it all happen?
A simplified view of these questions could be restated this way. Once we get SI:
- What would things be worth?
- What will each person do/work on?
- What should we do, collectively, given the first two?
I want to take a concrete case and see how far we can go with these questions.
For example, once we achieve SI, what would happen to the value of the sofa I am sitting on, the people who work to make this sofa, and what, if anything, should the collective do about that reality?
The Sofa Maximizer problem
SI without robots
The cost of a Sofa is approximately 30% materials, 20% factory labor, about 10% plant operating expenses, 10% logistics, 20% selling, general, and administrative expenses, brand, and Customer support, leaving 10% in profits.
Super intelligence, even without robots (SI physical instantiation), eliminates the jobs associated with selling, G&A, brand, and customer support. However, there is one significant component of this category that is not the cost of labor, but rather the fees the sofa company pays to other companies that assist them in acquiring their customers.
People will go to Amazon, find the product on Instagram, or ChatGPT will recommend it, and that intermediation will be a cost to the sofa maker that SI cannot bypass. If current numbers hold, that could be up to ½ to ¾ of sales, G&A, customer support, or approximately 10% of the top line.
From the perspective of the attention merchant, maybe their business is all SI doing most of the work (it won’t be zero cost because that SI needs servers, those servers need cooling, etc). Maybe if those businesses are all run by SI, they will be more efficient. The marginal cost of attention will drop, but because SI does not have a direct impact on network effects or economies of scale, promoting or selling the sofa on these platforms is likely to be a real cost for the sofa makers, at least for a long while.
Where else can SI change the cost structure and value of the sofa? We can imagine SI organizing resources at the plant and logistics to be more efficient; however, hiring and training humans still need to be more efficient, placing people in the right jobs at the correct times. SI could optimize energy consumption second by second and improve logistics in similar ways.
How much better can this be? Current systems and workflows are likely smart enough, so I don’t think there is a massive improvement to the currently software-controlled, pre-SI systems, so maybe a 10% reduction on plant opex and logistics? A total of 15% reduction in cost seems reasonable.
This last point is interesting: SI will be able to do things that no machine can do now, and do it better than humans. However, for many tasks, machines already perform very well, and the gain might actually be tiny. Imagine an SI that figures out how to build a printer that is 100X faster than current home printers. Unless it is sold at the same price, the sofa company probably won’t care too much about that gain in that part of their business.
SI with robots
So far, we are talking about non-instantiated SI. Now, let’s get the robots in here. An SI that had superhuman dexterity and the ability to act in the world should move the needle quite a bit. Let’s do another back-of-the-bowling estimation on the impact:
Factory labor will drop to practically zero. Logistics will drop a ton with autonomous movement, but it would not fall to zero because fuel is a significant cost. Let’s assume it halves. These two savings would represent another ~25% drop in price, for a total accumulated from SI of 40%. Let’s assume we are undercounting, and round off to 50%.
But we can’t forget that embodied SI will also introduce new costs. For one, a factory full of robots will consume significantly more energy, so that plant overhead will increase. And then, there is the cost of robots: real SI would have robots design and build themselves, so there will also be a constant drop in costs, but, like the sofa, there will be material, energy, a marketplace for attention on those products, and margins that will make the cost not zero. So, that’s part of the cost of the Sofa. How much? It’s hard to tell, but I feel comfortable assuming 10% of the total cost. So, let’s go back and call it a 30-40% total cost reduction thanks to full-blown SI.
30%?!?
My first reaction is that it feels tiny. I was expecting SI and its robotic extensions to slash costs by 90%.
Looking at our simple model of costs and margins, it becomes apparent that materials and energy are the two big, stubborn categories that are keeping SI from conforming to what I anticipated. Can SI dramatically change material costs and energy? I would imagine the answer is yes.
We can imagine a world where SI conceives of an atomic printer or some biolab process that can make wood, steel, and other materials much cheaper than growing or mining them. Maybe it is something more trivial but overlooked, or some currently unimaginable solution.
But these solutions will require a ton of energy. And as we mentioned, the robots and the computers will need tons of energy too. So, perhaps the SI finds a way to generate clean fusion energy at a low cost, or harness energy from the sun in non-obvious, highly scalable, and persistent ways, or maybe it fast-tracks a Dyson Sphere or some other galaxy-scale energy project.
Let’s call this the New World Scenario. And let’s label the world where things somewhat stay as they are, but now jobs are done by computers and their robots, as the Better World Scenario.
The Better World Scenario
~30% seems smaller than anticipated, but if achieved within a short timeframe, then it’s very significant from an economic standpoint, and most certainly for the job prospects of all the workers who lost employment due to automation. In turn, it is worth spending some mental cycles trying to figure out what it would mean for our key guiding questions of how much things would be worth in this world, what it would do to people’s jobs, and what our collective reaction would be.
