What’s Scary About AI Automation Economics

A new wave of Automation is on its way, one driven more by machines than by humans. Uber is creating millions of jobs that it will be destroying once its self-driving cabs hit the ground, truckers will be out of a job soon, low-skilled workers are already being replaced by machines in factories and McDonalds. That’s step 1, it’s already happening. Step 2 is happening tomorrow, and this time lawyers, doctors, engineers, designers, programmers, strategic planners, will have to adapt. At least a sizable proportion of them will be replaced by more efficient algorithms. Within 10 years you’ll hear of a flying prototype of a commercial airplane without a pilot. I’m calling it a wave, but at the speed at which AI is improving, it’s more like a tsunami.

Some people are proposing this sexy idea of a Universal Basic Income as the miraculous solution to save us from AI’s job destruction.

“jobs are destroyed, industry doesn’t create replacement jobs, so let’s be pragmatic and not starve people without a reason” — a Swiss UBI Enthusiast

Their opponents claim that basic income does not just remove the incentive to work, that it actively punishes work, in the form of steep tax. But it’s not the first time in history that people panic and call for a basic income. Do we really need that?

Automation is not a choice, it’s a result of market dynamics

Automation can take many different shapes but we can define it as any process or technology improvement that reduces the number of jobs that are needed to create the same amount of value:

Profit/Employee before/after automation

This increases the company’s profit by changing its cost structure. In particular it increases one ratio: the (company profit)/(number of company employees)* ratio. The resulting competitive advantage makes it impossible for competitors not to react, which makes automation practically impossible to stop. So this is happening whether we like it or not.

If we look at the past, automation has only moved jobs from a sector to another sector:

Automation is not a new thing: it happened many times before, decreasing the number of related jobs as productivity per job increased, for instance with the mechanization of the agriculture. Just the single invention of the tractor eliminated about 10 million farm jobs in the US.

Gradually, jobs went to the manufacturing sector, making the first move in what is called the “3 sectors economic theory”: as jobs in the prominent sector get automated, more jobs are created in the “next sector”. Recently, automation happened again in the already mechanized secondary sector, and more jobs went to the tertiary sector, where most of them are today in developed countries.

AI is the technology to automate the jobs of the Tertiary sector

Computers are good at processing loads of well formatted information and making rational decisions based on numbers. Until now, we had to provide the machine with both the data and the reasoning for the decisions. AI simply takes this to the next level by allowing the machine to making decisions by itself based on patterns it has seen previously.

If you take a step back for one second, this is probably what you do in your job. Think about it, aren’t your proud years of experience in your resume just that? A set of patterns you’ve seen previously that allows you to better figure out what to do?

Unlike us, computers can instantly transfer information to one another without a loss. Connected in a network, once one learns, all the others have received the same training. Open AI built a bot capable of beating the best players in the world at Dota, one of the most strategically complex games, where brute force can’t be the answer. Instantly, millions of instances of this bot could be put online and defeat all human players online.

A machine outsmarts the best in the world without using brute force

Boston Dynamic’s Atlas has become terrifying

For those who work in tech, we all know that machines will replace humans in a majority of the jobs we know today. Basically every task that we do more than once in our lives, AI will do this much better than us.

The debate today about AI is between two kinds of people:

• The ones that believe we won’t find any new job after that.
• The ones who believe new jobs will appear anyway, and there is nothing to worry about.

I’m in between: I believe we’ll find new jobs, but we should worry greatly about the economics involved.

Till now, automation has always been beneficial and created new jobs:

And there is a very mechanical market reason why it always happened this way.

Automation of a sector at first concentrates the profits within this sector to its pioneering companies, which rewards innovation and risk and attracts competitors. As competitors spawn, market dynamics slash down prices and cut margins, making the sector’s products become commodities (which means much cheaper).

If Humans were like machines, we would merely spend only what we need to cover our survival needs, and the entire economy would collapse alongside consumer spending. But we are not machines (not really, not yet…), in fact, consumers always spend the amount of money they can afford to, and create new needs along the way if more wealth is available. Since automation phases always happened slowly enough so that not everyone was out of a job in a day, competition, market dynamics and innovation were always given enough time to make the transition to a new sector happen.

Here is the catch: Software companies warp the profit distribution mesh like a black hole warps space-time

Profit per Employee distribution among companies pre-tech-boom (2002–2004)

The unprecedented competition level powered by the internet makes software a “one winner takes all” market, which has led to the creation of the greatest monopolies in history: Google, Facebook, Amazon, Microsoft are the most well known for the Western world, but there is a bunch of them for each market.

Facebook and Alphabet, each bring in over $1 million in revenue per employee — Apple and Visa rake in nearly $2 million per employee. As of this writing, just Apple is also sitting on a famous $250 billion cash pile, and isn’t doing anything with it. These companies are taking in all the profits of their market, and maintain monopolies that deny access to a competition that would create as many jobs as the market can afford to.

AI companies are in the straight continuation and amplification of that phenomenon. No one in tech wants to hear that, because everyone in tech is happy to be on the side of the winners, the juicy profits and salaries, at least temporarily.

