AI-Powered Learning

Teaching with AI: When the Technology Challenges Students to Think for Themselves

What if the true potential of artificial intelligence in education lies not in providing answers, but in inspiring better questions? Discover how innovative AI-powered teaching tools at NUS Business School are transforming learning, challenging conventional approaches and empowering students to think critically, make informed decisions and navigate an increasingly AI-driven world.

• AI in Education
• Educational Technology
• Teaching Innovation

AI can score an occupation’s tasks, recommend a statistical test or suggest a compensation decision in seconds. But a fast answer is not necessarily a sound judgement. A model’s assessment may be useful evidence, yet it remains a model judgement—not a measurement of the world.

For Associate Professor Song Zhaoli of NUS Business School’s Department of Management and Organisation, that distinction shapes three teaching initiatives. Rather than asking students to accept an AI-generated conclusion, the tools place evidence in front of them, invite them to challenge assumptions and require them to defend their own decisions.

The projects address three very different learning problems: understanding how AI may change work, designing a coherent pay system and making sense of statistical results. Their shared starting point is the same: students should do the thinking.

1. AI and the Future of Work: Making the future of jobs debatable

What will AI actually change about a job? FOJA—Future-Oriented Job Analysis—starts with the work itself. Students select an occupation, review its published O*NET tasks and inspect a language model’s assessment of each task: whether AI might replace, augment or leave it relatively unaffected. The tool also surfaces its reasoning and the human capability that may remain essential.

Students can disagree. They can override a task rating, examine how importance-weighted Replacement and Exposure measures change, and compare the result with an external AI Occupational Exposure benchmark. They then test different futures through AI capability, adoption, regulation and computing-cost scenarios, before writing three possible job descriptions for the occupation roughly ten years ahead.

From occupations to personal choices: Career Future

The companion app, Career Future, turns the lens towards students’ own plans. It builds an interest and work-values profile, compares fit and gaps against 2,030 Singapore Skills Framework job roles, and adds AI exposure and a 2035 outlook for occupations under consideration. Students can compare career options side by side and export their work as reports.

The aim is not to let a prediction choose a career. It is to help students explain why a direction makes sense—and what they may need to develop to pursue it.

2. FastCat: Learning compensation by building a pay system

Designing compensation requires more than calculating salaries. Decisions about job value, market competitiveness, merit and equity must work together. FastCat gives student teams responsibility for a simulated 250-person company, where they act as its first compensation specialist.

In the first phase, teams group jobs, define compensable factors and evaluate roles. Next, they benchmark salary-survey data, fit a market pay line and create grades and ranges. Finally, they place employees into that structure, allocate merit and bonus budgets, and check for pay gaps. Each phase feeds the next, so an early decision has consequences later.

An AI coach questions saved decisions and offers hints, but is instructed not to produce a team’s deliverable. Instructors can follow progress during the semester. Available in English and Chinese, FastCat also replaces a Microsoft Access-based teaching workflow that Mac users could not run.

3. StatLab for Excel: Making statistical reasoning visible

For students learning introductory statistics, mastering unfamiliar software can distract from understanding the analysis. StatLab brings the work into Excel through a dedicated tab and panel. Students prepare their own data, choose analyses and receive results with assumption checks, effect sizes and APA-style summary sentences.

The initial feature set covers descriptive statistics, chi-square, t tests, ANOVA, Kruskal–Wallis, linear regression and five chart types. A later release adds reliability measures and more advanced models. An optional AI coach helps students choose a test or interpret their last result; an instructor can restrict it to tutor mode.

On R’s classic sleep dataset, its Welch test reproduces the reference result: t(17.78) = −1.86, p = .079. StatLab remains in development and has not been released outside the course.

Across the three initiatives, AI takes on different supporting roles: analyst, scenario generator, coach and interpreter. It does not replace the student’s obligation to make a defensible decision. In FOJA and Career Future, students critique exposure readings and construct plans. In FastCat, teams justify a compensation architecture. In StatLab, they must interpret statistical evidence, not merely produce a table.

These are teaching tools developed and used at NUS Business School, not evidence of measured learning gains. Their significance lies in a consistent instructional design: AI makes complex work more accessible while leaving evaluation, critique and accountability with people.