Charlie Ball's answer to the question of how to choose an AI-proof career is that you mostly cannot choose one. Shielding a 45-year working life from fast technological change is, he says, a hard problem, and the better question is not which professions are safe but which set of skills lets a young person keep adapting as the labour market shifts. What he is confident about is narrower: AI is unlikely to replace work built largely on face-to-face human contact any time soon. Pressed to name fields that may hold up better than others, he gives six.
Research and development comes first. Degrees in science, technology, engineering and maths have long carried good career prospects, and while AI will inevitably reach entry-level lab work, the creative and imaginative parts of the job stay with people. The point of R&D is to find what is not there yet, and AI is mostly able to compare information against what is already in its database.
Engineering follows the same logic from the other end. AI will be used more and more in design and modelling, but engineers are still needed in physical environments where human oversight is required. Ball thinks a large share of engineering tasks will be hard to hand over, and that people will still be required on major projects — civil engineering, bridges, buildings. Those areas, in his view, are likely to stay durable.
On creative work and entertainment his case rests on taste. AI can produce art, but Ball's view is that people broadly do not like it very much, and that audiences will get better over time at spotting it. Demand for human-led creativity will persist, he argues, because everything AI makes today is a rework of ideas that already exist: it does not originate, and it does not produce a recognisable voice of its own.
Medicine, he says, will still need doctors in fifty years, though the content of the job will change. People want medical news delivered by people, particularly serious diagnoses and other difficult conversations. However good automation becomes, a parent with a sick child will want to see a human GP. AI can assist with diagnosis in specific cases, but it will still get things wrong, and where someone has to carry responsibility for a decision, that someone will be a person.
Nursing and midwifery he treats as a case where the technology is welcome. AI and adjacent tools will spread through many medical roles, and Ball thinks that is broadly useful given how badly short of staff these jobs already are. But sick people will still want to be cared for by people, and people will still need other people to look after them in old age. He cannot picture a robot midwife.
Education closes the list. Robot teachers will not replace teachers either, despite England's continuing teacher shortage. Technology already helps organise how teaching is delivered, but human teachers are needed to connect material to real life and to support students beyond the immediate academic task, and that part of the job keeps its value.
Read together, the list is not really about cognitive difficulty. Nothing on it is protected because AI cannot do the thinking; the bridge is designed with modelling software, the diagnosis is assisted, the lesson plan is organised by tools. What each of these jobs has is a body in a room and a named person who is answerable when it goes wrong. Ball says so directly about medicine — someone has to take responsibility — and the same structure holds for the midwife, the site engineer and the teacher. That is a more useful filter than any list of degrees: look for work where physical presence and legal accountability are the product.
By that filter, the creative entry is the weakest of the six. It does not rest on presence or liability; it rests on the claim that audiences dislike AI output and will learn to detect it. Preferences are the least durable thing to build four decades on, and detection has a habit of losing to generation. If the defence of creative work is that people can tell, the defence is one model release from needing a rewrite.
There is a second wrinkle Ball names without drawing the conclusion. Two of his six — nursing and midwifery, and teaching in England — are described in the same breath as chronically short of people. Their resistance to automation may owe less to what AI cannot do than to demand running far ahead of supply. Those are stable jobs for as long as the shortage lasts, which is not quite the same promise.
Which leaves the part of the argument that should worry an eighteen-year-old most. The one concession Ball makes is that AI will reach entry-level lab work. That is not a peripheral loss: entry-level work is where careers are built, the years in which a junior researcher becomes the senior one whose imagination is supposedly safe. If the technology takes the bottom rung of the ladder while leaving the top intact, then a field can be AI-resistant for everybody already standing on it and closed to everyone trying to get on.