When a robot or AI salesperson visits the engineering or IT department with the latest gizmo, the potential customer has a form that they need to fill out and submit to the Financial Analysis department for approval. The purpose of the form is to provide the Analyst with the information that he or she needs to determine the financial impact of the purchase and implementation. The form usually quantifies expense savings, generally in labor, against the costs of the equipment or program. If the purchase is found to increase shareholder return, it will be approved. If not, it will be rejected.
So that is the first important point. Automation lowers costs and increases profits or it isn't done. What this almost invariably means as the first step is that someone loses their job and the shareholders earn more as a consequence. So far, not good. In fact, it plays right into the 1% narrative. It is a transitional phase and the situation changes in a second step.
However, before we get to step two, we are going to pause and do a little Economics. An economy has X hours of human input that results in Y output. So, Y/X = productivity or the amount of production that can be done with a unit of labor input. There is a labor constraint on the economy and through all of history, it has been so. Output is limited to Y/X times the number of hours that are available for productive activity. The number of hours can only be affected (given constant population) by changing the number of hours per worker or by changing the percent of the population that works. Both of these are severely limited in their range.
For example, if there are 120 million people who are willing to work full time and productivity is $200,000 per worker, Gross Domestic Product is limited to $24 trillion. However, economies do not function perfectly and there are always unutilized hours (unemployed or underemployed people). However, there are two things that are clear. One, the only way to sustain growth in standards of living is to increase productivity (X/Y) and that involves automation. Two, as long as X/Y is not infinite, full employment is possible; it simply requires an increase in GDP.
In other words, suppose X/Y doubled to $400,000. If GDP remained $24 trillion, 60 million people would lose their jobs. However, if GDP grew to $48 trillion, there would still be 120 million jobs and, essentially, the standard of living would double. This is always the case. If X/Y = $1,000,000, without economic growth, there would only be 24 million jobs (80% unemployment). However, if the GDP grew to $120 trillion, again, there would be 120 million jobs.
This is the explanation for why the Luddite fallacy is a fallacy, today, as it was in the 19th Century. However, there is a potential fly in this particular ointment. As we see, today we live in a labor constrained economy. It is true that there is unemployment and there are people who would work if they could get a job. However, if we completely solved those two problems and everyone who wanted a job could get a full time job, the economy would still be labor constrained at about 112% of today's GDP.
There is the possibility that with a sufficiently high X/Y the economy would become either resource or demand constrained before it became labor constrained. In other words, there would be some permanent unemployment as is feared. While this may be an issue at some time in the future, it is not in the near term.
First, the economy is becoming progressively less resource dominated which argues against a resource constraint. In other words, as the economy grows, a smaller percentage of GDP is comprised of energy, steel, cement, etc. and more is comprised of design, knowledge, content, etc. For example, a Rolls Royce costs ten times the cost of a Toyota. However, it does not use ten times more steel. Its design component comprises an enormous amount of the price difference. As we become more affluent, the economy will continue to become more design and content laden.
Second, while small parts of the economy are already demand constrained, we are a long way from the majority of the economy being constrained in this way. For example, lowering the cost of flash memory will not greatly increase demand. In other words, very few people want more flash memory but don't buy it because it is too expensive. Flash memory is essentially demand constrained. However, housing, automobiles, services, the primary components of end product consumption are far from approaching demand constraints. In other words, there is some point at which people will say, 'Enough house', but we are not even close to it.
Consequently, we can safely conclude that incomes can increase without meaningful constraints, but that it will go to the owners of enterprises, not to the employees. That is not an untrue statement and is a component of why workers' wages have stagnated over the past two decades. However, this is not the end of the process.
To this point, companies have fewer employees and higher profits. However, to the extent that free enterprise works, the lower costs will translate into lower prices as companies either attempt to increase their market share or defend their current market share. This price competition is a major reason why predominantly unregulated economies have historically outperformed centrally planned and heavily regulated economies.
Companies will continue to drop their prices until the incremental decrease in price does not pay for itself through higher market share or economies of scale. Lower prices, of course, benefit all people who buy products and services. So, on the surface, lower prices look like a good thing. However, to Economists, comprehensively lower prices, called deflation, is a bad thing and they will take measures to 'fix it'.
Because the Central Banks are full of Economists who believe that the best state of affairs is inflation in the 2% to 3% range and that deflation is bad, when automation causes deflationary pressure, they will make it go away by printing more money. Historically, this has usually been done by lowering interest rates. However, as deflationary pressures have continued, interest rates have fallen to close to zero and they can't use that method anymore. So, they have started buying debt, usually either government or mortgage debt.
Here is the really cool part. Usually, as is the case in the U.S., the Central Bank is granted a modest return and, beyond that, all the proceeds from printing money goes back to the government. Lately, that amount has been around $80 billion per year in the U.S. There is essentially no down side to this. It is a process by which a portion of the increase in GDP ends up in the Treasury without anyone being taxed.
Something similar happens when the Central Bank buys mortgage bonds, but rather than the government benefiting, home owners benefit through lower mortgage rates. The U.S. Federal Reserve did this and right now mortgage rates are historically quite low. In all likelihood, they will return to Treasury Bonds in a QE-IV.
