Where does the limit of our consent to automation lie?

A recent study from Harvard Business School, by James Riley and Simon Friis, makes us pause over a surprisingly simple question – albeit only on the surface. How much work are we willing to hand over to machines? And what actually determines where we draw the line?

The results – based on responses from 2,357 people asked about 940 occupations – are, at first glance, reassuring for technology advocates. Americans declare acceptance of automation for about 30% of professions today, and when they are asked to imagine artificial intelligence more advanced than today’s – cheaper and more efficient than a human – this percentage almost doubles, to 58%. Almost all respondents, 94–96%, endorse the use of AI as a tool that supports human work.

And yet, these numbers hide something more interesting than statistical acceptance. Riley points out that this jump – from 30% to 58% – suggests that our resistance to AI is largely not about value, but about technical feasibility. We doubt whether the machine can really do it, rather than whether it should try. When we assume that it will do better, a significant share of the objections disappears.

But what about the rest? About 42% of professions remain in the sphere of ambivalence – neither unambiguous agreement nor opposition. And 12% of jobs evoke what the authors call a moral disgust toward automation. These are roles that – regardless of the AI’s capabilities – we would like to preserve for people: clergy, childminders, artists, athletes, funeral directors. One might ask: what do they have in common? What makes something stronger than cost accounting stir within us here?

Perhaps it is that some tasks cannot be reduced to their result. It also matters who performs them and how. A funeral requires presence, not just service. Raising a child is not providing a service. Art, as Riley notes, can be valuable not because it was created, but because someone created it. The value of certain things lies in the human gesture itself, not in its effect. This is a surprisingly difficult thought – because it forces us to recognise that not everything that can be measured by effectiveness can be reduced to effectiveness.

The authors recall historical parallels: genetically modified food, nuclear energy, stem cell research. All potentially useful. All of them encountered social constraints that pure economic rationality could not disarm.

What does this mean for organisations that are already planning AI implementations today? Riley suggests caution: start where utility is obvious and resistance is low. Be open about where and how the technology is used. And remember that in some industries, “cheaper and faster” can simply reduce the value of the product in the eyes of the customer.

The boundary we set is not fixed. But it is worth asking now where it runs – and what we actually want to keep for ourselves on the other side of it.

Read more here: https://www.library.hbs.edu/working-knowledge/riley-performance-resistance-2026?utm

This post is part of the project “People and Algorithms in Organisations: Competences to Work in the Digital Environment” (DIGIT_People and algorithms, funded by the NAWA – Narodowa Agencja Wymiany Akademickiej (Polish National Agency for Academic Exchange).