Gabriella gives the teams an authentic task so that they use knowledge rather than just receiving it. She provides customised generative AI to enhance the students’ academic skills, monitoring its use and giving feedback. Group work fosters a sense of belonging and a high level of engagement.

Teaching area

Lectures and workshops

Course

MSc Development Economics core module: Macroeconomic Analysis. Workshop numbers: 36 in 2020/21, the first year of TBL, and 18 in 2025/26 (total 117 students in six years of TBL). Mostly international students; around a quarter of the students are from the UK (mainly white). Approximately half are female.

Description

1 Preparation: flipped classroom

I upload the lectures, recorded lecture videos and a set of three homework exercises linked to the lectures. I then ask AI to generate a customised study guide aligned with the topic of the week, the lecture and the homework tasks.

For the application exercises in the workshop, I create an ad hoc custom AI. This AI is designed to ask probing questions, highlight conceptual gaps, request clarification and act as a sceptical expert. It probes students and encourages critique rather than passive acceptance.

In an early workshop, I instruct the AI to make a mistake to test whether the students identify it. This approach also prompts students to question AI outputs more generally, as they cannot assume that responses are always correct – even within a controlled or ‘protected’ environment such as a custom-built AI, which is designed to reduce hallucinations. Students therefore learn to treat AI-generated content as a starting point for analysis rather than as the authoritative truth.

2 Readiness assurance: flipped classroom

In class, I organise the groups to ensure diversity of gender and background, and a balanced mix of prior academic performance (based on Term 1 results), so that no team has a systematic advantage.

Students first complete an individual online multiple-choice quiz based on the lecture material. I do not reveal the answers at this stage. They then retake the same quiz in their groups using scratch cards, where the correct answer is revealed immediately by a star. Groups compete with one another, which increases engagement and motivation.

Students quickly realise that they perform better when they collaborate (Cagliesi & Ghanei 2022). At times, I intentionally include a problematic or ambiguous answer. Groups may raise a formal challenge by indicating which alternative answer they believe is correct rather than the one revealed by the scratch card. If their challenge is justified, they receive additional marks.

I collect the scratch cards, review any contested questions, provide a mini-lecture to clarify common misconceptions and mistakes, and address group queries. In some cases, I subsequently revise my slides to improve conceptual clarity.

3 Apply concepts: workshops

The weekly workshops deepen learning after students have consolidated their core knowledge. Groups solve significant problems that they have not encountered before, often based on real-world scenarios or theoretical extensions of the topic. All group work is completed in class. Students discuss alternative approaches, justify their reasoning and agree on a collective solution.

I act as a facilitator, circulating to provide targeted guidance. I introduce AI in the third workshop, once group dynamics are established. In each workshop, I use a different custom AI for one or two of the application exercises. I actively monitor how students use it. At the end of each session, groups submit their AI interaction logs. I provide feedback on their reasoning and the quality of their AI engagement. This process has strengthened their critical thinking and improved how effectively they use AI as a tool.

At the end of each week, I circulate a ‘lessons learned’ document summarising key conceptual insights, common errors, workshop reflections and observations on AI use.

4 Grading

The students are graded for teamwork and for individual contributions. Individual grades reflect attendance, contribution to team tasks and anonymous peer evaluation (students give each other feedback on their effort, which contributes to their individual grade).

Case study: Team-based learning with AI

As I’ve become more confident in designing and constraining custom AI tools, I’ve progressively introduced more demanding AI-integrated assessments. My first use of AI in assessment (2022/23) adopted a deliberately cautious, assistive approach. I began with structured critique before gradually advancing towards model construction and forecasting in later iterations.

In the initial year, students were given a macroeconomic prompt alongside a response generated by a general AI model. Their task was to critique the reasoning, identify conceptual weaknesses, and improve both the argument and its underlying assumptions. The emphasis was on critical evaluation rather than production.

In 2024/25, I introduced a custom-AI reverse-engineering challenge that used AI in a more augmentative role. Students were provided with synthetic datasets simulating a contemporary macroeconomic debate. They evaluated whether the data plausibly represented real-world dynamics, inferred the prompts that might have generated the dataset and iteratively refined their own prompts to reproduce a comparable dataset, documenting each refinement step. Finally, students used their reconstructed datasets to produce three forecasts (pessimistic, neutral, optimistic) and propose policy responses, testing macroeconomic reasoning and forecasting logic.

The progression moved from critiquing AI outputs to using AI as a modelling tool, requiring students not merely to use AI, but to interrogate its assumptions and limitations. Through the process, they learnt to evaluate the strengths and limitations of using AI-generated data for forecasting and decision making in macroeconomics and economic research.

