AI won't take your job. The AI-literate knowledge worker will.

Rob Dixon

May 25, 2023

4 min read

Six months ago ChatGPT launched. Since then Facebook, Amazon, and Google have spent billions catching up, Elon Musk has asked for all development to be paused, and Italy banned it outright and then changed its mind. I've just given a talk at LTSE2023 on what it means for higher education. The recording is at the bottom, and this is the argument.

The phenomenon isn't the AI. It's the access.

AI isn't new. Retrieval AI, the kind that looks things up, has been in the hands of tech workers for years. They used it to build applications for the rest of us. What changed in November is that generative AI reached out of the IT department and into everyone's lives. A non-technical person with a browser and a no-code tool can now build, in a day, something that would have cost millions a year ago.

That is why this keeps coming up and why it isn't going away. Nobody is going to resist that much commercial power.

Two curves at once

The first curve is the cost of software. A coder with an AI assistant writes something like ten times the lines they used to. The bottleneck that limited the internet boom, not enough people who could build things, is dissolving.

The second is knowledge work. Goldman Sachs put 300 million jobs in the exposed column, with office administration, legal, engineering, and finance near the top. Within four weeks of the ChatGPT API appearing, Octopus Energy was handling nearly half its inbound customer enquiries automatically. Its CEO called generative AI a freight train about to hit the labour market, and most of us don't know it's coming. Add those two curves up and this goes faster and wider than the internet did, with the graduate job market changing under our students' feet while they're still on a three-year course.

The 4x knowledge worker

Here's the picture I keep coming back to. Take everything a knowledge worker does, from gathering the material to drafting, formatting, and promoting the result, and colour in blue whatever the AI can already do. Most of the page is blue. What's left uncoloured is the human's job: clarifying the input and checking the information is right, deciding who the output is for and what shape it should take, and adding the critical narrative. The new thought. The questions. The juice.

A researcher working this way doesn't stop reading. They stop spending their week on the gathering and the typing, and spend it on the thinking, and the same material becomes an 8,000-word article, a one-pager, a student version, and a business version in an afternoon. ARK Invest call that the 4x knowledge worker. So my line for the room was this: AI isn't going to take your job. The colleague who's fully AI-literate and operating this way will. With the caveat that every productivity leap in history has created new work, and I don't see why this one would be different.

How it works, in one paragraph

Retrieval AI looks data up, is always accurate, and repeats itself. Generative AI holds no data at all. It's a compression algorithm: the opposite of big data. Billions of dials, trained by running text through until "for my birthday I'm going to make you a..." reliably comes out as "cake", then tuned by humans rating its answers. It predicts. That's why it can write and why it will confidently invent research on sheep farming in central Manchester if you ask it about something nobody has published on. In a rich field it's accurate. In a thin one, check.

Why the strategy committee won't work

Universities plan like positivists. Analyse, debate in committee, implement, on the assumption that a correct strategy exists and only needs finding. In a stable environment that's fine. In this one there is no correct answer sitting there to be discovered. Sam Altman doesn't know how it ends. Neither does Satya Nadella. Neither do I.

What works in turbulence is what we already teach: lean startup, effectuation, small experiments, rapid iteration, and a willingness to pivot. And it depends on a threshold most senior people haven't crossed yet. Only by using the tools do you move from unconscious incompetence to conscious incompetence, the moment you realise you have no idea what this means. That sounds like bad news. It isn't. It puts you level with everyone at the frontier, and it's the only place from which a sensible discussion about assessment, research, or professional services can start. Literacy first. Then formulate. Then keep iterating.

I've started calling the sequence get urgent, explore, formulate, iterate, embed. There's a paper coming. For now the practical version is a one-day Festival of AI Literacy: the whole faculty, leadership, students, and a few local business people in one room, a morning on what's coming for the job market, and an afternoon with hands on the tools, so people leave able to do something, not just told about it. I've run versions of this with company boards and small businesses, and the next-day results are what convinced me. One engineering firm in Middlesbrough had technical data loaded into a sales chatbot by the time I rang the MD the next morning.

The pilot in the storm

I closed with Saras Sarasvathy's image. In normal flight a pilot plans the route and executes it. In a storm, the pilot can't control the storm. They control the throttle, the altitude, and their own calm, and they tell the passengers they're all in it together. We don't have to control this. We have to focus on what we're good at, and on the one thing each of us can change, which is our own literacy.