This semester, I opened an AI tool before I opened a textbook. I used it to untangle difficult concepts, structure my thinking, and navigate research I did not yet understand. So did most of my peers. We did not think twice about it.

But here is what we did not think about: what happens when we graduate into a workforce where employers are doing the same thing — faster, cheaper, and without needing a salary?

The data tells a story we are not ready for.


Chart 1 of 5
Source: QILT Graduate Outcomes Survey 2024, Figure 3. Domestic undergraduate full-time employment rate. Dashed lines mark COVID-19 (2020) and ChatGPT launch (November 2022). Hover for values.

After a record recovery to 79.0% in 2023, graduate full-time employment fell by 5 percentage points in a single year — the sharpest decline since the survey began. The fall arrived exactly as AI tools became embedded in Australian workplaces.

To understand why, we need to look at what those tools actually became capable of in the same period.


Chart 2 of 5
Source: MMLU (Massive Multitask Language Understanding) benchmark. Scores = best-performing model that year across 57 subjects including law, medicine, mathematics, economics and history. Human expert baseline = 89.8% (Hendrycks et al., 2021). Hover for values.
“MMLU tests the same subjects graduates spend years studying: law, medicine, economics, history, mathematics. In 2021, AI failed it. By 2024, it outscored the average human expert.”

This is not gradual drift. In three years, AI went from a failing score to exceeding human expert performance on a graduate-level test. The tasks it now performs well — legal analysis, financial reasoning, scientific synthesis — are precisely what most graduate careers are built around.


Chart 3 of 5 — Multivariate
Sources: QILT Graduate Outcomes Survey 2024 (Table 6); WEF Future of Jobs Report 2025. Employment rates verified from QILT. AI exposure derived by mapping WEF (2025) occupational exposure data to QILT study areas. Hover for details.

Business and Law graduates are currently employed at above-average rates, but face the highest AI automation exposure — their risk is structural, not yet reflected in figures. Creative Arts sits in genuine double exposure: below-average employment AND significant AI risk. Even IT graduates are at just 67.8% — below the national average.


Chart 4 of 5 — Multivariate
Source: WEF Future of Jobs Report 2025, Section 3 (Skills Outlook). Demand change figures are approximate values from WEF Figure 3.4. Skill rankings and directions confirmed by WEF report text. AI disruption level based on WEF (2025) Box 3.1 task displacement ratings. Hover for details.
“The skills declining most sharply are not electives — they are foundations of most undergraduate curricula: writing, numeracy, data entry, and coordination. The skills growing fastest are barely taught at all.”

The skills growing most rapidly — AI literacy, creative thinking, resilience — are often treated as electives. The skills declining fastest — reading, writing, and numeracy — are the ones universities test most rigorously.


Chart 5 of 5 — Multivariate
Sources: QILT GOS 2024 (Table 6: employment by study area; Table 10: % in managerial/professional roles as proxy for skills utilisation). WEF Future of Jobs Report 2025 (AI Exposure = WEF occupational data mapped to QILT study areas). Health skills utilisation is approximate weighted average. Colour scale inverted for AI Exposure — green = lower exposure. Hover for values.

The most striking finding is Law. Law graduates have 79.3% full-time employment — above the national average — yet only 43.8% are in professional or managerial roles. Most are working in jobs that do not use their legal training. Add a 52% AI exposure for legal tasks, and the picture sharpens: Law graduates are employed, but not in law, and AI is about to make that harder.


What This Means

This is not a story about whether AI is good or bad. It is a story about preparation, timing, and honesty.

Three things need to change.

1. Curricula must shift toward skills AI cannot replicate. Critical thinking, ethical reasoning, interpersonal communication, and AI fluency are not soft skills. In the next decade, they will be the hard ones.

2. Students in high-risk fields deserve honest information. A student choosing Law, Creative Arts, or Commerce in 2026 is not choosing a safe path. They have a right to make that choice knowing what the data shows.

3. Australia needs a national conversation about graduate preparation for an AI economy. The data exists. The research is clear. What is missing is the urgency to act on it.

The graduates who will thrive are not the ones who ignored AI — they are the ones who understood what it can and cannot do, and built their careers around the difference.


References

Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., Agarwal, S., Herbert-Voss, A., Krueger, G., Henighan, T., Child, R., Ramesh, A., Ziegler, D. M., Wu, J., Winter, C., … Amodei, D. (2020). Language models are few-shot learners. Advances in Neural Information Processing Systems, 33, 1877–1901. https://arxiv.org/abs/2005.14165

Chowdhery, A., Narang, S., Devlin, J., Bosma, M., Mishra, G., Roberts, A., Barham, P., Chung, H. W., Sutton, C., Gehrmann, S., Schuh, P., Shi, K., Tsvyashchenko, S., Maynez, J., Rao, A., Barnes, P., Tay, Y., Shazeer, N., Prabhakaran, V., … Fiedel, N. (2022). PaLM: Scaling language modeling with pathways. arXiv preprint arXiv:2204.02311. https://arxiv.org/abs/2204.02311

Gemini Team. (2023). Gemini: A family of highly capable multimodal models. arXiv preprint arXiv:2312.11805. https://arxiv.org/abs/2312.11805

Hendrycks, D., Burns, C., Basart, S., Zou, A., Mazeika, M., Song, D., & Steinhardt, J. (2021). Measuring massive multitask language understanding. International Conference on Learning Representations (ICLR 2021). https://arxiv.org/abs/2009.03300

OpenAI. (2023). GPT-4 technical report. arXiv preprint arXiv:2303.08774. https://arxiv.org/abs/2303.08774

Quality Indicators for Learning and Teaching. (2024). 2024 Graduate Outcomes Survey national report. https://www.qilt.edu.au/docs/default-source/default-document-library/2024-gos-national-report.pdf

Quality Indicators for Learning and Teaching. (2024). 2024 GOS national report tables [Data set]. https://www.qilt.edu.au/docs/default-source/default-document-library/gos_2024_national_report_tables.zip

World Economic Forum. (2025). The future of jobs report 2025. https://reports.weforum.org/docs/WEF_Future_of_Jobs_Report_2025.pdf

Anthropic. (2025). Claude (claude-sonnet-4-6) [Large language model]. https://claude.ai

RMIT Library. (2024). Artificial intelligence (AI): Acknowledgement and referencing guidelines. https://rmit.libguides.com/referencing_AI_tools

Acknowledgement of Generative AI Use

Generative AI tools (Claude, Anthropic) were used in the preparation of this assignment to assist with the initial structuring of the narrative framework, R code suggestions for Plotly tooltip formatting and interactivity, and early-stage outline drafting. All data sourcing, analytical interpretations, chart design decisions, visual encoding choices, narrative content, and final R code were developed and verified independently by the author. All data values were sourced directly from the cited primary sources. AI assistance is acknowledged in accordance with RMIT Library guidelines (RMIT Library, 2024).

Additional acknowledgements: The World Economic Forum’s Future of Jobs Report 2025 data explorer was used to cross-check skills demand figures cited in Chart 4. The Stanford HAI AI Index 2024 interactive data tools were used to verify benchmark performance trends cited in Chart 2.