When Does Collaboration Become Compulsion?
Imagine an environmental team considering a new AI system for detecting an invasive species. The case for using it is easy to make. There is too much land to survey regularly, too many species with invasion potential, too little funding, and a growing need to identify new populations before they become established. An automated system could process thousands of photographs or audio recordings, alerting researchers to signs that we otherwise might not encounter until much later.
At first, the system is an additional source of information. Fieldworkers compare its classifications with their own observations, question surprising results and decide when its recommendations are useful. Gradually, however, the infrastructure around the technology changes. Monitoring practices are redesigned to produce the data that the models need. Funding is directed towards extending coverage. Reports adopt its outputs, while management plans begin to depend on its predictions. Employing more field staff becomes harder to justify when automated monitoring is already available.
No single decision has made AI compulsory. Nevertheless, refusing it now would mean dismantling practices, expectations and investments built around its use.
When organisations talk about responsible (or “ethical”) AI, attention often turns to how AI systems can be improved and governed: how their outputs can be made more reliable, their workings more transparent, their risks reduced, and their use subject to human oversight. These are important questions. But they can quietly assume that the decision to use AI has already been made, neglecting the earlier question of whether AI belongs in a particular activity at all.
A person may be “in the loop” while having little power to change the loop.
The increasingly salient idea of “human-AI collaboration” in AI Ethics discussion risks concealing this shift. Collaboration sounds voluntary and relatively equal. It suggests a human choosing to work with a tool while remaining free to shape the relationship. Yet many encounters with AI take place within institutions that have already selected the system, defined its purpose, and established what counts as “legitimate” or “successful” use. A person may be “in the loop” while having little power to change the loop.
Similar pressures appear in universities. Researchers are encouraged to adopt AI in order to work more quickly, process more material, or demonstrate methodological innovation. Students encounter it in assignments, writing, searching, and study support. Staff are often told that automation may relieve workloads that are becoming unsustainable. These uses are not necessarily harmful, and some may be genuinely valuable, but the difficulty arises when the availability of AI begins to reshape what institutions expect people to achieve.
Across many industries, as we’ve seen from the rise of computation and automation over the past century, increased capacity is rapidly followed by increased expectations. If a task is assumed to be faster with AI, workloads can expand accordingly. If automated analysis permits larger datasets, smaller and more interpretive forms of inquiry may appear insufficiently ambitious. If AI-assisted writing becomes the norm, choosing to write without it may be treated as uncommon personal preference rather than a legitimate way of thinking and honing skills through language. What begins as an optional convenience can alter the standards against which everyone is judged, including those who would prefer not to use it.
This is why the right to refuse AI deserves a place within debates about AI ethics. By refusal, I do not mean rejecting every use of AI, nor imagining that pre-AI practices were necessarily better or neutral or fair. Refusal can be selective and situated. It might mean deciding that a particular system is inappropriate for a particular task, that its costs outweigh its benefits, or that something valuable would be lost by allowing it to mediate an activity. A field ecologist might value an AI-assisted survey while resisting replacement of long-term observation work. A researcher might use software to organise material but decline to delegate interpretation. A student might experiment with an AI tool and still want spaces in which their writing develops without it. An employee might accept the relevance of AI for a task while objecting on ethical grounds to the environmental impact of the available tools. These positions are attempts to negotiate where AI should and should not be given a role.
For refusal to be meaningful, however, it must be more than an opt-out written into a policy. People need to be able to question an institution’s choices without being characterised as inefficient, fearful, or opposed to innovation. Most importantly, they should have some influence before AI becomes embedded in the infrastructure of their work, while its purpose, boundaries, and desirability remain genuinely open to discussion.
The resulting shift in responsibility away from the individual alone is key. It is not enough to tell people that they remain free to disregard an AI system after an organisation has structured its resources, workloads, and expectations around using it. Institutions that introduce AI should be prepared to explain not only whether it works, but why it is needed, what practices it may displace, and how those who decline it will be protected from disadvantage.
AI may help environmental researchers detect species sooner, help academics navigate large bodies of material, and help students approach difficult tasks. But the value of these possibilities should not make adoption inevitable. If collaboration is to be an ethical relationship rather than a polite name for accommodation, it must leave room for hesitation, negotiation, and limits. Sometimes acting responsibly with AI may begin with asking how it should be used. At other times, it may begin with preserving the possibility that it should not be used at all.
