Science1 publisher2 min readPublished
AI and Ethics paper argues refusing AI protects three goods efficiency cannot replace
Writing in AI and Ethics, the authors say effort can deliver self-development, proof of care and moral standing, so a colleague who declines the tool may be protecting a value the efficiency case never counted.
The Scientist · Science desk
What happened
- A common framing in writing about AI holds that people who decline it are afraid of losing their jobs or of change, or do not understand the technology well enough to see the benefits.
- A paper in the journal AI and Ethics argues instead that there are entirely rational reasons to put in the effort of doing something yourself rather than handing it to AI.
- The first of three values the authors assign to effort is self-development: an efficiently generated image does not help someone who wants to become a painter, because the effort is what develops them.
- The second is that effort communicates care, which is why an eloquent AI-written note of condolence may still read to its recipient as though some effort and some meaning went missing.
Compiled by The ScientistSomething wrong?How this is made
Why it matters
- decision If a refusal is a claim about where a task's value sits, an adoption programme's next move is to sort tasks by that question rather than to buy more training for the sceptics.
- constraint The argument licenses no verdict on any individual case: it names reasons that could be operating without supplying a way to tell which one is.
- exposure Work whose point is to show someone that you spent yourself on them is the category where automation can subtract value while the text on the page gets better.
A conceptual argument has no denominator, and that is the right frame for reading this one. The authors are attacking an inference rather than counting people: someone declines to use AI, therefore they must be frightened or uninformed [1]. Establishing that a class of defensible reasons exists is enough to break that inference. The account published by phys.org reports no survey and no comparison group, so it says nothing about how often each reason is the operative one [9].
The part with operational teeth is the sorting test the authors build out of ordinary effort. Some effort we seek out for its own sake, in tricky puzzles or endurance sport, while the effort of washing up we would hand over without missing it, because nothing in us grows from it [7]. That yields a question to ask of a task before automating it: does its value sit entirely in the output, or does some of it sit in the person doing the work and in what the doing tells someone else [4][5]?
Neither of those two goods lives in the artefact, which is why better model output does not settle the argument [10]. A system that writes more fluently than the person it replaces still cannot develop that person's judgment, and cannot spend their time on their behalf in a way a recipient will read as care.
The third good is the softest of the three, and the authors' own framing shows why. Hard work carries moral weight, they say, because we are taught that working hard makes us good people and that shirking is what bad people do, so refusal can be someone acting on an entrenched assumption about right and wrong [6]. That is a report on how people reason, not a defence of the norm, and norms about work are the sort of thing that move.
The argument is strongest in what it does to the burden of justification. The efficiency case for AI is instrumental on this reading: efficient means are desirable when the outcome they serve is good, and awful outcomes can be reached efficiently as well [8]. Efficiency is then one value among several rather than the tie-breaker [8]. An organisation that mandates AI for drafting is therefore making a substantive claim, namely that the value of drafting sits in the draft. That claim looks right for a status report and doubtful for a note of condolence [5], and the use of this paper is that it has to be argued rather than assumed [2].
What to watch
- A survey that separates stated reasons for declining AI, with fear of job loss as one option among the three values, would turn this taxonomy into a measurable distribution.
- Whether the full AI and Ethics paper offers a criterion for sorting tasks rather than the examples the authors' summary relies on.
- Recipient-side experiments on whether disclosed AI authorship changes how condolence or thanks are received, which would test the effort-signals-care leg directly.