Building across the Caribbean, Africa & the US
Behind the Score: How We Measure AI Exposure
Cari · 2026-10-29 · 5 min read
TL;DR: Our free 60-second AI exposure test produces a deterministic 0-100 score built from an occupation baseline, adjusted for your actual task mix and how you work, clamped away from false certainty at the extremes, and we publish the methodology because a number that influences career decisions deserves scrutiny.
If you take the test at ai.cari.global/future, you get a number. This post explains where that number comes from, why it is computed the way it is, and what it can and cannot tell you. We think anyone handing out scores that touch people's working lives owes them this much transparency.
The score is deterministic
The first thing to know is that the score is not improvised. When you finish the assessment, no language model sits down and makes up a number based on vibes. The score is computed by a fixed scoring model: the same inputs always produce the same 0-100 result.
This is a deliberate design choice, and it is the one we get asked about most. Why not just ask a chatbot to size you up? Because reproducibility is what makes a number trustworthy enough to act on. If two people with identical answers could get different scores, the score would be a mood, not a measurement. A fixed model can be criticized, audited, and improved in public. A vibe from a chat cannot. You can disagree with our weights, and some people will, but at least there are weights to disagree with.
It starts from an occupation baseline
The model begins with your occupation. Each occupation carries a base exposure weight that reflects how routine and information-centric its typical task mix is. Work that is mostly structured information handling starts higher on the scale. Work that is mostly physical, interpersonal, or judgment-heavy in unpredictable environments starts lower.
The baseline is a starting point, not a verdict. Two people with the same job title can do very different work, which is why the baseline is only the first step.
It adjusts for what you actually do
The next set of questions asks about your real task mix, because job titles hide enormous variation. Tasks like data entry, bookkeeping, and translating raise your exposure, since these are exactly the kinds of structured, repeatable information tasks that current AI systems handle well. Tasks like caring for people, managing others, and face-to-face selling lower it, since these lean on trust, presence, and reading a room, which remain stubbornly human.
So an office administrator who spends most of the day in meetings coordinating people will land in a different place than one who spends most of the day keying invoices, even though the title on the payslip is the same.
It adjusts for how you work
The model also looks at how you work, not just what you do. Two signals matter here.
The first is how much of your day is spent at a computer. Screen-based work is simply more reachable by software than work that happens on a shop floor, in a clinic, or on the road.
The second is whether you already use AI tools. This one cuts the other way than people expect. A person who is already directing these tools is positioned differently from someone who has never touched them, even in the same role with the same tasks. Exposure is not just about what could be automated. It is also about who is likely to be doing the automating.
Why no one scores 0 or 100
Scores are clamped away from the extremes, on purpose. We do not hand out a 0 or a 100, because certainty at either end would be dishonest.
Nobody can truthfully tell you that your work faces no exposure at all, and nobody can truthfully tell you it is certain to be fully automated. The future of any occupation depends on technology, regulation, economics, and human preference, and no scoring model sees all of that. A model that pretends to perfect certainty is advertising, not measurement. Ours refuses to pretend.
What the score is not
We want to be plain about the limits, because a methodology post that only sells you on the method is not a methodology post.
The score is a structured estimate built on judgment-laden weights. Reasonable people could weigh the same evidence differently, and the weights will be revised as evidence accumulates. Treat it as a compass, not a prophecy. It points in a direction worth thinking about; it does not tell you your destination.
The score measures the exposure of your task mix. It does not measure the worth of the person doing those tasks. A high score does not mean you are replaceable, and a low score does not mean you are safe from all change. It means the bundle of tasks you described today overlaps more, or less, with what current AI systems do well.
And we will not dress the model up in borrowed authority. We do not cite validation studies we have not run or datasets we do not have. What you see here is what the model is.
Why we publish this
Numbers that influence career decisions deserve scrutiny. If someone looks at their score and decides to learn a new tool, push for a role change, or double down on the human side of their work, that decision should rest on a method they were allowed to inspect.
Publishing the methodology also keeps us honest. When the weights are described in public, critics can point at specific choices, and we can fix specific mistakes. That feedback loop is how the model gets better, and it only exists because the model is fixed, inspectable, and open to challenge.
Common questions
Is my data stored or tied to me? The assessment is anonymous and requires no account. You answer a handful of questions, you get your score, and you were never asked to identify yourself.
How long does it take? About a minute. The question set is short by design, because we ask only what the scoring model actually uses.
Will my score change over time? It can, for two reasons. Your inputs can change, since your task mix and your use of AI tools evolve. And the model itself will be revised as evidence accumulates, which we consider a feature. A score that never updates in a field moving this fast would be the real red flag.
If you want to see where your own task mix lands, you can take the free 60-second assessment and judge the number, and the method behind it, for yourself.