Can a two-hour workshop teach future teachers AI skills?
Spanish trainee teachers who attended a coding, robotics and machine-learning workshop reported stronger familiarity and attitudes than a separate control group. With no pretest or performance task, the study cannot show that the workshop caused durable skill gains.
The short version
- What happened
- The quasi-experiment included 144 first-year trainee primary teachers at three Spanish public universities: 105 attended the workshop and 39 at another university formed a non-equivalent control group.
- Who it affects
- Trainee teachers may gain confidence from a short practical introduction, but they still need assessed and sustained development before consequential classroom use.
- What remains unproven
- No pretest, random allocation or active comparison condition was used.
Keep in mind
No pretest, random allocation or active comparison condition was used.
All outcomes were self-reported immediately after a two-hour intervention; no objective competence or retention was measured.
The 105-person workshop and 39-person control groups came from different universities and were unequal in size.
The machine-learning component was introductory and did not involve hands-on model training.
Exploratory learner profiles were not externally validated and should not be treated as stable types.
Source analysis · 1 linked source. Primary analysed source date: 13 October 2026.
Read the full report · 7 min The evidence, methods, implications and limitations
The answer: the workshop changed reported perceptions, not proven competence
First-year trainee primary teachers who completed a two-hour hands-on workshop reported higher knowledge, familiarity and enthusiasm than students in a separate control group. The largest group differences concerned familiarity with block-based programming and game engines and views of coding and robotics as active-learning tools; each produced an effect size of r=0.50. Self-reported knowledge of programming and machine learning differed more modestly, at r=0.29. The results make a short, practical introduction worth testing in teacher-education courses that struggle to create space for digital skills.
They do not demonstrate that two hours taught usable AI competence. Students were measured only after the workshop, the groups came from different universities and nobody completed a programming, debugging or machine-learning performance task. The control group followed its usual course rather than an equally engaging alternative activity. Differences could therefore reflect earlier experience, institutional context, selection, novelty or active participation, as well as the workshop itself. The paper is most informative as a feasibility signal about attitudes and perceived familiarity—not evidence that teachers became ready to evaluate or teach AI.[1]
Who took part and what they did
The sample comprised 144 first-year students on primary-education degrees at three Spanish public universities. Two universities supplied the 105-person workshop group and a third supplied the 39-person non-equivalent control group. Women made up 90.5% of the workshop group and 82.1% of the control group, reflecting the gender pattern of the participating courses. Most students were aged 20 or younger. Participation was voluntary, anonymous and unpaid, and the researchers report ethics approval and informed consent.
Workshop students rotated through three practical stations using Scratch, educational robotics and MakeCode, with challenges intended to introduce computational problem solving. The machine-learning component operated at an awareness level; students did not train a model and the study did not directly assess ML skills. That distinction matters for readers: exposure to the idea that examples can shape a model is different from being able to test data quality, evaluate error across groups, challenge automated output or design a safe classroom exercise.[1]
How the researchers measured the result
Participants completed a 20-item questionnaire on a four-point scale. Five items covered perceived programming and machine-learning knowledge, six covered block programming and game engines, and nine covered active-learning applications such as engagement and inclusion. Eleven experts reviewed the items. The full instrument had high reported internal consistency, with Cronbach's alpha of 0.911, and an exploratory factor analysis found a three-factor structure explaining 68.4% of variance. These checks show that responses formed a coherent questionnaire; they do not transform perceptions into an objective skills test.
Because scores were non-normal and the groups were unequal, the authors used Mann–Whitney U tests. Workshop participants scored higher on all three dimensions. For programming and ML knowledge, U was 2,821.5 with p=0.00046 and r=0.29. For block programming and game engines, U was 3,373.5 with p below 0.00001 and r=0.50. For active learning, U was 3,380.0 with p below 0.00001 and r=0.50. Statistical significance says the observed rank differences are unlikely under the test's null model; it does not repair the non-randomised design or show long-term learning.[1]
Why the missing pretest changes the conclusion
A pretest would have shown whether the groups began at comparable levels and how each person's answers changed. Without it, a higher post-workshop score might reflect the intervention, but it might also reflect different students or universities. Random assignment was not used, and access and curriculum alignment determined which institutions joined each condition. The workshop group also received a structured active-learning experience while the control group received no alternative session, so the study cannot separate coding and AI content from the broader effect of doing something practical and engaging together.
The article identifies reluctant, intermediate and enthusiastic response profiles among workshop participants, but the clusters are exploratory and context-dependent. They were derived from the same self-reports and were not validated against later behaviour, classroom practice or academic performance. Labels can help instructors think about differentiated support, yet they should not become permanent categories attached to trainees. A student who reports low confidence after two hours may need time or prior knowledge, not a remedial identity.[1]
What the result means for teachers and schools
For teacher educators, the practical lesson is that a brief workshop can fit inside an existing course and may make unfamiliar tools feel more approachable. It can provide a shared vocabulary before deeper work on lesson design, assessment, bias, privacy and appropriate use. The study also found comparatively limited familiarity in specific technical areas despite favourable overall attitudes. That gap matters: enthusiasm can make a teacher willing to experiment, but safe classroom use requires enough understanding to recognise when a system is unreliable or inappropriate.
For schools, the paper does not justify replacing sustained professional development with a single session. Future teachers will need repeated practice, feedback, subject-specific examples and opportunities to examine real model errors. They also need routes to opt out of tools that expose children's work or data without proper governance. A short introduction may be the first rung of a programme, followed by assessed tasks and supervised classroom application. The workshop was funded through a Spanish Ministry of Science, Innovation and Universities project; the authors declared no conflict of interest.[1]
What would change the assessment
A stronger trial would measure participants before and after training, randomly allocate them where feasible, and give the comparison group an equally active workshop on another topic. Objective tasks should test whether trainees can create and debug a simple programme, explain how training examples affect a classifier, identify a misleading output and design a classroom activity with safeguards. Follow-up after a term or teaching placement would show whether any gains persist and transfer into lesson planning or practice.
Larger samples across countries, age groups and teacher specialisms would help test generalisability. Researchers should report prior experience and access to technology, because a workshop may widen gaps if confident students benefit most. Measures of workload, accessibility and classroom inclusion would also make the result more useful to institutions deciding where to invest. Until such evidence exists, the careful conclusion is that this Spanish microintervention was associated with more positive post-session self-reports. It is a credible starting point for programme design, not a certificate of AI readiness.[1]