The Automation Ladder
Anything you can define, you can eventually automate — so the real job is choosing what's next.
- Difficulty
- Moderate
- Time to result
- ~ongoing to results
- Steps
- 6
- Confidence
- 87%
Wolfram's answer to AI-and-jobs is a repeating ladder drawn from technological history. A big chunk of work people had to do for themselves gets automated; those jobs go away; but that same automation enables many new categories of things people can do. His key move is the asymmetry underneath it: once an objective is defined, automation for it is achievable in principle. What isn't automatable is the question of what to do next — because there is an infinite set of things you could do, and which ones we choose is a matter for humans. The cycle then repeats: what is a matter of human choice eventually gets standardized, gets automated, and you move on to another stage.
Origin
Wolfram grounds the pattern in the economic history he's watched: 150 years ago most Americans did agriculture by hand, machinery automated it, and people said nobody would have anything to do — but the automation enabled entirely new types of work. He then lived the pattern personally, spending 40 years automating low-level programming and watching what remained for the humans who used his tools.
Core principles
- 01Once you've defined an objective, you can build automation that achieves it — it might take 100 years, but in principle you can.
- 02Automation destroys a category of jobs and opens many new categories that the automation itself enabled.
- 03The cycle repeats: human choice, then standardization, then automation, then a new stage.
- 04What survives automation is deciding what to do next — an infinite set with no defined objective.
- 05New jobs cluster around what is still a matter of human choice.
- 06The first jobs automated are the ones closest to the automation itself.
- 07Choosing what to do next requires computational thinking — you must define the objective precisely enough to execute.
How to run it
- 1
Identify what's currently done by hand
Find the big chunk of work that people have to do for themselves — the category where humans supply the effort because no machine does it yet. This is the rung about to be automated.
- 2
Watch for the objective getting defined
Automation follows definition. Once you've defined an objective you can build automation that does that objective — maybe it takes 100 years to get there, but in principle it's doable. The moment a task's objective is crisply stated, its clock has started.
Pro tip Machine translation was defined as an objective at the beginning of the 1960s and took an extra 60 years to actually happen — but it happened.
Watch out A long lag is not a reprieve. The 100-year version arrives too.
- 3
Expect the jobs to go and the categories to open
Telephone switchboard operators plugging wires in did go away when switching was automated. But that automation opened many other categories of jobs — the telecommunications infrastructure that now enables things like podcasting.
Pro tip Look at what the automation enables downstream, not just at the displaced role.
Watch out 'Then nobody's going to have anything to do' has been said at every rung and has been wrong at every rung.
- 4
Move to the human-choice layer
New jobs get created around the things that are still a matter of human choice. Left to its own devices, an AI has an infinite set of things it could be doing; which things we choose to do is really a matter for us humans. Position yourself there.
Pro tip Of everything out there to compute, the slice human civilization has cared about so far is a tiny tiny tiny slice — the choosing work is nowhere near exhausted.
- 5
Learn to define what's next precisely
Once the routine work is automated, most of what you have to do is figure out what you want to do next — and that's exactly where computational thinking comes in, because you must think about what you're trying to do in computational terms to define it.
Watch out This is harder, not easier. When low-level programming is automated away, people say things have become so difficult now — because the crank-turning was the part where you didn't have to think much.
- 6
Expect your new rung to standardize too
What you do eventually gets standardized, and then it gets automated, and then you go on to another stage. Treat the ladder as permanent, not as a one-time transition you survive.
In the wild
There was a time when telephone switching required operators physically plugging wires in. People warned that automating telephone switching would destroy all those jobs. It did — those jobs went away.
→ But the automation opened many other categories of jobs and built out the telecommunications infrastructure that now makes video communication and podcasting possible — enabling work nobody in the switchboard era could have named.
Several years back, people warned that machine learning would put all these people out of jobs. Wolfram found it amusing, because he knew perfectly well which category would be hit first: machine learning engineers themselves, since machine learning can be used to automate machine learning.
→ The prediction illustrates his rule that once a thing becomes routine it can be automated — and the people closest to the automation technology are not exempt from it.
Wolfram spent a large part of his life automating low-level programming through Wolfram Language. Users report doing in an hour of their time what would have taken a month writing standard programming language code.
→ In some segments of the world low-level programming is already extinct. But many people had concluded they could get a good job by learning C, C++, Python or Java — and Wolfram says that thing to spend your human time on is not necessary. What's left for those users is figuring out what they want to do next.
Common mistakes
Assuming automation leaves nothing to do
150 years ago most people in the US did agriculture by hand. Machinery automated it and the fear was that nobody would have anything to do. It turned out they did — because that very automation enabled a lot of new types of things people could do. The lump-of-labor instinct has failed at every rung of the ladder.
Investing your career in the crank-turning layer
Wolfram is blunt that low-level, turn-the-crank programming should be extinct already after 40 years of effort to automate it. People who bet on it as a durable good job bet on the layer that automation reaches first.
Expecting the post-automation job to be easier
When the routine is automated out, people say things have become so difficult now. The routine work was where you didn't have to think much — you turned the crank and got a piece of code written. What's left is the hard part: figuring out what to do next.
Is it for you?
Best for
Anyone deciding what skills to invest in, and anyone reasoning about AI's employment effects from a historical rather than a panicked baseline.
Not ideal for
People needing immediate income protection during a displacement — the framework is structural and offers no timeline for individual transitions.
From the transcript
“once you've defined an objective you can build automation that does that objective maybe it takes 100 years to get to that automation but you…”
“yes those jobs went away but that automation opened up many other categories of jobs”
“a lot of new jobs get created around the things which are still sort of a matter of human Choice what you do eventually it…”
“if you've automated out all of that what you realize is most of what you have to do is figure out so what do I…”
“the first category of jobs that would be impacted were machine learning Engineers because machine learning can be used to automate machine learning”
From the episode
Stephen Wolfram: AI, ChatGPT, and the Computational Nature of Reality
Stephen Wolfram