The Impact of AI, Revisited (2018 → 2026)
An update to my September 2018 post, The Impact of AI.
Eight years later
In September 2018 I wrote a short post asking three questions about AI and jobs. ChatGPT was still four years away, and self-driving cars were my main example.
This is that post, revisited. I kept the three questions and graded my own answers. Then I added the two topics I only pointed at back then: universal basic income and the post-scarcity economy.
Will AI only replace monotonous, repetitive work?
2018 me: No. It starts there and heads for more complex jobs.
2026 verdict: right, and I undersold it. The driving example aged well. By March 2026 Waymo was giving 500,000 paid driverless rides a week across 10 US cities, up from 50,000 in May 2024 (TechCrunch).
What I got wrong is the order. I assumed physical, routine work would go first. AI went for writing, code, analysis and support instead, the desk work that looked safe in 2018.
So far the damage is at the entry point. Stanford researchers using ADP payroll data found employment of 22–25-year-olds in AI-exposed occupations is 19% below where it would be had it kept pace with less-exposed peers. The cause is fewer hires, and the same paper finds no economy-wide job loss yet (Brynjolfsson, Chandar and Chen).
Will AI collapse the world economy?
2018 me: As AI takes jobs, people lose purchasing power, demand falls, and the companies that automated lose their customers. My example was the Netflix engineer who can no longer afford Netflix.
2026 verdict: the logic holds, the timeline doesn't. There has been no collapse. The US has added about 3 million white-collar jobs since ChatGPT arrived, and real wages in professional services are up 5% (Washington Monthly).
The loop I described is still unsolved, though. One company can cut payroll and keep its customers. Every company cannot do it at once, because one firm's payroll is another firm's revenue.
AI can produce plenty. The open question is how people get the income to buy it, and that is where UBI comes in.
Will AI replace jobs only in developing economies?
2018 me: Outsourced routine work is exposed. But a smart, less experienced worker with AI tools can compete with a veteran in the developed world.
2026 verdict: both halves now have evidence. In a study of 5,179 customer-support agents, an AI assistant raised productivity 14% on average and 34% for novices, with minimal gain for experienced agents (Brynjolfsson, Li and Raymond). That is the leveller I was hoping for.
The other half worries me more. Support and back-office work is exactly what outsourcing economies sell. If a tool makes a novice 34% better, the buyer eventually asks whether the novice is needed at all. Those countries also have the least budget for a safety net.
Universal basic income: a patch for the demand problem
UBI is a regular, unconditional cash payment to every adult, with no work requirement. If wages stop being how most people get purchasing power, something has to replace them. That is the whole argument, and it targets the exact loop I described in 2018.
What the pilots show
| Pilot | Design | Result |
|---|---|---|
| OpenResearch, US | $1,000/month for 3 years; 1,000 recipients, 2,000 controls | Employment down 2.0 points; 1.3–1.4 fewer hours worked per week; no change in job quality |
| GiveDirectly, Kenya | $22.50/month for up to 12 years; about 200 villages | No drop in total work; shift from wage labour to self-employment; the 12-year arm invested more than the 2-year arm |
Both answer the laziness objection: people mostly keep working. Neither answers the harder questions.
Why I'm only cautiously for it
- Cost. $1,000 a month for roughly 260 million US adults is about $3 trillion a year, by my rough math.
- Pilots don't test the real scenario. They pay a few thousand people in an economy where jobs still exist. Nobody has run UBI at a scale that could move prices and rents.
- Money isn't meaning. A cheque replaces income. It doesn't replace the structure, status and purpose people get from work.
- Who pays. If AI's gains concentrate in a few firms, the funding has to come from there, through taxes on automated profits or a public dividend like the one Alaska pays from oil revenue.
I still land on yes. I would rather start small, with a partial dividend that grows as automation does, than debate it until the day we need it.
The post-scarcity economy
In 2018 I left one line telling readers to look up "post-scarcity economy". Here is the longer version.
Post-scarcity means most goods and services can be produced with so little human labour that they become close to free. Economics is about allocating scarce things. Post-scarcity is what happens when the scarce list gets short.
What is actually becoming abundant
Digital goods got there first. Another copy of a song, a film or a map costs almost nothing to make.
Intelligence is following. The price of querying a GPT-3.5-level model fell from $20 to $0.07 per million tokens between November 2022 and October 2024, a drop of more than 280 times (Stanford AI Index).
What stays scarce
- Land and housing in places people want to live
- Energy, compute and the minerals behind them
- Human attention, trust and care
So the near future looks lopsided. Anything made of bits or cognition gets abundant, while atoms and locations stay as scarce as ever.
Abundance doesn't distribute itself
Cheap production only helps people who have some income, or who get the goods provided publicly. A product at 10% of today's price is still out of reach at zero income.
Ownership decides the rest. Whoever owns the models, the robots and the data centres captures the surplus by default. Post-scarcity is the destination and UBI is one bridge to it, which is why I put the two topics in the same post.
Where I land
The 2018 post was right about direction and wrong about sequence. AI came for complex work sooner than I expected, and it came for desk jobs before driving jobs. The economy hasn't collapsed, but nobody has fixed the demand loop either.
UBI won't fix everything, and the funding math is hard. I still think some version of it is the most practical bridge to an economy where intelligence is cheap and jobs are fewer.
I ended the original with two videos. Both are worth rewatching with 2026 eyes.
The Last Job on Earth: Imagining a Fully Automated World (Guardian Animations)
Humans Need Not Apply (CGP Grey)
Sources
- Waymo's skyrocketing ridership in one chart, TechCrunch, March 2026
- Canaries in the Coal Mine?, Brynjolfsson, Chandar and Chen, Stanford Digital Economy Lab, revised August 2026
- How AI Broke the Entry-Level Job, Washington Monthly, May 2026
- Generative AI at Work, Brynjolfsson, Li and Raymond, NBER
- Unconditional Cash Study: employment findings, OpenResearch
- Early findings from the world's largest UBI study, GiveDirectly
- AI Index 2025: State of AI in 10 Charts, Stanford HAI
Comments
Post a Comment