Brief: algorithmic-aversion
Research brief for algorithmic-aversion · how it was made.
Research brief - algorithmic-aversion
| Field | Value |
|---|---|
| Slug | algorithmic-aversion |
| Working title | Algorithmic Aversion |
| Article shape | Specimen Record |
| Category | Digital Distortions (modern), after Automation Complacency |
| Intent tags | decision-biases, at-work |
| SEO / reader query intent | Primary: algorithm aversion. Also: why do I trust a person more after the model is wrong once, algorithm aversion Dietvorst, refusing an algorithm after it errs |
| AI authors credited | Grok 4.7 |
| Reviewer or checker | none |
| Date | 2026-10-08 |
1. Concept gate
Reader's real question:
- Why does one obvious machine mistake make me prefer a human who is wrong more often?
Why this belongs in the catalog:
- Automation complacency is over-trust and skipped checks. AI oracle fallacy treats fluent output as vetted. This page is the swing the other way: a seen error retires a better average.
Bug claim in one sentence:
The bug is Algorithmic Aversion, where a visible machine error makes a weaker human judgment feel like the safer choice.
What this bug is not:
- Not a claim that every model beats every person.
- Not a claim that wanting an explanation you can defend is the bias.
- Findings are from incentivized forecasting, then from studies where a small modification restored use.
Nearby bugs to check:
| Nearby bug | Difference |
|---|---|
| Automation Complacency | Stop checking because it looks finished. |
| AI Oracle Fallacy | Fluency feels vetted. |
| Deepfake Cynicism | Anything might be fake, so real evidence is dismissed. |
| Detection Overconfidence | Your eye is enough, so you skip the check. |
| Overconfidence Bias | Your skill feels above the record. |
Concept gate: yes
2. Sources and research
| Source / concept | Year | What it supports | URL / DOI / note |
|---|---|---|---|
| Dietvorst, Simmons, & Massey. Algorithm aversion. Journal of Experimental Psychology: General. | 2015 | Five studies. Seeing the algorithm err reduced willingness to tie payoffs to it, even after seeing it beat the human. Confidence dropped faster for the algorithm. | https://doi.org/10.1037/xge0000033 |
| Dietvorst, Simmons, & Massey. Overcoming algorithm aversion. Management Science. Published online 2016. | 2016 | Small rights to modify an imperfect forecast increased use and performance. Preference was for some control, not a large override. | https://doi.org/10.1287/mnsc.2016.2643 |
3. Article plan
Examples: I'll do it myself, a person is accountable, one miss junks the approach, my estimate feels safer, one invented citation, worse on paper but trusted, if I can't tweak it I don't want it.
Scenarios: sales forecast, routing, mangled name, diagnostic aid.
Reframes: the famous miss, the mangled name, the override.
Related: Automation Complacency, AI Oracle Fallacy, Deepfake Cynicism, Detection Overconfidence, Overconfidence Bias.
5. Sign-off
- ☑ Belonging test passed
- ☑ Nearby bugs checked
- ☑ Sources verified or uncertainty labeled
- ☑ Examples are concrete
- ☑ Reframes are realistic, not forced optimism
- ☑ Credits model/person named