Generative AI is all the rage in 2023.
Maybe it’s a fad, maybe it isn’t, but if it lives up to the hype (or really, even if it comes anywhere close), it could have dramatic economic implications, not all of which bode especially well for carbon-based labor.
I’ve read several apocalyptic accounts documenting the outlook for workers in industries seen as especially vulnerable. Initially, I found such accounts to be hyperbolic, but I have to admit that after experimenting with publicly-available generative AI tools over the past two months, and extrapolating from their current capabilities to what may be possible if they evolve along the same arc as other technology, the outlook for humans in certain professional capacities does indeed seem dire.
In an expansive new note, Goldman took a deep-dive into generative AI and one of the key takeaways was their estimate of labor substitution.
As usual, I’ll spare readers the details of the bank’s methodology and skip straight to the point. “Roughly two-thirds of US occupations are exposed to some degree of automation by AI, and of those occupations which are exposed, most have a significant — but partial — share of their workload (25-50%) that can be replaced,” the bank’s Joseph Briggs and Devesh Kodnani wrote.
The charts above give you a sense of how existential this could potentially be. Goldman weighted their estimates by the employment share of each occupation in the US Occupational Employment and Wage Survey and aggregated to the industry level.
“One-fourth of current work tasks could be automated by AI in the US, with particularly high exposures in administrative (46%) and legal (44%) professions and low exposures in physically-intensive professions such as construction (6%) and maintenance (4%),” the bank wrote. The same is generally true in Europe.
Obviously, exposure varies by country. In some locales, it’s especially acute. For example, in Hong Kong, nearly a third of full-time equivalent employment is exposed to AI replacement.
The good news for less developed nations, emerging markets, agricultural societies and some Latin American countries, is that ChatGPT can’t cultivate crops or run a cocaine processing plantation. But the further “up the line” you go (so to speak), the more perilous the situation is for human workers.
The report, which spans 20 pages, isn’t amenable to any sort of “summary treatment,” although I suppose AI could “write” a decent synopsis in milliseconds. As noted above, the headline takeaway is simply the bank’s estimate of the prospective substitution effect.
Goldman described the potential for “significant disruption” in labor markets, and said that globally, generative AI “could expose the equivalent of 300 million full-time jobs to automation.” Bottom line: Nearly one-fifth (18%) of the work performed on planet Earth “could be automated by AI on an employment-weighted basis,” the bank wrote.
A case in point thanks to Tom’s Hardware:
“Nvidia Tech Uses AI to Optimize Chip Designs up to 30X Faster”
My 7 year old loves playing the game Minecraft. There’s an in-game item called a “command block” that allows you to program the game to do virtually anything. Unfortunately, using command blocks requires knowledge of both game-specific syntax and Java code. I’m only passingly familiar with the former, and know nothing of the latter. But now thanks to OpenAI-powered Bing search, I present to you the following:
/execute @e[type=arrow] ~ ~ ~ /summon tnt ~ ~ ~ {Fuse:0}
That’s the code to make an arrow fired in the game rain blocks of lit TNT. My kid is gonna love it.
AI is going to be incredibly disruptive–like the invention of the internet itself–but it’s going to make the future a lot more interesting. This power will be harnessed–just like the internet–for both good and evil. But it’s going to happen much much faster.
Technological change since the dawn of the industrial revolution has always been disruptive, but societies have mostly managed to digest that disruption since the rate of change has been manageable. Even then, there was major social upheaval starting with the Luddites and never really ending.
The future, assuming we live to see it, will be incredibly interesting.
Anecdotal, but we’re already seeing this happen at my company. I’m still early enough in my career where I know it’ll be a requirement for me to be well-versed in generative AI’s capabilities to remain gainfully employed. Many use cases still require a decent amount of technical expertise, but companies are racing to integrate generative AI into their products.
If the twitter is to be believed, the way you search for anything on the web will soon change: https://twitter.com/rowancheung/status/1640430038985199616?t=RVaykMMcZhGQCdarlcP5dg&s=19
While I don’t want an economic crash, I’m salivating at the idea of a market downturn so I can go all-in on tech this year. I still think this explains a significant portion of the market resilience we’ve been seeing (in addition to the Fed put which hasn’t really gone away IMHO).
A few days ago I ordered an expensive piece of decorative art from an on-line/catalog company. When I received it it was poorly packed and totally destroyed. Yesterday, I called customer service and got connected to what was a poorly conceived, pre-beta AI driven chatbot, definitely not programed to deal with destroyed product problems. After fifteen minutes of dancing with the bot, I got shoved into and endless loop and soon disconnected on the company’s end. I called back the same number, but without the 1- prefix and got a person. In three minutes I was assured of a solution and how to file my report with a person. So far all the bots I have dealt with are not yet ready for prime time. These things can kill people in the healthcare system, already have, in fact, on more than one occasion. I am glad I am old enough I won’t have to put up with much of this crap. This is not being done to make our lives better folks. It is being done to make a very small group much richer. Fortunately, my daughter and son-in-law are both creator managers of this stuff (though recently laid-off by their companies selling out to a couple of punks).
Selecting “0” (zero) if it is not listed as a menu option can occasionally still override the menu and send you to a live person.
I’ve heard from an Oracle employee that they have been forbidden to use ChatGPT-generated code, at least at this time. The problem that management apparently is trying to avoid right now is questions concerning copyright law (and yes, code can be copyrighted in some cases).