A shortcut to help think about the Better World scenario we described above, where we see a rapid ~30% drop in the cost of making a sofa, is to think about a similar disruption to sofa production from the recent past: China.
It turns out that the rise of China as the manufacturing center of the world had a similar impact. Specifically for sofas, apparently, it drove down prices by about 30%, and it did it fast.
So, what happened when China started making the world’s sofas? Many things could have happened: margins could have expanded, the dollar value of the market could have contracted as people buy the same amount but pay less, suppliers could have mushroomed as the startup costs for the Sofa business dropped, and so forth. This is what actually happened:
First, people began buying more luxurious sofas and replacing them more frequently. Supply grew, but the most prominent players grew increasingly concentrated/consolidated to get economies of scale.
In the U.S, furniture-related jobs dropped about 30%, with those jobs mostly moving into logistics and ecommerce, where workers mostly found lower wages and lower job satisfaction.
This seems to be a robust pattern that has repeated itself throughout modern history. As things get cheaper, the consumption budget items do not seem to drop. Instead, people move up the scale (I wrote about a similar dynamic with US house prices here).
For people in China, the manufacturing boom created tons of good jobs, which triggered a massive migration from provinces to cities, and, as we saw,
SI is likely to have a comparable impact to China’s industrialization on the production and consumption patterns of sofas, characterized by increased variety, luxury, and more frequent upgrades at lower prices.
The jobs, however, will have different: China’s rise directly impacted jobs in Western countries. But SI will impact jobs everywhere: The sofa companies likely have sales/GA/brand in the US, customer support in Mexico or India, and manufacturing in China. All those jobs are likely gone with SI.
In countries with large domestic markets like the USA, SI might actually make it easier to reshore all of the supply chain, but this will have a marginal impact at best in employment, as a lot of the jobs the sofa company generates in the US could still be lost due to SI.
Was China really a robot?
Here is an interesting twist:. There is a big study that claims 87% of the job losses in manufacturing from 200 to 2010 were due to automation, and only 13% due to trade and globalization. This is consistent with the growth in output per worker, which began even before China reached its stride: from 1996 to 2016, U.S. manufacturing output grew by 71%, while manufacturing employment dropped by 30%. Some economists over at MIT did the math and concluded that manufacturing robots replace somewhere between 3 and 6 workers.
China, robots, or both, the effect seems comparable. The policy implications, however, might be dramatically different.
The value of a sofa (and everything in it) in a Better World
Let’s get back to the initial questions.
First: what would sofas be worth in this SI, Better World scenario?
With what we have learned so far, it is safe to assume sofas will be cheaper, there will be more and better options with sofas, and it is likely people will move up in quality/value, and maybe change sofas more often
.Jobs in sofa manufacturing will mostly disappear. We also saw both of these happen when China and the prior wave of automation became involved in the sofa business.
It is less clear what jobs will be available after sofa factories replace people with SI robots. Part of the lessons from the 2000s tell us that new jobs will be created thanks to the growth of new industries on the back of SI, but last time around, these jobs were not a full replacement and were not equivalent in terms of wages. Happening here would be, in good deal, a function of policy decision.
What else?
Raw materials that go into sofas (and other products) and the energy needed to make them (including, and especially, the energy required by SI) will conservatively hold their value, or more likely increase, as demand for iforcommodities rises as production rises (more and better sofas in the market).
This has support from our analogy: China’s industrialization consumed vast amounts of commodities, including many of those included in our sof, a like steel, wood, cotton, leather, and foam.
This growth ushered in a commodities super cycle that drove a ~250% gain in Gold, 1,300% in steel, 700% in cotton, 100% in wood, 50% in foam, and other massive spikes in other commodities like oil, aluminum, copper, and others.
How about sofa companies? Looking at China, they ended up with tons more factories, and also much bigger factories. The US lost sofa manufacturers, but the companies that could command the consumer attention and purchase intent grew, so we got fewer but bigger stores.
What does this mean in an SI scenario? SI could drive a China-like commodities super cycle, and commodity traders (physical and financial) will likely have a great run.
Then we have companies building SI.
If you substitute the Chinese factories and their workers from the recent past, and replace them with the SI of the near future, the companies that provide the software and robots will grow in quantities and in size, and they will capture a good deal of the new value. This will drastically accelerate the “software eats the world” trend.
As more sofas are bought more often, the companies that can promote and sell these products to customers will also benefit from SI. From those companies, some will leverage SI sooner to run their business (in a similar fashion to who we explore SI role in sofa manufacturing), and those will win and get very big.
What about the people
What would happen to jobs?