Outcomes

The integration of role-based custom AI within TBL produced several positive outcomes.

Academic skills and student support

Deepened conceptual understanding

Student feedback indicates a stronger grasp of complicated macroeconomic concepts, particularly in modelling, data simulation and policy interpretation. Students reported that AI prompts pushed them to explain assumptions and refine arguments rather than simply retrieve information.

Enhanced critical thinking

Students highlighted that AI challenged their assumptions and encouraged them to question outputs rather than accept them at face value. The deliberate use of AI error and probing questions fostered scepticism and analytical reasoning.

Improved assessment preparation

Students found AI particularly helpful for essay planning, argument structuring and synthesising macroeconomic data. They reported greater confidence in linking theory, evidence and policy.

Development of AI literacy

Students became aware of prompt sensitivity and the importance of precision in shaping AI responses. They recognised both the strengths and limitations of AI-generated content and articulated that AI cannot replace critical thinking.

Fostering sense of belonging

The TBL structure fostered strong group cohesion and a sense of belonging. Peer evaluation comments and observed classroom behaviour indicated loyalty to teams and shared responsibility for performance. Students recognised that individual preparation directly influenced collective outcomes. The integration of AI within teams appeared to reinforce this dynamic. Because AI responses required collective interpretation and refinement, students engaged in deeper intra-team discussion, particularly when the custom AI was designed with a limited number of questions each team could ask. This constraint encouraged deliberate consultation before interacting with the AI. AI interaction logs showed evidence of collaborative reasoning, joint critique and shared prompt refinement. In this sense, AI did not replace teamwork; rather, it acted as a catalyst for more structured and reflective group dialogue.

Student engagement

• Attendance remained as high under the TBL structure as it was before, and peer evaluation reinforced individual accountability.
• In 2024/25, TBL produced higher submission rates, higher averages and lower standard deviations – or a combination of these – in almost all cohorts.

Comments

There are identifiable risks in integrating AI into classroom practice. AI-generated responses can create false confidence because outputs often appear coherent and authoritative. Without careful structuring, there is a risk of cognitive outsourcing, where students rely on AI rather than engaging critically with the material.

AI also tends to over-explain and default to lecture-style responses. Without explicit guardrails, it can reduce productive struggle and shorten the reasoning process. For this reason, behavioural limits were embedded within custom AI roles to prevent over-explaining and passive consumption. Students were required to articulate reasoning before receiving feedback and to submit AI interaction logs for review.

The experience suggests that AI integration requires intentional design rather than open access, explicit guidance, guardrails and reflective practice. When properly designed, AI becomes a pedagogical partner rather than a shortcut.

Student feedback

Students have highlighted the benefits of the TBL approach:

Learning under the TBL pedagogy really helped improve my team working skills and communications, and while elements of the course were very challenging, with collaborative efforts we succeeded.

The Team Based Learning programme has been a wonderful experience to learn, work in a team, interact with the class and teachers, present your work etc.

(Student quotes from Cagliese and Ghanei 2022).

Reflections

• Integrating AI into TBL has been iterative and time-intensive. Customising AI for each session required sustained preparation, testing prompts and managing AI ‘drift’ – its tendency to revert to over-helpful explanations or initiate new directions.
• AI literacy cannot be assumed. In early sessions, some students initially accepted incorrect AI outputs without sufficient scrutiny. This reinforced the need to explicitly teach scepticism and embed structured constraints.
• Over time, tightening behavioural rules (limiting explanations, restricting the number of AI questions and enforcing reasoning-first responses) shifted AI from an answer provider to a structured dialogue partner. The experience confirmed that the effectiveness of AI depends on disciplined pedagogical design rather than technological novelty.

Links

Cagliesi, C. and Ghanei, M. (2019) Can We Mend the Gap? Team Based Learning (TBL): Developing Diversity into Inclusivity, Advance HE (accessed 26 March 2026)

References

Bowen, R. S. (2017) ‘Understanding by Design’, Vanderbilt University Center for Teaching (accessed 26 March 2026)

Cagliesi, G., and Ghanei, M. (2022) Team-Based Learning in Economics: Promoting Group Collaboration, Diversity and Inclusion, The Journal of Economic Education 53.1: 11–30 (accessed 26 March 2026)

Michaelsen, L. K. (2002) ‘Getting Started with Team-Based Learning’, in L.K. Michaelsen, A.B. Knight and L.D. Fink (eds.), Team-Based Learning: A Transformative Use of Small Groups, Sterling, Virginia: Stylus Publishing

Wiggins, G., and McTighe, J. (2005) Understanding by Design, expanded 2nd ed., Alexandria, Virginia: Association for Supervision and Curriculum Development