Here is the counterintuitive part. The Central Bank can never sell the bonds and when they mature, they must replace them. They bought them to increase money supply and eliminate deflation. Selling them or allowing them to mature will undo that. It, then, would be deflationary because it would take money out of circulation and would need to be fixed just like the original purchase.
What that means is that when the Central Bank buys government bonds, the interest paid becomes income to the Central Bank and from there it goes back to the government for a net zero interest expense. The Central Bank can never accept payment on the principle or it will undo the deflation fix. So, when the Central Bank buys government bonds, they functionally disappear. They never have to pay back the principle and the interest they pay is refunded to them.
When the government spends tax dollars, it does not stimulate the Economy. The government spends it instead of the taxpayer, but there is no increase in total spending. But when the government spends the money that it receives by selling bonds to the Central Bank, the taxpayer keeps their money and is free to spend it AND the government spends the money without ever having to pay it back. This is very stimulative and GDP grows.
So, in the final step, the people who lost their jobs at the beginning of the process get jobs back because X/Y and GDP have increased such that, in the end, jobs remain constant. However, the process takes time. In a way, that is good, because the people who lost their jobs lost them to robots and AI and that loss is permanent. They need to retrain for jobs that have not been automated. And, of course, two or three years of unemployment while being retrained is a severe hardship for mid career workers.
Many Economists, accustomed to productivity gains that fall comfortably within traditional economic growth rates have difficulty imagining substantial technological unemployment. Yet, it is and always has been the case that economic growth and technological unemployment balance each other. In the long term, through the mechanisms described above, that is correct. However, if technological unemployment is very high, there is a lag and the net unemployment can, temporarily, get very high.
Suppose that structural unemployment is 4% and over ten years GDP is going to increase 10X. Productivity must grow 10X which is the equivalent of 90% of jobs being replaced by robots and AI. Further, let's suppose the process described above takes 3 years, on average, from the loss of jobs to automation to the eventual re-employment. If the Transformation takes 18 years, unemployment will need to be 90%/6+4%= 19% during that time. If it takes 24 years, unemployment will average 90%/8+4%=15.25%. On the other hand, if it takes 12 years, unemployment will average 90%/4+4%=26.5%.
If we slow down the Transformation, the unemployment peak will be less. However, the total pain will not be less. The total number of unemployment years is the same. The benefit will be in an increased ability of the economy to ameliorate the pain since it will be less in any given year. The problem is that we have mechanisms that allow us to delay the onset of the Transformation and, not surprisingly, it appears that government and industry are availing themselves of them, but we don't really have good mechanisms for delaying full implementation.
For example, some governments have asked Google to delay the introduction of their driverless vehicle technology and Google seems to be agreeable. On the other hand, once they introduce it, it is not clear how we slow full implementation. In other words, our behavior seems to be shortening the Transformation and, in the process, increasing the average and peak unemployment rate.
If we can shorten the cycle, say, from three years to two years, we then can make some significant strides to lessening the trauma of the Transformation. For example, in the 18 year example, rather than an average unemployment of 19%, it will average 90%/9 + 4% = 14%. We also can lessen the pain by using our 'deflation fix windfall' to increase unemployment benefits to two years and to fund retraining expenses.
We also see on the horizon some spectacularly disruptive individual events that are imbedded in the 90% job loss. What happens if SAP and Oracle get together and eliminate the tasks currently done by AR, AP, Payroll clerks and staff Accountants? This is within current technological capabilities and primarily require the various accounting software packages to be able to talk to each other directly. It will tend to happen all at once, because, for the most part, only two companies need to upgrade their software and it is over. What happens when three million clerks and accountants hit the unemployment line as almost the same time?
It appears that once over the road drivers start getting replaced by robot drivers, competition will drive this to be a one year event. A typical 18 wheeler may cost $0.75 per mile to operate of which about $0.50 is for the driver. If a robotic driver cuts the driver cost in half, suddenly there will be trucks that are charging $0.50 per mile. The unconverted trucks have a choice of losing money or converting. If it is a profitable implementation for one company, it will be profitable for all and the conversion of the national fleet will happen very rapidly.
Also, if the promise of drastrically reduced accident rates holds up, it is likely that the companies that insure the 18 wheelers will demand that the change out takes place as quickly as possible.
An ameliorating factor is that the faster the transformation the faster will be the deflation which leads to larger amounts of government funds that can be used to increase unemployment benefits and retraining grants. This is a theme that will come to dominate the political dialogue over the next 10 to 15 years.
So, to summarize
- Companies buy productivity enhancing robotics and AI. They lay off workers and profits increase.
- Companies 'spend' their increased profits to gain or defend market share. Prices fall.
- Falling prices, when nearly universal, creates economically destructive deflation.
- Central Banks respond by creating money by buying government bonds.
- 'Deficit spending' findanced by debt that carries no net interest expense and doesn't need to be paid back is economically stimulative and creates jobs. Unemployment decreases, X/Y is higher and standards of living are increasd.
It is unlikely that governments will respond in an enlightened way. However, the more we talk about it, the more we 'spread the Transformation meme', the more likely it will be. It would be wonderful if this article went viral. However, because it contains no kitties, I'm pretty sure it won't.