When is it better for an AI assistant to say nothing?

PursueUpdated 10 October 20268 sources

Short answer

Whenever speaking would cost more attention than it gives back: when the person can't take it in, the evidence is thin, or nothing needs to change. The 2018 framework for just-in-time support (Nahum-Shani et al.) counts giving nothing as a designed option, and declining uncertain cases makes a system's other answers more reliable (Geifman and El-Yaniv, 2017).

01Silence is a decision

The framework for just-in-time support treats each moment as a decision point where providing nothing is one of the options, and warns that repeated prompting creates burden[1]. A 2026 review adds that people often can't engage at the very moment they need help most[2].

Decision theory frames each alert as a trade between the value of telling someone now and the cost of interrupting them[3]. Even useful content is disruptive at the wrong moment[4], and a notification costs attention even when it is ignored[5].

02Why false alarms are expensive

People learn to ignore systems that cry wolf; false alarms are a common cause of 'disuse'[6]. Every unnecessary signal makes the next useful one weaker.

On the machine side, a system that may decline uncertain cases trades coverage for reliability[7]. Staying quiet is a legitimate action, not a failure.

03Silence is not the same as broken

One distinction is easy to lose: choosing not to speak is different from being unable to. A classic security principle says systems should fail to the safe side[8]. By analogy, an assistant that can't work reliably should say so, not go quiet and let you think there was nothing to say.

04Where the evidence stops

No study we found directly tests whether assistants that stay quiet more often lead to better outcomes. The sources support a design stance and testable hypotheses, not a proven rule.

Something to try today

Look at the apps allowed to interrupt you and ask of each one: when did it last tell me something I needed at that moment? Silence the ones you can't answer for.

Pursue

How Tre-Up approaches this

Pursue has three possible answers: one small step, one question, or nothing to change. Saying nothing is designed in, and when Pursue can't work reliably it tells you so.

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. How we write these notes

  1. 1.Nahum-Shani I, Smith SN, Spring BJ, Collins LM, Witkiewitz K, Tewari A, Murphy SA (2018). Just-in-Time Adaptive Interventions (JITAIs) in Mobile Health: Key Components and Design Principles for Ongoing Health Behavior Support. Annals of Behavioral Medicine 52(6):446–462.University of Michigan · Northwestern University · Pennsylvania State University · University of New MexicoOur summarydoi:10.1007/s12160-016-9830-8
  2. 2.Nahum-Shani I, Murphy SA (2026). Just-in-Time Adaptive Interventions: Where Are We Now and What Is Next?. Annual Review of Psychology 77:679–703.University of Michigan · Harvard UniversityOur summarydoi:10.1146/annurev-psych-121024-044244
  3. 3.Horvitz E, Jacobs A, Hovel D (1999). Attention-Sensitive Alerting. Proceedings of UAI 1999, 305–313.Microsoft ResearchOur summaryNo DOI
  4. 4.Mehrotra A, Pejovic V, Vermeulen J, Hendley R, Musolesi M (2016). My Phone and Me: Understanding People's Receptivity to Mobile Notifications. Proceedings of CHI 2016, 1021–1032.University of Birmingham · University College London · University of Ljubljana · University of CalgaryOur summarydoi:10.1145/2858036.2858566
  5. 5.Stothart C, Mitchum A, Yehnert C (2015). The attentional cost of receiving a cell phone notification. Journal of Experimental Psychology: Human Perception and Performance 41(4):893–897.Florida State UniversityOur summarydoi:10.1037/xhp0000100
  6. 6.Parasuraman R, Riley V (1997). Humans and Automation: Use, Misuse, Disuse, Abuse. Human Factors 39(2):230–253.Catholic University of America · Honeywell Technology CenterOur summarydoi:10.1518/001872097778543886
  7. 7.Geifman Y, El-Yaniv R (2017). Selective Classification for Deep Neural Networks. Advances in Neural Information Processing Systems 30 (NeurIPS 2017).Technion – Israel Institute of TechnologyOur summaryarXiv:1705.08500
  8. 8.Saltzer JH, Schroeder MD (1975). The Protection of Information in Computer Systems. Proceedings of the IEEE 63(9):1278–1308.Massachusetts Institute of TechnologyOur summarydoi:10.1109/PROC.1975.9939

Keep reading

All notes
  1. 01When is the right moment for a reminder or a nudge?When the person is able to take it in, which changes from moment to moment. In a trial with 1,255 users of a workplace well-being app (Bidargaddi et al., 2018), the same message made people about 4% more likely to use the app in the next day, and about 12% more likely at 12:30 on weekends.Timing and receptivity
  2. 02How 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.Trust and uncertainty

Tre-Up is built on this research.

Pursue, the AI inside Tre-Up, is designed to notice when your day starts to slip and offer one small next step. The iOS app is in development. Invitations go out in small groups; leave your email and we will contact you when yours is ready.

Your email is used only for your invitation and important product updates. It is never sold. Unsubscribe at any time. Privacy policy