AI’s gains in university-level reasoning tests are sharpening the question of what graduates need to bring to work. Law lecturer Armin D. Ashenberg argues universities should teach students to direct AI, spot its mistakes and improve its output, while keeping the subject knowledge needed to judge the result.
Watch Desk analysis
What happened
In an article republished by Silicon Republic, Ashenberg points to research showing that AI model scores on university criminal law exams rose from an average of 52.5% in 2023 to 76.3% in 2025. In that later study, models on average scored above an average university student on critical-thinking questions. He also cites a 2024 reasoning study in which newer AI models performed better than earlier ones.
Those are results from specific tests, not proof that AI can replace an entire profession. Ashenberg also describes a 2026 pilot in which healthcare professionals, mostly nurses, accepted half of deliberately wrong AI recommendations and rejected about one in six correct ones. The examples point to two different challenges: systems are improving at some assessed tasks, while people still need the judgement to use their advice well. Read the article.
Why it matters
A degree may need to prepare students not only to do work themselves, but to judge and improve work produced with AI. That is a more demanding brief than simply adding an AI tool to a course and hoping for the best.
The article also notes that a Reserve Bank of Australia assessment has found early signs of difficulty for younger workers in industries more exposed to AI. That is a signal worth watching, not evidence that AI alone caused graduate jobs to disappear. Ashenberg says some graduate roles have been cut or tasks shifted to AI, while consulting firms have attributed cuts to other reasons.
Our read
The useful takeaway is not that graduates should panic or that universities can solve this with a prompt-writing workshop. Students need enough grounding in their discipline to recognise when a system is wrong, and enough practice working with AI to make its output better. The awkward bit is that this may require more learning, not less.
What to watch
- Whether universities build AI use and critical evaluation into subject teaching, rather than treating them as bolt-on skills.
- Whether further research finds similar changes in exam performance and workplace outcomes.
- Whether graduate hiring changes can be linked to AI use, rather than simply arriving alongside it.
Discussion spark: Should universities make AI collaboration a core skill in every degree, or risk short-changing the deep subject knowledge needed to check its work?
Sources and evidence
- Source update (19 November 2026, 13:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.