The AI transformation is fundamentally different in scale and speed

Technology doesn’t give a chance if you’re not ready. We’re a cavalry regiment meeting an F-22.

As many of you might not know, I’m a history nerd. I’ve read dozens of books, hundreds of wiki pages, and I’m subscribed and listened to hundreds of hours of podcasts debating what some would consider things of the past. As remote as it can seem, the AI automation wave appears to me very similar in nature to what has been one of the greatest shocks in history: the breakout of World War 1.

I’m not talking about AI starting a war, I’m talking about change, in scope and speed. Back in 1914, technology and industrial capacities had made a leap forward of several orders of magnitude since the last major wars. No one knew what this was going to look like, and no one was ready for what happened when things sparked. Nations had to deal from day one with unthinkable level of casualties by their (and today’s) standards. We’re talking about industrial death: tens of thousands on a daily basis, when in everyone’s memory tens of thousands dead was the rate of the bloodiest months of major wars before that. That was a 2 orders of magnitude increase in the number of casualties overnight, and governments had absolutely no idea what to do about it, and it went on, and on, and on.

Since the last automation wave happened, the pace of progress has been tremendous:

• The speed at which software can be produced and deployed has nothing in common with the speed at which other forms of automation were deployed before. We easily see a 3 orders of magnitude increase in both go to market speed and reachable market size. A software that removes the need for certain kinds of jobs can be deployed worldwide through the internet, from day one and at virtually no cost. BOOM: with a clear value proposition, an entire profession is out of job, worldwide.
• Unlike in the times when anyone could come and compete with a slightly improved product, AI runs on what is 21st century’s gold: data. In fact it’s common knowledge that the main driver of AI technologies is not their algorithms, but the data to train them. And for many real world problems, only a handful of the biggest companies concentrate ownership of it. It’s like if they each were Saudi Arabia in 1920.
• The amount of training required by humans to fill the newly created jobs is also orders of magnitude greater than what was needed before. We’re talking non-trivial things: engineering, creative work. All of which with an highly increased competition level between workers.
• Last, but not least, because software is so efficient at improving productivity, companies can increase their profit/employee ratio and concentrate profits like never in history.

I don’t doubt that humans will find other things to do with their time and money. We’ll invent new things. But what I’m concerned with is that we won’t be able to cope with such changes, and that unless there are preventive social programs in place, some will lose their home, some will starve.

As I said, the problem in this coming automation phase is not much in its nature, which is to increase the profit/employee ratio, but in the speed at which it will destroy jobs. No one wants to prevent automation from improving work and living conditions, but while doing so we should be clever enough to engineer a way to smooth out transitions.

Here is my stab at such a program, that would be different from the much publicized UBI (Universal Basic Income).

Companies need to have a duty towards society to create as many jobs as they can afford to

Roughly applying Pareto’s law to the profit/employee ratio per company, we can isolate the companies that have a much greater level of automation than others, the ones with extreme profit/employee ratios:

Creating an incentive to create jobs for companies that can afford to

Companies creating less jobs per profit unit than average would see part of their extra profit relative to the average of companies (that is, what remains after all costs being paid) taken away in the form of a tax. This creates an incentive for companies to hire and train more employees (even low-skilled) to reduce the extra profit and extra tax. On the other hand, the money collected through the extra tax would allow to create replacement jobs or better train people.

How this is better than UBI: The tax auto-regulates by nature and only increases along with automation

Here is how it would impact a single company’s dynamics:

• The attractiveness for companies to accumulate profit and creating huge cash piles without creating jobs decreases
• If companies have a better idea on what to do with the money than giving it away in taxes, creating additional jobs will come to them [tax percentage]% cheaper than before, as it will reduce their taxable extra profit by the same amount.
• Companies can keep attracting the best people as there is no decrease in the competitiveness of compensation (taxes are taken on the extra cash pile)
• Since only part of the extra profits get taken away, there is still an incentive to try to be smarter, to keep automating, and enough money to re-invest in innovation

And how it would impact the system as a whole:

• The tax amount that can be used in creating new jobs and re-training workers increases if too many companies are seeing extreme profit ratios, as automation is responsible for destroying jobs.
• The tax decreases when inequalities decrease. Unemployment level could keep being high because of other factors without increasing the burden on well performing companies as automation would not be responsible.
• Companies are not penalized for restoring profit by automating if they still contribute their fair share to job creations.

Thanks to James Shen, Joy Zhang, Varad Kishore, Alexandra Alimbekova and Nicolas Dinh for reading drafts of this.

→ 👋🏻 Thanks for reading. If you liked it, please ‘clap’ several times or share it so that more people can see it, and I keep writing on similar topics.

Subscribe to get my essays here, in which I try to give a helicopter view on technology related topics.

More about me here. You can also check out some of my other top articles:

• 8 Things I learned while leaving a startup to join a big firm
• How cryptocurrencies can finally bring value to markets
• Why no-one can prevent Elon Musk from dwarfing the airline industry
• Rails Migrations Tricks [Guide + Code] — Cheatsheet included
• The downside of “Doing things that don’t scale”

*This is in contrast with outsourcing, which more likely increases the (company revenue)/(number of company employees) ratio.

_____________________