Recommended Free Tools
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
In one randomized experiment involving more than 1,300 teachers in Greece, teachers were more likely to accept an unfairly harsh grade when it was attributed to AI than when it was attributed to another teacher. That is evidence of deference in one grading scenario—not proof that teachers generally trust AI more, or that they accept every AI recommendation.
What did the Greek teacher experiment find?
Monash University’s 20 August 2026 release describes a randomized experiment in which more than 1,300 teachers in Greece evaluated a grading scenario. Teachers were more likely to accept a harsh grade when told AI had assigned it than when told another teacher had assigned it. The result points to a risk: an unfair AI output may go unchallenged even when a teacher has a reason to question it. Monash University’s account of the study does not provide the full methods, effect sizes, or a complete explanation of how unfairness was defined, so the size and boundaries of the effect cannot be assessed from the release alone.
The same release reports that 48 per cent of teachers in the Greek study sample used AI tools at least weekly for lesson preparation, while 16 per cent actively encouraged fellow teachers to adopt the technology. These are study-reported figures from that context, not estimates for teachers everywhere. Monash also reports that younger teachers, teachers with postgraduate qualifications, and teachers who described themselves as technologically confident were least likely to challenge the harsh AI grade. The release gives no subgroup effect sizes and does not establish why those groups were less likely to object.
Does this mean teachers trust AI more than people?
No single study establishes a general ranking of trust. The studies measure different behaviors: whether someone challenges an unfair grade, asks for advice, follows a recommendation, or becomes more willing to use a tool after training. Those outcomes are related, but they are not interchangeable measures of trust.
#1 Best Overall
| Study | Participants and task | What it measured or reported |
|---|---|---|
| Monash University, 2026 | More than 1,300 teachers in Greece; a grading judgment involving a harsh grade attributed to AI or another teacher. | Teachers were more likely to accept the harsh grade when it was attributed to AI. The university release does not report effect sizes or full methods. Source |
| Kaufmann and Budescu, 2020 | 435 middle- and high-school teachers in three online experiments; computerized expert advice about student profiles. | Participants requested and followed computerized expert-model advice less often than human advice. This was not a direct test of generative AI grading. Source |
| Nazaretsky et al., 2022 | Six in-service high-school biology teachers in Israel; professional development around an AI tool for grading constructed biology responses. | Teachers initially expressed algorithm aversion and a belief in human grading superiority. After learning about the tool and confronting human disagreement, their views became more nuanced and willingness to use the studied tool increased. The small study does not establish population-wide effects. Source |
The contrast matters. Teachers may be less willing to seek or follow computerized advice in one task and still defer to an AI-attributed grade in another. A broader, non-teacher-specific summary from HEC Paris describes “algorithm appreciation”: people sometimes follow algorithmic advice more than identical human advice, even when told about errors. It also reports that complex performance information reduced appreciation. That finding offers context for how people respond to algorithmic advice, but it is not evidence about teachers in the Greek grading experiment. HEC Paris’s summary.
Why might an unfair AI grade go unchallenged?
The Greek experiment shows a difference in acceptance based on whether the grade was attributed to AI or a teacher; it does not establish what motivated each participant. It would be speculative to conclude that teachers accepted the score because they viewed AI as objective or infallible.
One useful distinction is between trust and uncritical compliance. Trust, using wording from Lee and See (2004) quoted by Nazaretsky and colleagues, is “the attitude that an agent will help achieve an individual’s goals in a situation characterized by uncertainty and vulnerability.” A teacher can reasonably use a tool while still checking its output. The practical danger arises when reliance exceeds the teacher’s ability or opportunity to evaluate a recommendation.
The professional-development study offers a possible reason that views can shift: teachers discussed both the AI tool and disagreement among human graders, then became more open to using the tool. Because the study involved only six teachers and one grading system, it cannot show that this approach prevents overreliance elsewhere. It does suggest that informed use involves more than simply being told that AI can make mistakes.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How can schools use AI grading without blind deference?
Monash’s Professor Rigissa Megalokonomou warned that “Teachers are deferring to AI outputs without sufficiently questioning them, and errors are quietly going uncorrected.” The implication is not that schools must reject AI grading. It is that a system should support teacher judgment rather than quietly replace it.
- Make review possible: Teachers should be able to inspect a recommendation and the student response it applies to, rather than receiving only a score.
- Keep correction authority with educators: The 2022 professional-development study’s authors recommend that teachers be able to review, modify, or override AI recommendations.
- Train with real examples: Professional development should include practice examining outputs and discussing where human graders may also disagree. The small Israeli study found a change in attitudes after practical exposure and discussion, but did not prove a universal training effect.
- Separate a tool’s usefulness from its authority: Weekly use for lesson preparation, as reported by 48 per cent of teachers in the Monash release, does not demonstrate that the same teachers should accept automated grading without review.
- Track challenges and corrections: Schools can examine whether teachers flag disputed scores and whether those flags lead to review. This is a practical oversight measure, not an outcome established by the cited studies.
Megalokonomou also said, “It’s not enough to just tell people AI can be wrong.” For school leaders, that means pairing warnings with the time, training, access to evidence, and authority teachers need to question an output.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
