I suspect that any labor that is rules-based — work governed by math, statistics, law, science or engineering — and that doesn’t require physical human dexterity will be the easiest to replace. HR is on that list too, all of it below the C-level. Next will come the so-called “creative” labor. Then, finally, the manual labor.
I’m still unclear where the new jobs are that AI will create, outside of those required to build AI and AI-tangential business.
The people paid to model this have arrived at roughly the same first step. Goldman Sachs economists estimated in March that generative AI could end up automating about 300 million jobs over the next decade, and the shape of the estimate matters more than the headline number: they put 46% of administrative tasks within reach of automation, against 6% of jobs in construction. A paper posted the same month under the title “GPTs are GPTs” found that “around 80% of the U.S. workforce could have at least 10% of their work tasks affected,” with higher-income jobs potentially facing greater exposure. And on May 1, IBM’s chief executive told Bloomberg the company expects to pause hiring for back-office roles that AI could do — HR among them, roughly 26,000 people — saying, “I could easily see 30% of that getting replaced by AI and automation over a five-year period.”
Manual labor sits last for a reason the roboticists named decades ago. Hans Moravec wrote in 1988 that it is “comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers, and difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility.” Thirty-five years on, the paradox holds. Reading a contract turns out to be easier to teach than lifting a box.
And just an aside — while I believe humans are immensely complex and sophisticated, our single greatest accomplishment wielding those traits may just have been the thousands of years we’ve spent documenting our own outputs, in a way a computer can now learn our reasoning from. Every textbook, statute, proof and procedure was assiduously written so that a stranger could follow it. What we took the trouble to write down, we made easy to learn — we just assumed the learner would be human.
After ingesting that vast body of human creation, I don’t feel machines will have much of a challenge picking up where we may shortly leave off. Knowing all that we know, why is it so hard to believe a machine can do the things we’ve so thoroughly documented how to do?
I’m blatantly of the rather aggressive position that s**t is about to change at warp speed — and that’s unlikely to be how it actually goes.
