Preparedness for artificial intelligence (AI) is everywhere in policy right now: in government departments, in workplaces, in schools. But most of that attention has gone into building the tools and getting people to use them, rather than asking what skills people need to use them critically and confidently.
One of those skills that has been largely overlooked is numeracy.
Last week we launched Count on It: AI, Numeracy and Social Mobility, commissioned by the National Numeracy Leadership Council. The research combined desk research, interviews and a focus group with business and sector leaders, and a nationally representative survey of over 2,000 UK adults conducted by Ipsos. Here is what we found about the role numeracy plays in AI literacy, and what happens if it continues to go unrecognised.
The numeracy behind AI literacy
Functionally, using AI is relatively simple. For most people, if you can send a text message, you can write a prompt. But what’s sometimes much harder is judging what comes back. AI tools can be fluent and assured without always being accurate, and that gap is not always visible to the person reading the output. Even where an output is accurate, it will often tend to confirm what someone already believes rather than challenge it, meaning it can reinforce a person’s existing view without them realising it.
Recognising that gap depends on a basic technical understanding of how these tools work. They generate responses by identifying statistical patterns in their training data, not by verifying facts. Assessing whether an output is plausible, noticing when a figure looks wrong, or recognising that a response reflects likelihood rather than certainty are, in practice, numerical judgements. They draw on a sense of scale, an understanding of proportion, and basic statistical literacy.
That connection is not widely recognised. Among survey respondents aware of AI tools, just 25% agreed that good numeracy skills are needed to use them, compared with 42% who said the same of literacy.
Who stands to lose out
Our interviews pointed to two groups facing particular risk. The first are those using AI without the numeracy needed to question what it produces. They may accept outputs at face value in decisions that matter, including finances, health information or employment. Among survey respondents who had used AI tools, 77% considered them trustworthy on accuracy, a level of confidence at odds with what is known about how often these tools get things wrong.
The second group is aware of AI but not using it, often due to low confidence. Awareness in our survey stood at 94%, but only 53% had actually used AI tools. Both groups skew towards respondents in lower socioeconomic grades. 70% of respondents with a higher socioeconomic grade (AB) had used AI tools, against 36% of respondents with a lower socioeconomic grade (DE).
There is also a question of what happens over time, as AI becomes part of how people handle everyday numerical tasks. The evidence here is still emerging, but stakeholders in this research described a possible cycle. Relying on AI to complete numerical tasks, rather than working through them directly, may weaken the skills needed to evaluate what AI produces. That could reduce the ability to catch errors, increase reliance on the tool, and further limit the development of those skills.
Neither of these risks is solved by expanding access to AI, or simply encouraging more people to use it. Doing so will not close these gaps on its own.
Building skills that last
The report makes five recommendations. The central ask is for government to recognise numeracy as a foundational condition for AI literacy, naming it in AI strategy and funding decisions, and to commission a funded successor to the Multiply adult numeracy programme with AI-relevant numeracy built in from the outset. The remaining recommendations focus on education. They include naming critical AI literacy and applied numeracy in the curriculum and the new digital V-level, setting procurement requirements so that AI tools used in schools support active numerical engagement rather than substituting for it, and equipping teachers, across the existing workforce and the training pipeline, to deliver all of this.
At our launch event in the House of Commons last week, one point came through clearly in the discussion. A population confident in numeracy is better placed to navigate whatever comes next, regardless of which tools or policies define the moment. Building that foundation, rather than reacting to each new technology as it arrives, is what will let people adapt as AI itself continues to change. That is the case this report makes.
Read the full report here: Count on It: Numeracy, AI, and Social Mobility
If interested in contributing to future Education and Skills policy, please contact Rhiannon Tuckett-Jones at (Rhiannon.Tuckett-Jones@policyconnect.org.uk).
Related content
