The Next Decades: Forces Shaped by Machine Learning

There have been many articles on this, but I’ve only come across few that grasp the economic repercussions of machine learning. Here I hope to offer what I know of the subject, and if you find it instructive, I hope you share it; the goal is for this information to reach someone who can use it.

A while back, I found myself wondering about how many total labor hours is required to sustain Earth’s population (if you have this number, please send). Every person on the planet has a set number of hours of work per unit of time alive to sustain their needs. If we were to set the unit of time to one month, that person would require meals, a roof, electricity, water, transportation, etc. That time would be the cumulative hours used to perform basic maintenance on the infrastructure that provide our standard of living for that month, divided by the number of beneficiaries of its output. Tools, such as machinery, reduce that number by some factor (a kind of inverse proportionality). One after another, our inventions became more sophisticated and that reduction factor has gone up, as did our quality of life. Now, we are at the cusp of a technology that will render the greatest reduction in that labor number yet. How can this be bad?

The potential danger lies in the possibility that the reduction factor being so large that it will unhinge our supply-and-demand laws, which underpin our current system of economics — theoretical and applied (which is about everything the treasury knows). In depressed times, the federal reserve can lower interest rates by injecting money into the economy (the details of which are spared in order to keep the scope narrow); the banks too lower their lending rates. As businesses become more liquid, they begin making capital investments, which results in people getting hired again and the beginnings of economic recovery.

The threshold that we could be crossing in the not-too-distant future is when machines can perform tasks as well as if not better than humans for a sizeable portion of jobs at a fraction of the cost. To be clear, I’m not talking about walking androids, as they’ll be more costly than what most small business can afford (not to mention overkill) but small electronic devices that can run image-recognition, natural-language-processing in the cloud. We already have the hardware technology and manufacturing capability, all that’s missing is the software. Again I won’t get into the details here but to say that companies like DeepMind, OpenAI, IBM and a few others are making advancements at a breakneck pace. The biggest hurdle has been crossed: the direct alignment of R&D with profit, the hardware to simulate parallelism at sufficient speeds, and data availability. It’s almost safe to say that AI is a solved problem by this point, and all that’s missing is optimization before the graph can goes vertical (research strictly speaking, not adoption).

If this happens, the number of hires per dollar injected will decline proportionally to the progress in ML research. In other words, businesses will be less likely to hire people, weakening the government’s ability to counteract recessions by putting people back in the work force.

What makes this scenario unique is unlike previous recessions, demand picked up after enough time has passed, in this one the displacement of human labor in favor of machine labor will take hold as the new equilibrium. It follows that there is a nontrivial threat to a significant percentage of blue-collar jobs, and in turn a threat to the general welfare of the middle class. Why this is bad should be fairly obvious.

While the time scale for this is on the order of decades, our best recourse is thorny at best. The remedy to which I allude is basic-income, a concept not too distant from its communist cousin — socialism. And herein lies the problem: The hallmark of recent Western civilization, our identity, is capitalistic. Any preemptive implementation of anything akin to basic-income will need to pass through Congress where it could easily be filibustered on purely political grounds.

Perhaps worse is viability. Finding the money to pay for population-wide basic-income will require deliberating our current habit of leveraging against the future, which itself is dependent on income of future generations. When basic-income is added to the mix, the cycle becomes an infinite recursion. Whatever changes made domestically will also likely extend to international financial systems for compatibility. (I’m nearly at the knowledge of this so I will stop here.) Historically, nothing of this scale has ever been attempted, and it’s reasonable to suppose that there exists few, if any, models designed for these situations.

The nearest bridge in financing basic-income is a refactor of our method of taxation. Part of the reason the tech sector has done as well as it has is due to an externality: job displacement. Automation renders job-categories obsolete. Recovery has been more or less at parity with rates of obsolesence, but as pace picks up, it would be unwise to leave this unaccounted. One thought is to add an additional variable in the calculation of sales tax. I’ll leave the details for someone more qualified than myself, but the gist is that some constant k, let’s call it the “job-displacement constant” is applied to the sales tax as a multiplier to offset the effects of jobs expected to be displaced in that sector sector or sub-sector. There may be nuances, such as it being only applied after certain thresholds, but the basic idea is there. Revenue from this tax would then go on to fund for basic-income.

There are a few uncertainties with this approach. For one it assumes that the periodic sale of new inventions will be enough to support basic income for the whole population; the other is that it leaves maintenance costs unaccounted; yet another is offshoring, which requires nearly all countries to sooner or later apply this method of taxation multilaterally. There are likely others of which I am unaware at present.

One drawback to this proposal is the negative effects it will have on the pace of innovation. While the argument may be valid, I tend to think it’s not as bad to slow down as it’s often made out in the media, especially amidst all this flux. Regulatory agencies have not kept up, and this tax may induce a much needed breaking mechanism to prevent us from fish-tailing ourselves into trouble.

Despite whatever faults the proposal may bear, at its fount is a principle, I believe, should be more or less invariant: If basic-income is indeed the our best option, financing it should come out of the job displacement externality.

I am nearly finished.

There has been a concerning tone among certain segments of the Valley who believe that because something is decades away, immediate attention is unwarranted. What they seem to fail to account is magnitude and nature of the change. Because the fix falls in the realm of legislation and politics, change occurs on vastly different time scales than what those in the tech sector are used to, especially one of this proportion. This fact is a result of dissimilar incentive systems that is unlikely to change. And while those in the Valley may be right, the consequences of them being wrong could be catastrophic.

Notes: This topic bigger than any one person, and the premise more ambitious than what I can give credence to. Due to time constraints, I wasn’t able to pull together citations or supporting figures; ideas for this have come from observing developments from afar.

Feel free to put down your thoughts, including criticisms. If valid points are made, I will add them to article.

If you found it useful, please share it. My hope is that either my understanding is pruned by criticism, or that it can be helpful to someone out there in chartering the next steps.

(Sorry for the disorganization of paragraphs; had to type the rest out on my phone)