UKRI’s AI skills problem is a pathway problem, not a course-count problem
A new study finds gaps in literacy, disciplinary application and responsible use across the UKRI-supported community. Its proposed user typology could turn fragmented provision into navigable pathways.
What happened
Innovation Research Caucus Report 88 combined a literature review, 33 interviews, a doctoral-training consultation and two workshops, then proposed five user types and six system opportunities.
Why it matters
Research funders and institutions need role-sensitive routes that connect AI, domain, ethical and collaborative capability, with evaluation of access and retention.
The UKRI-supported research and innovation community has substantial AI investment but an uneven route from awareness to capable use. Innovation Research Caucus Report 88, published on September 10, identifies gaps in AI literacy, discipline-specific application and responsible and ethical use.
The study ran from October 2025 to March 2026. Its full report documents a literature review, 33 qualitative interviews, an online consultation with Centres for Doctoral Training and two workshops that tested emerging findings. Interviews were concentrated in universities, research councils and publicly funded research institutes, with only a small number of industry representatives. The findings therefore describe recurring needs rather than population prevalence across every part of the UKRI-supported community.
The authors propose five overlapping user types: AI Workers, AI Adapters, AI Developers, AI Leaders and AI Skills Champions. They combine this typology with three knowledge areas—AI tools and technologies, safe and ethical use, and domain knowledge—and with technical and non-technical skills. The report also identifies three broad shortage areas: AI literacy, responsible use, and people able to apply or adapt AI in a domain.
This framing matters because a catalogue of courses is not a development system. Someone judging model output, an engineer adapting a method and a specialist building new systems do not need the same sequence or evidence of competence. Without visible routes, learners must infer prerequisites and institutions cannot see where provision is missing.
Treat the typology as a hypothesis
The report sets out six connected opportunities: resource Skills Champions; curate training and signposting; strengthen responsible-AI guidance; evaluate retention mechanisms; foster inclusive training cultures; and convene people, data and compute. These are system-design recommendations, not measured effects. The qualitative sample is suitable for finding themes, but not for estimating how common each gap is or proving that one pathway model improves outcomes.
A separate Skills England evidence report provides a useful comparison. Drawing on 23 workshops, 10 case studies and a survey of 536 responses, it emphasises practical, reachable, integrated, modular, expandable and sustainable training. That broader evidence supports role-linked pathways, but also shows that navigation alone is insufficient: learning needs realistic tasks, access, governance, reinforcement and outcome monitoring. Its employer survey is not fully representative and its case studies are illustrative, so it does not validate the UKRI typology either.
Make pathways observable
A useful implementation would attach each user type to entry criteria, task examples, risk boundaries, learning options and evidence of proficiency. Institutions could then measure who finds an appropriate route, who drops out, which disciplines remain underserved and whether trained people can complete relevant work safely.
Champions can improve local navigation, but the role needs time, authority and escalation support. Otherwise it becomes an informal helpdesk layered onto existing workloads. Retention needs separate measures: training more people does not solve capability loss if academic pay, career structure or infrastructure drives them away.
Portfolio governance is the connective tissue. A funder can maintain a versioned map of provision, show which user types and disciplines each offer serves, and retire duplicative material when evidence changes. Common assessment patterns can make learning portable without forcing every discipline into one curriculum. Data and compute access should be treated as prerequisites where practical work depends on them.
Evaluation should test whether pathways reduce search time, improve access for underrepresented groups, produce demonstrated proficiency, support safe application and retain technical talent. A larger course count could otherwise increase fragmentation while leaving the same people excluded.
The report's most transferable insight is that AI capability is composite. Funding a tool course without domain judgment, responsible-use practice and collaborative review may increase activity without increasing dependable research. The Skills Atlas can provide a shared vocabulary, but each institution should validate pathways with access, proficiency, application and retention evidence.