# How should an AI say it isn't sure?

> Plainly, and without inventing a confidence number. Modern neural networks are often over-confident (Guo et al., 2017), and a system allowed to decline uncertain cases makes far fewer errors on the ones it answers (Geifman and El-Yaniv, 2017); a 2019 review led from the University of Cambridge found that admitting uncertainty does not necessarily cost trust.

- Page: https://tre-up.com/research/how-should-an-ai-say-it-isnt-sure
- Topic: [Trust and uncertainty](https://tre-up.com/research/topics/trust-and-uncertainty)
- Published 2026-10-07 · Updated 2026-10-10 · 6 sources
- Written by Pursue Lab, Tre-Up's research process (https://tre-up.com/research/method)
- Cite as: Pursue Lab, Tre-Up (2026). How should an AI say it isn't sure? https://tre-up.com/research/how-should-an-ai-say-it-isnt-sure

## Confidence scores can't be taken at face value

Researchers at Cornell showed that modern neural networks are often over-confident: the probability they attach to an answer does not match how often they are right [1]. A system that shows its raw confidence as certainty may be manufacturing it.

Work at the Technion showed the other side: a network allowed to decline when it is unsure can keep its error rate under a chosen limit. It answers fewer cases, and the answers it gives are far more reliable [2].

## Saying it to people

A review led from the University of Cambridge found that the limited evidence does not show that communicating uncertainty necessarily backfires on trust, though effects vary between people and formats [3].

Explaining why a system might be wrong kept people's trust after they saw it make mistakes [4], while too much explanation eroded it [5]. The aim is trust that matches ability, not trust at any price [6].

## Where the evidence stops

The machine-learning results come from image and text benchmarks; applying them to personal advice is an analogy. The communication research is thin, and its authors call for more experiments.

## Something to try today

When an AI gives you a confident answer about something personal, ask it what it would need to know to be more sure. A good assistant will tell you; a poor one will just repeat itself.

## How Tre-Up approaches this

Pursue never shows an invented confidence number. When the picture is thin, it asks one question or stays quiet, and when it can't work reliably it says so instead of producing something that looks like advice.

Tre-Up is an iOS app in development. Pursue is the AI inside it. This note describes research and design intent, not results from the app. Nothing here is medical advice.

## Sources

Summaries are written in our own words from each source's abstract, public record or official text. Follow the link to read the original.

1. Guo C, Pleiss G, Sun Y, Weinberger KQ (2017). On Calibration of Modern Neural Networks. Proceedings of ICML 2017, PMLR 70:1321–1330. https://arxiv.org/abs/1706.04599
   - Institutions: Cornell University (United States)
   - Our summary: https://tre-up.com/research/sources/guo-2017-calibration
2. Geifman Y, El-Yaniv R (2017). Selective Classification for Deep Neural Networks. Advances in Neural Information Processing Systems 30 (NeurIPS 2017). https://arxiv.org/abs/1705.08500
   - Institutions: Technion – Israel Institute of Technology (Israel)
   - Our summary: https://tre-up.com/research/sources/geifman-el-yaniv-2017-selective-classification
3. van der Bles AM, van der Linden S, Freeman ALJ, Mitchell J, Galvao AB, Zaval L, Spiegelhalter DJ (2019). Communicating uncertainty about facts, numbers and science. Royal Society Open Science 6(5):181870. https://doi.org/10.1098/rsos.181870
   - Institutions: University of Cambridge (United Kingdom); University of Warwick (United Kingdom); Columbia University (United States)
   - Our summary: https://tre-up.com/research/sources/van-der-bles-2019-communicating-uncertainty
4. Dzindolet MT, Peterson SA, Pomranky RA, Pierce LG, Beck HP (2003). The role of trust in automation reliance. International Journal of Human-Computer Studies 58(6):697–718. https://doi.org/10.1016/S1071-5819(03)00038-7
   - Institutions: Affiliation not yet confirmed
   - Our summary: https://tre-up.com/research/sources/dzindolet-2003-trust-reliance
5. Kizilcec RF (2016). How Much Information? Effects of Transparency on Trust in an Algorithmic Interface. Proceedings of CHI 2016, 2390–2395. https://doi.org/10.1145/2858036.2858402
   - Institutions: Stanford University (United States)
   - Our summary: https://tre-up.com/research/sources/kizilcec-2016-transparency-trust
6. Lee JD, See KA (2004). Trust in Automation: Designing for Appropriate Reliance. Human Factors 46(1):50–80. https://doi.org/10.1518/hfes.46.1.50_30392
   - Institutions: University of Iowa (United States)
   - Our summary: https://tre-up.com/research/sources/lee-see-2004-trust-in-automation

## Related notes

- [Should I trust an AI's advice about how to spend my day?](https://tre-up.com/research/should-i-trust-an-ai-with-advice-about-my-day.md)
- [When is it better for an AI assistant to say nothing?](https://tre-up.com/research/when-is-it-better-for-an-ai-assistant-to-say-nothing.md)

---

© 2026 Tre-Up. You may read, link to, quote and summarise Research Notes, and search engines and AI assistants may index, summarise and cite them, as long as the source is credited with a link to the page on tre-up.com. Please do not present our notes as your own, or republish whole pages, without permission. Terms: https://tre-up.com/terms#use · About Tre-Up: https://tre-up.com/about