The government has a forecast for these jobs. Our scores can't tell them apart.

In July the US Bureau of Labor Statistics published its forecast for the decade to 2034, and named eighteen jobs that artificial intelligence is expected to lift or sink. Data scientists: up by a third. Procurement clerks: down by a twelfth.

This site has a number for both of those jobs. Data scientists hold up at 43. Procurement clerks hold up at 41.

Two points apart, for a forecast that is forty points apart. That is not a rounding problem or a bad week for our arithmetic. It is the clearest demonstration we have of what a score here is, and what it will never be.

The official forecast, against our scores

Every one of the eighteen jobs sits between 36 and 55 on our scale. The scale runs from 30 to 91.

The nine jobs the BLS expects to grow average 47 on this site. The nine it expects to shrink average 45. Run the correlation between the two columns and it comes out at 0.14, which in plain terms means: knowing our score tells you almost nothing about which way the official forecast points.

The sharpest case is a pair that share a desk. Medical secretaries and medical transcriptionists are the same job on the international list, so this site gives them one number: 47. The BLS expects the first to grow by 4% and the second to shrink by 5%.

Why they disagree

Because they are answers to different questions, and only one of them is a forecast.

A score here is a description of today’s work. The ILO’s raters went through the tasks in a job and said how much of each one today’s AI tools could do. We average that and subtract it from 100. Nothing in it knows about demand, budgets, interest rates, an ageing population, what health insurers will pay for, or whether a firm takes its savings as more output or fewer staff.

A BLS projection is made of exactly those things. It starts from the size of the labour force in 2034, works through what each industry is expected to produce, and only then asks how many people of each kind that takes. Technology is one input among many, and it enters as a judgement about staffing rather than a measurement of tasks. The result answers “how many of these will there be”, which is the question most people actually have.

So the two can disagree without either being wrong. A job can be full of tasks a machine can do and still grow, because the work itself is growing faster than the machine is taking it. Software developers are the clean example: they score 47, among the most exposed professional jobs we measure, and the BLS still expects 268,000 more of them by 2034, the biggest increase of the eighteen.

Zoom out and they do agree

One thing is worth holding on to before this turns into “numbers are useless”.

Look at the shrinking end of the list and you keep reading the same kind of job: secretaries, clerks, people who move information between forms. Our own rankings put clerical support at the bottom of all nine occupation groups, averaging 47 while every other group sits between 61 and 85. The two measures are pointing at the same part of the economy.

That is the honest summary. At the level of a whole group of jobs, task exposure and the official forecast tell a similar story about office and administrative work. At the level of one job, ours cannot pick the winner, and it was never built to.

What the score is for

It answers a smaller question than “will this job exist”, and it answers it about now rather than about 2034: of the things I do all week, how many are the kind a machine is already good at?

That is a question about your Tuesday, and the answer is useful even when the forecast is cheerful. A growing job whose tasks are being hollowed out is still a job that changes under you: the work shifts, the skills that count shift, and the person doing it has to shift too. Data scientists are forecast to grow by a third, and they score 43. Both of those things are true, and the second one is about what the day feels like.

When someone tells you a number proves your job is safe or doomed, the first question is which kind of number it is. A forecast is a claim about how many. A score here is a claim about what the work is made of. Only one of them needs a crystal ball, and it isn’t ours.

Find your job, and read its task list before its score.