Let’s start with the now: we have the impression that employment levels have remained generally stable, but if you look at hours of work, these have been going down monotonically for 20 years.
This becomes far more striking if you look far enough back. In a world without work, Daniel Susskind remarks, “British agriculture produces more than five times the output that it did in 1861, but the proportion of the total UK workforce employed in it has fallen from 26.9 percent to 1.2 percent”. Where did all these workers go? They went into manufacturing. In the 70s, manufacturing employed a quarter of all employees. Today, it uses less than 10%. And where did all those workers go? They are now in the service sector.
Steady unemployment rates can be attributed to employers transitioning from one industry to another as their jobs become automated. But what explains the decline in hourly work? To some degree, this is the nature of the service sector; by some estimates 1 in 3 Americans are now working on the Gig economy - but, fundamentally, this is a function of automation: there is no new sector to jump to, so employees, in all industries, are seeing their hours of work cut down by the complementary contributions or technology.
The challenge with SI is that it is unlikely to be complementary. For many jobs, SI will do it all. Farms already run mostly on machines; manufacturing increasingly so as well. In my area, some taxis are starting to drive themselves. Most of my customer support interactions - in-apps and over the phone - are with AIs.
So, what happens to jobs in SI? One option is to assume SI is like other technologies and that the future will be like the past. If that is the case, new industries will emerge, and the challenge will be to help people from past sectors to transition to new ones (this is harder than it sounds, and we have failed to do this well in the past). The other option here is that SI is different, and that there is a permanent loss of work. The challenge here is less what will people do, and more, how will people earn a living: work used to be the primary form of wealth distribution, but, as we can see, we have fewer hours of work to go around as it is, and with SI, this number will collapse.
SI, people, and policy
SI seems inevitable. As companies, institutions, and countries race to build it, the biggest challenge and opportunities they will face will be from regulation. Regulations will help SI come into being, or will block it altogether. It will give space to exercise its abilities and powers, or it will not. And, it will likely inform, impact, and maybe even create its own regulatory environment.
Public policy will also be critical to the future of work, or whatever it is we do to earn a living, or simply freeride on it.
The power of SI cannot be modulated individual by individual or company by company. Most likely it will require the greatest effort in industrial and labor policy, fiscal and monetary policy, especially as they drive wealth redistribution and even foreign policy, as it affects the material, energy, and products (chips!) source around the world, which provide the material substrate of SI.
There is a long history of successes and failures of industrial policy: protectionism that worked well in East Asia and that worked poorly in Latin America; state-backed financing of new industries like Solar, that worked poorly in the US but fairly well in China, and so on. Like everything, there are a million ways to screw something up, and a few ways to get it right.
The general sentiment from the big tech companies is that, no matter how much they thought they needed to invest in AI in a year or two, today they believe they underinvested. Massive growth will require boatloads of capital, massive Real Estate build-ups, chip production, and vast amounts of energy (Google and Microsoft are already at ~30tWh; top 4 tech cos probably are in the 10B/yr in electricity bills). This will require zoning, permitting, trade, and energy, which the government can help enable or block. If one thing can get in SI’s way, it is the government. And, at this stage of the game, government and capital are the most powerful levers to move faster.
The COVID test-drive
The labor side of the equation is more complex. How would people live if they couldn’t work?
When people could not work due to COVID, the solution, at least in developed countries, was the government. Just in stimulus checks (and there were loans and other forms of help), the government sent about one trillion dollars in direct transfers to people during COVID.
But being unemployed is not a virus. Suppose we can trust humans to remain humans through the rise of SI. In that case, no-merit wealth transfer will not lead to a stable social equilibrium: people don’t like others to take what is theirs, and most people don’t want to feel despondent and dependent on others, and need a sense of self-worth.
Past solutions won’t work either: in the past, as people moved from farms to factories and from there to service jobs, we leveled up people’s skills, getting everyone through high school, and then moving more and more people through college. But, there will be little to no jobs where any education can enable people to compete with SI.
But really, what about the people?
My guess here is that we all become politicians, or maybe more accurately, public servants. For one, regulation will be key to SI, and humans don’t have and won’t want to advocate for democracy to SI. As topics emerge, SI will likely help us come up with optimal policies and come to reasonable compromises faster, but people’s input and decisions won’t go away. So, in a way, we will all be politicians managing SI boundaries, funds, decisions, and so on.
But, more importantly, humans, if we stay more or less the same, will always need other humans for love, care, and support. Today, many of these tasks are unpaid. But, in the future, I think, caring for others will likely be the way people earn their living; the government, however it looks with SI around, will move wealth around, not like Covid nor like welfare, but for jobs that exist today, that we know are valuable, but that don’t have the wealth, the incentives nor the mechanisms to compensate adequately.