Take 2 on AI Takeover
Yesterday, I had a rather long post about one writer’s positive take on the automation/artificial intelligence/robot trend coming fast at us. I didn’t buy in his positive thesis, even though I wanted to believe in something positive, but his thesis just didn’t convince me – it was too free market oriented. I did want to bring up another somewhat positive viewpoint but the post was way too long by the end, so today is going to be a very brief post on this other somewhat positive thinking on the automation trend.
But first, I do want to mention a brief research I did after publishing the post: I went looking for book reviews to see what critics were saying. I found mostly readers’ reviews and only one “critic” review and all were positive. None expressed any disquiet. That is understandable because his arguments were very compelling on the surface. The reason why his arguments failed for me was because of all of the talk about inequality, about people wanting real change, about people agreeing to the idea that millionaires and billionaires should be taxed more, and so on. There’s a real angst in the air. So, his arguments about the beneficence of the wealthy just didn’t ring true to me.
And now, on to the second somewhat positive view.
This is Paul Krugman’s viewpoint on presidential candidate Andrew Yang’s idea of providing universal basic income because of the coming jobs demolition from AI. The economist has argued that the data does not support the idea that there has been (or will be?) mass unemployment by technology: if we are undergoing massive technological replacement by machines, then productivity should be soaring but it’s not – it is flat. He feels that the fear of the technological revolution may be overblown.
Oh, productivity is flat?
Looking at government data
Okay, I’m looking at a government site, Bureau of Labor Statistics, and I see that there are two types of productivity figures: labor productivity and multifactor productivity. So, which one is Krugman using? Maybe I can look at both? I’m going to download the CSV version for both.
Okay, there’s a heck of a lot of information coming over – 60 field headers for the labor productivity by major sector. There’s a lot to play with but I’m going to stick with the field category “Productivity”. I’m pulling over 4 datasets: labor productivity by major sectors; labor productivity – durable and nondurable; multifactor productivity by major sectors; and multifactor productivity – durable and nondurable.
I’ve played with the data and I see where Dr Krugman is saying productivity statistics appear to be flat. Here are some screenshots from Power BI where I see the flattening out since 2010. I also see a strange drop off for the year 2019. I’m not sure if that means the government has quit collecting data or what. I hope not! Maybe next year there will be a correction in the data.
Labor Productivity
Multifactor Productivity
I’m not going to upload all of the multifactor charts I did because they pretty much tell the same thing.
If you want to get the data yourself
If you want to see what data I used, here’s are some screenshots and the web address to get there. I’m just using the field for Labor Productivity (2012 = 100) just to get a quick take on what is going on. Also, to get the quarter and year to go in sequence, I added a column in the table at the query stage and created the column by example. Then I changed the data type from text to date and that led to the correct ordering sequence. The date type changed Q1 or 01 to January, Q2 or 02 to February, etc. but it didn’t matter for this quick take. When you look at the chart you can’t tell that is going on.
At the BLS site, I went ahead and clicked on the Labor Productivity hyperlink until I came to the labor productivity page: https://www.bls.gov/bls/productivity.htm
Then I scrolled down the page until I found the following. I clicked on the CSV files.
On that page ( https://www.bls.gov/lpc/tables_by_sector_and_industry.htm), I chose the following 4 links:
My analysis of what I’m seeing
There is a huge uptick during 2009-2010 and then productivity flattens out. I can see the gradual rise since 1980, with a bit of an uptick around 1985 – 1987, up through 2009 where another jump in productivity occurs. But since around 2010, productivity appears to flatten out, especially in the manufacturing sector. Manufacturing productivity is actually pretty weak and you would think that is where a lot of the automation is occurring.
Those are very interesting results and rather shocking, so maybe technology won’t be all that adverse. However…my experience with just Excel tells me something is off. Just using Excel alone allows me to do things more quickly and more accurately. Once I start working faster, I then move on to other things that we never do because of lack of time. And maybe it is that other thing is what is going on.
But, but, but. If we are still driven by that shareholder primacy, then at some point, the shareholders or that profit motive will drive to full automation. The question is: will new jobs be created as quickly or faster than jobs are destroyed? Can workers/employees learn new skills fast enough? My fear is that the answer to both questions is no: new jobs will not be created quickly enough and workers will not be able to learn such new skills quickly enough. There is a potential for massive and extremely painful dislocations.
The other thing that I have thought of is, and this is just a thought: what if this AI takeover works in an exponential fashion? Exponential is where capability or speed doubles after “x” amount of time, much like that Moore’s Law. Technology appears to act in an exponential fashion. So, if AI takeover acts in an exponential fashion, and there is nothing to really indicate it does act that way other than that it is a technology, then job loss would be experienced in an exponential fashion. If that is the case, then once job loss hits 1% of the working population, then it will just take just 6 or 7 doublings to take over all jobs.
- 2%
- 4%
- 8%
- 16%
- 32%
- 64%
- 100%
There are a couple of questions to ask about this exponential rate.
- Is there an upper limit to what AI can do: does it take over all jobs or are there some jobs it just cannot do? What is the percentage of jobs that it cannot do?
- What is the time frame for the doubling? One year, a year and a half, 2 years? If it’s 1 years, then the 100% would be in 7 years but if the doubling is 2 years, then we are looking at 14 years.
- And where are we in the process? Are we close to 1% or have we already passed it?
Again, this is just a thought. They always say with exponential, in the beginnings you don’t notice it happening and then in the end, it just happens so rapidly that it’s incomprehensible.
And I prefer Dr. Krugman’s take over the author’s.
You must be logged in to post a comment.