I recently came across this interview with Bill Gates where he proposes a sort-of ‘robot tax’, with the basic premise being that a worker is taxed on their income – therefore, if a robot replaces that worker then the robot should be taxed as well.
This argument seems logically sound at a glance, but upon further analysis it seems like it needs a lot more thought and elaboration.
[Issue of Partial Automation]
This is actually an issue that I ran into to some extent at my own job. Lets say that in my 40-hour work week, I spend 70% of my time on Task 1, and 30% of my time on Task 2. Lets assume for the sake of simplicity that I’m an hourly worker.
Now, lets say that my company gets a device that completely automates Task 2, and now I’ve only got 70% of my former workload. Now I only need to work 28 hours a week, and I’m getting paid 30% less because I’m an hourly worker.
In this case, would the “Task 2” automation device be taxed – or since it didn’t technically replace me entirely, would it just be equivalent to a hardware upgrade? For example, if a lumberjack went from using a manual saw to using a chainsaw, their efficiency would go up – just as my efficiency went up with the introduction of the “Task 2” automation device. This leads to the next issue…
[Issue of Defining a ‘Robot’]
In Bill Gates’ interview, he uses factory workers as an example – implying that the robots will be physical entities that will be performing physical labor.
What about software though? Say that some company develops software that can automate all data entry jobs. Another company develops software (AI) that can make medical diagnosis and perform psychological evaluations. A third company develops software to automate the roles of President, the House of Representatives, and the Senate, giving true democratic power to the constituents.
In all of these cases, there are no traditional ‘robots’ taking our jobs – yet it’s clear that people are becoming obsolete as a result of the software. Should software qualify as a ‘robot’ in this case?
Or lets go back to the discussion of ‘hardware upgrade’ vs ‘robot’. Lets take the lumberjack example to the extreme and say that we start out with just a plain human. The human obviously has some difficulty taking down trees, because they were provided with no tools to perform their job. Lets look at some ‘upgrades’:
- We give the lumberjack an axe. The lumberjack is now able to produce 5X as much wood as he was able to when he had no tools at all. Most people would agree that there are no robots yet, yet there’s an important point to be made here – one lumberjack with an axe can replace five plain humans. If a company required 100 plain humans before, they could reduce their workforce to just 20 lumberjacks with axes and maintain their old efficiency. The company effectively just swapped out 80 humans in favor of 20 axes.
- We give the lumberjack a chainsaw. The lumberjack is now able to produce 25X as much wood as he was able to when he had no tools at all. Most people would agree that there are no robots yet.
- We give the lumberjack a power-suit that turns both of his arms into chainsaws and gives him super-human strength and speed. The lumberjack is now able to produce 100X as much wood as he was able to when he had no tools at all. Is there a robot at this point? Or is this still just ‘hardware upgrades’?
- We give the lumberjack a remote-controlled power-suit, otherwise identical to the one mentioned in the previous step. The difference here is that the lumberjack can now sit in the safety of his vehicle while the power-suit cuts down the trees. Note that the lumberjack still has to be active for the entirety of the day at this point in order to remotely control the robot. The efficiency multiplier is still 100X. Is there a robot at this point? Or since the lumberjack still has to control the power-suit is it just a ‘hardware upgrade’?
- We give the lumberjack a programmable power-suit, otherwise identical to the ones mentioned in the previous two steps. The difference here is that the lumberjack can program the power-suit at the beginning of the day, leave, and then return 8 hours later to gather the wood. The efficiency multiplier is still 100X per-robot, but the lumberjack could program many robots in a single day. Is there a robot at this point? Or since the lumberjack still has to program the power-suit is it just a ‘hardware upgrade’?
- We make a power-suit with artificial intelligence that can act on its own. The lumberjack just skips stones all day. The efficiency multiplier is still 100X per-robot, but the lumberjack’s company can purchase as many robots as they want to continue to increase efficiency without adding any more humans into the equation. Most people would agree that at this point, you’re dealing with robots.
My primary point with the ‘thought experiment’ above is that it’s not a trivial task to determine what the definition of a robot is. And then on top of that – it seems clear to me that the ‘issue’ that we should be focused more on is not ‘robots’ but rather just increases in production (per human).
