Algorithm Delegation: How Embedded AI Facilitates Agency Transference in Medical Services

Summary

This paper integrates psychological-distance (construal-level) and human–AI agency theories to investigate how embedded AI — algorithms physically integrated into tangible medical devices — affects patients’ willingness to delegate medical decisions. Across four studies (a large text-mining analysis of 224,433 reviews and three controlled experiments totalling 677 participants) the authors find that embedding AI in physical form reduces psychological distance to the technology, increases trust and willingness to delegate healthcare decisions to algorithms, and raises expectations of better health outcomes. The effect is moderated: priming an analytical mindset with numerical accuracy information weakens the delegation effect. The paper draws design implications for reducing algorithm aversion in medical services.

Key Points

Embedded AI (AI integrated into physical devices/wearables) reduces psychological distance between users and algorithms.
Lower psychological distance increases people’s willingness to delegate medical decisions to algorithms.
Participants expect better health outcomes when they delegate to embedded AI across multiple studies.
Evidence combines large-scale text mining (n = 224,433) with three controlled experiments (combined n = 677).
The delegation effect is attenuated when participants are placed in an analytical mode via numerical accuracy information.
Design factors (embodiment, form factor, presentation) matter: making AI feel ‘closer’ can reduce algorithm aversion in healthcare contexts.
Data are available from the corresponding author on request; research funded by ENSILIS/FCT and HaDEA.

Content Summary

The authors propose that embedding AI into tangible medical artefacts (for example, smart wearables, bedside devices or embedded sensors) reduces the psychological distance people feel from algorithmic decision-makers. Reduced distance fosters perceptions of proximity and social presence, which in turn increases willingness to transfer agency — that is, to let algorithms make or support medical decisions. The manuscript reports a multi-method programme of research: a large-scale text-mining analysis to detect real-world signals, followed by controlled experiments that test causal mechanisms (psychological distance as mediator, analytical mind-set as moderator).

Key empirical findings: embedded AI consistently lowers perceived distance and raises delegation willingness; participants also expect better health outcomes when using embedded AI; however, when participants are prompted to think analytically (for example, shown numerical accuracy metrics), the positive effect of embodiment on delegation decreases. The authors discuss how these findings interact with prior work on algorithm aversion, anthropomorphism and trust, and provide practical suggestions for designers of medical AI and digital health services.

Context and Relevance

As healthcare systems scale up AI for diagnostics, triage, wearable monitoring and telehealth, patient acceptance is a major barrier. This study is important because it identifies a tangible, designable lever — embedding AI into physical devices — that can reduce resistance and increase uptake. The results are directly relevant to product teams building medical wearables, device manufacturers, clinical implementation teams and policymakers concerned with responsible deployment of AI in care settings. It also highlights a caution: making AI feel closer isn’t a universal fix — analytic scrutiny can reverse the effect, so transparency and accuracy remain critical.

Author style

Punchy. If you work on health-tech, medtech or patient-facing AI, read the methods and experiments — they give both robust evidence and clear design cues you can use to make algorithmic assistants more acceptable without glossing over the trade-offs when users adopt an analytical stance.

Why should I read this

Short version: embedding AI into real, touchable devices makes people more likely to hand over medical choices to algorithms — handy if you build wearables, telehealth tools or bedside tech. The paper backs that up with one massive text dataset and three experiments, and points out when the trick stops working (hint: when people are shown hard numbers and nudged to think analytically). Saves you the time of trawling the literature and gives practical design takeaways.