Recently I had the opportunity to present on one of the most pressing questions facing higher education today: what does the rise of generative AI mean for how we produce knowledge, uphold academic integrity, and teach students to think critically?
At the heart of this discussion is the politics of knowledge production. The tools we use to find, generate, and assess information are never really neutral – they reflect the biases, privileges, and power dynamics embedded in their design. This has always been true of knowledge production but GenAI makes these dynamics harder to spot and easier to scale in ways that risk obscuring truth and perpetuating bias.
Identifying bias
One of the clearest illustrations of this is how AI image generators reproduce social stereotypes. Research into tools like Midjourney found some troubling patterns: all images returned for terms such as “journalist,” “reporter,” or “co-correspondent” exclusively featured light-skinned people, a trend described as “racial hegemony built into the system,” likely reflecting a lack of diversity in both training data and the AI industry itself, while also producing images that reinforced gender bias in terms of the types of professional roles occupied by men and women.
There have also been concerns raised from Indigenous communities, about the risks of GenAI co-opting sacred knowledges and impeding upon Indigenous knowledge sovereignty. Recently, researchers found that prompts to include visual data of Aboriginal Australians surfaced deeply concerning images, often with regressive visuals of “wild,” “uncivilised,” and sometimes “hostile native” tropes. More broadly, AI technologies fuel the spread of misinformation about Indigenous people, often conflating distinct communities with other global Indigenous peoples and drawing on inappropriate or non-Indigenous sources to do so.
For academics and educators, this raises urgent questions about bias and ethics, particularly when students use GenAI to research, illustrate, or generate content, they may unknowingly encounter and reproduce these perspectives.
Prompting as a critical skill
A practical response (though not a complete solution) is developing stronger prompt engineering skills which some have called “micro managing” AI. An example of this is MIT Sloan’s guidance on effective prompting describes this as “programming with words”, emphasising that the quality of AI output hinges largely on how questions are framed. Strategies include providing rich context, being specific about the task and audience, and building on the conversation across a session rather than treating each prompt as isolated. This kind of careful, deliberate engagement doesn’t eliminate bias or inaccuracy, rather, it can support the cultivation of critical reading and thinking skills that remain central to academic work.
Institutional gaps
There is a growing need for clear, principled institutional guidance on GenAI in ways that are not focused on policies such as plagiarism detection, but instead forms of applied frameworks for choosing tools ethically and understanding their limitations. Some universities have developed comprehensive approaches. For example, Purdue University offers detailed library guides helping students and staff navigate AI tools with academic integrity in mind.
As GenAI tools continue to evolve rapidly, many staff are proactively exploring and experimenting with different tools on their own terms. This type of grassroots engagement has value in building collective knowledge and capability across the institution. To support this, external resources can be incredibly useful, including benchmarking tools like LM Council, which allow academics to compare the efficacy of different LLMs and identify which are best suited to their specific teaching and research purposes. As institutional guidance continues to develop, these kinds of resources can help staff make more informed, ethically grounded choices in the meantime.