Maybe profits vs. humans employed will be an important metric in the solution to this issue of automation / robots. If a company is putting the money back into the economy by spending lots of their revenue on new construction, R&D, or perhaps just providing it as dividends to their investors, then maybe the increased revenue gained through robots and automation aren’t so bad. The issue arises when companies make revenues and then they just sit on it. This just further widens the wealth inequality.
This idea of profits vs. humans employed would have to be expanded on a lot, however, just due to the fact that it leaves a lot of room for exploitation. For example, we wouldn’t want companies blowing all of their revenues on cocaine and hookers just to keep their ‘profits vs. humans employed’ numbers down to avoid taxes. Realistically, you’d be looking at a metric more like ‘revenues that were not spent on essential operating costs, new constructions, R&D, or dividends vs. humans employed’. That’s about as far as I’d like to expand on this particular ‘issue’, as one could continue to elaborate quite a bit on this topic. Even the metric I’ve bolded above is fairly simplistic…
Part of the appeal for ‘robots’ and automation is that it opens up the possibility for some new ‘higher-level’ jobs.
When I say ‘higher-level’ jobs, I’d use the example of programming. Early on, programmers had to write their programs primarily in assembly. Then on top of assembly, they built languages like C/C++ which allowed for more complex programs to be developed more rapidly. Then on top of C/C++ you get scripting languages like Python, which once again speeds up the development process.
As you can see, with improvements in the tools available to workers – it allows for the workers to reach much farther than they could before. My point with this ‘higher-level’ example is that robots and automation don’t necessarily result in an overall increase in unemployment.
Like perhaps instead of manually constructing rocket ships using a team (say 500 engineers), we get to the point where we can mass produce them with perfect precision using colossal 3D printers. As a result of the introduction of the 3D printers, 500 engineers were laid off at the rocket company – but maybe now there’s an increased demand for deep-space pilots / flight attendants (for the expected increase in space tourism), and there’s an increased demand for engineers to build up cities on our newly-founded Mars colonies.
Should the rocket company in the scenario above be taxed on it’s 3D printer that replaced 500 engineers? Or should derivative jobs be taken into account in some way? Seems like it’d be difficult to measure this sort of thing objectively.
Another issue I wanted to bring up here was the idea of ‘non-essential’ jobs. Some of the appeal of automation and robots is that they can replace the jobs that people don’t want to do. This may cause an increase in traditionally ‘non-essential’ jobs. This would include basically anything that people view as something you’d do for pleasure.
Some examples being jobs like professional “cuddlers”, prostitutes, massage therapists, staff at vacation resorts, personal trainers, bartenders, artists, performers, etc. While many of these are arguable, my point is just that if I was in a money crunch – I would likely not find myself spending money on the services I’ve listed above, and I would survive all the same (although with some possible negative impacts).
Lets say that after many jobs have been replaced by automation, society changes such that a lot of people take up employment in these ‘non-essential’ positions. In general, one might expect that society would be subjectively happier overall, and that people would find more enjoyment out of their jobs. This is under a few pretty large assumptions, however:
- People pursuing these non-essential jobs would be more likely to seek employment doing things that they personally enjoy and find meaning/value in, instead of pursuing jobs strictly to ‘make ends meet’.
- With more of these ‘non-essential’ offerings available, people would be statistically more likely to find a non-essential service that they personally enjoy/value enough to spend their money on.
- People are able to make a livable wage through these non-essential jobs (possibly supplemented by Universal Basic Income?).
My point in this ‘thought experiment’ is that even if automation doesn’t end up creating derivative jobs like I discussed above – it could result in an overall increase in the happiness of the population. Should the taxes somehow be related to the subjective happiness (averaged) across the workers that the automation/robots are replacing? For example – maybe a company isn’t penalized with increased taxes so much if they’re automating jobs that tend to produce “sad” employees, but they are penalized with increased taxes if they’re automating jobs that tend to produce “happy” employees. This idea seems very far fetched and unlikely… but it’s certainly not completely outside of the realm of possibility.
My conclusions at this point (without further discussion) are as follows:
- Targeting robots for taxation seems like it’d cause more confusion than good.
- Targeting profits vs. humans employed seems like it’s getting closer to a feasible solution.
- Looking at the averaged subjective happiness of the population seems pretty far fetched and unlikely. I think focusing #2 above is likely the most reasonable approach.
Perhaps I will revisit this subject at a future time if I see that the discussion is actually gaining any political traction.