GenAI Is Like a Flesh-Eating Bacterium
(draft) How this wave of innovation isn't like the others.
Before diving in, two declarations: First, I love what Generative AI is capable of doing, and am optimistic we can figure out how to use it intelligently. Second, we're all trying to sense our way into the future; this is one such attempt, meant as observations, not clairvoyance.
That said...
The huge wave of innovation around Generative AI has led to many conjectures about how it will affect employment. Opinions range from GenAI just being a "normal" technology (so don't worry!), to it heralding the advent of Fully Automated Luxury Communism (less likely than pigs flying), to GenAI catalyzing a general economic meltdown.
It's this last extreme possibility that has me worried, for several reasons. The reason I'll cover here is that the GenAI revolution isn't like previous revolutions.
There are always new jobs!
In the 90s, I got a cold call from a researcher who was investigating the future of paper. When we talked, Adina Levin, who became a good friend, proceeded to ask me great questions, which I tried to answer to the best my perspective allowed. Each time I thought she was done asking questions, she'd ask another good one.
Our conversation made me notice that the "paperless society" was a mirage on the horizon: every time we'd wipe out one use of paper (sent a paper check recently?), another would pop up. If you look back at the Amazonification of shopping, you'll wish you'd invested in cardboard makers when Amazon was just an innocent creek.
Conventional wisdom holds that every prior wave of automation or innovation wiped out some jobs, but created new ones:
- Cars obsoleted blacksmiths (and many other trades; "peak horse" in the US seems to have happened in 1915), but brought gas stations, auto repair shops and more.
- Airlines replaced most rail and ship travel (well, in the US), but took us farther, faster than ever, while employing quite a few people.
- The printing press wiped out scribes, but certainly sparked many new industries, not to mention multiple wars and revolutions.
- Electric lights unemployed lamplighters, candle makers and a few whalers, but ushered in a transformative wave of industrialization and allowed us to turn night back into day.
Each of these revolutions fueled a replacement market with novel jobs.
Those new jobs weren't obvious to the people populating the eclipsed markets. The first cars were noisy, slow, and unreliable (see Red Flag Laws). There were no freeways or gas stations. It would have taken a first-rate clairvoyant to foresee the complex and overwhelming vehicular landscape we now take for granted — and are trying to undo — never mind the variety of jobs needed to keep it moving.
Many smart folks like Matt Stoller are saying the doomsayers are overreacting. I hear them, but I'm still worried, partly because we're still early in this shift, yet people are drawing broad conclusions already. This post isn't meant as a conclusion, but as a warning about a phenomenon we should keep an eye on, because it affects how we act over time.
This time is different
What's different? When GenAI masters a task — say consolidating regional sales spreadsheets and writing a summary report — that task disappears into the machine. It no longer needs to be done by humans. (Though someone should check the results — for now.) The task is still being done, at ever-lower cost, but it doesn't spawn a new task to offset the work loss because the work isn't lost at all.
So GenAI can "eat" tasks the way that necrotizing fasciitis eats human flesh. Alas, the simile is memorable but imperfect: it turns out the bacteria that we call flesh-eating don't bite, they secrete toxins that kill off human cells, making a very deadly mess. No chewing involved. But I bet the simile stays with you a while. And the disappearing of tasks is going to feel very real, very soon.
Problem is, we won't notice it because we're lulled by those previous replacement waves, plus the false confidence from knowing that the new jobs weren't apparent early on.
It's become cliché to cite Taleb’s Black Swan, but you may remember that he describes how every day is great for the turkey up until just before Thanksgiving. He doesn’t see the one bad day coming:
Consider a turkey that is fed every day. Every single feeding will firm up the bird’s belief that it is the general rule of life to be fed every day by friendly members of the human race “looking out for its best interests,” as a politician would say. On the afternoon of the Wednesday before Thanksgiving, something unexpected will happen to the turkey. It will incur a revision of belief. (The Black Swan)
I like "revision of belief." It's like the "rapid unscheduled disassemblies" at SpaceX. I have a feeling we're living through a Turkey Moment. This time is different, but we are blind to the change.
Timing matters
Previous, similar shifts in employment happened more slowly, so the traumatic bits had more breathing room to adjust.
Take, for example, the Industrial Revolution. Around 1590, Queen Elizabeth I refused William Lee's patent application for the stocking frame (making stockings was a huge cottage industry, keeping many households afloat through piecework), saying:
Thou aimest high, Master Lee. Consider thou what the invention could do to my poor subjects. It would assuredly bring to them ruin by depriving them of employment, thus making them beggars.
Karl Polanyi's The Great Transformation (1944) is one of my favorite history books (here's a taste of what it taught me). It describes the changes that happened in Britain during the early Industrial Revolution. Polanyi, an economic historian, describes the many efforts taken to try to protect those affected by the onrush of industrialization, including a variety of Poor Laws and the Speenhamland System, a kind of wage guarantee based on the price of a gallon bread. They didn't work very well. Industrialization ran roughshod toward a future of "dark satanic mills," which provoked their own waves of reform laws.
A more recent statistic I like to cite: At the time of the US Civil War (say 1865), roughly 80 percent of the population was engaged in growing or raising food. By World War I (1918), that number drops to 20 percent. Factories and offices (and agricultural progress) pulled most of the country off the farm in half a century. Today, the number is 1.5 percent. The shift away from farm work really changed America.
My, those old times look quaint.
GenAI's automation of work is happening so quickly that many folks are jumping to conclusions based on slim early data coming from companies that have experimented with it.
What's worse, organizations likely don't have a good picture of the extent of the change today. So-called Shadow AI masks much of the damage: employees use the technology but don't say they are. How do you incorporate that data into your decision-making, especially if trust is missing?
The mismatched timing is a big part of the problem.
Which jobs?
Dario Amodei's warning about white collar work. Entry level jobs going away.
Jensen Huang's example is long-haul trucking: tktk
Frontier work now is Agentic AI, where we can leave the AIs alone and they will do the work. tktk.
Takes ever fewer workers to do the same work. tktk
Whole categories of work will disappear, and the people who were doing those jobs will have a hard time retraining for something that matches their old income. The GenAI Job Reentry Dilemma.
As GenAI spreads unchecked through organizations, it is tearing holes in the workforce. But because GenAI Automates Tasks, Not Jobs (so far), tktktk.
tktktk
If such a thing comes to pass around GenAI, I do not promise to eat my hat, but I will be very surprised.
GenAI will eat many new jobs — but which ones?
John Henry, with complications
In the race between humans and machine intelligence, machine intelligence has huge advantages. Humans, as they age, get a bit smarter, but also get more expensive. They want raises. They are hard to manage. They earn longer vacations. Machine intelligence just gets faster cheaper better.
US Businesses Have Given Workers Good Reasons to Mistrust Them
Now there's a fancy new power tool on deck, and it has really sharp blades.
Trust implications
The
Without trust, we can't collaborate to find better, more adaptable solutions.
Open questions
This post leaves several juicy nuggets on the table that are fodder for future posts:
- How will workers move "up" the knowledge chain?
- Will organizations flex hard enough to absorb these changes humanely?
- Will we actually use AI to augment workers, instead of replacing them?
- How should we rearrange work?
Conclusion here.
This situation reminds me of Kranzberg's First Law of Technology (of six): Technology is neither good nor bad; nor is it neutral.
Where are you on the "GenAI will eat all jobs! to Don't worry, there are always new jobs!" spectrum?
#futureofwork #dangersofAI #automation
This article is cross-posted on Substack here, Medium here and LinkedIn here. It's also here in my Brain.