Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 24 of the 25 references it cites were found in a registry or an index; none of the 17 sources its claims cite is checked for support yet.

Adaptive Cognitive Mechanisms to Maintain Calibrated Trust and Reliance in Automation

Open access
Lebiere, Christian, Blaha, Leslie M., Fallon, Corey K., Jefferson, Brett · 2021 · Frontiers in Robotics and AI

Abstract

Trust calibration for a human–machine team is the process by which a human adjusts their expectations of the automation’s reliability and trustworthiness; adaptive support for trust calibration is needed to engender appropriate reliance on automation. Herein, we leverage an instance-based learning ACT-R cognitive model of decisions to obtain and rely on an automated assistant for visual search in a UAV interface. This cognitive model matches well with the human predictive power statistics measuring reliance decisions; we obtain from the model an internal estimate of automation reliability that mirrors human subjective ratings. The model is able to predict the effect of various potential disruptions, such as environmental changes or particular classes of adversarial intrusions on human trust in automation. Finally, we consider the use of model predictions to improve automation transparency that account for human cognitive biases in order to optimize the bidirectional interaction between human and machine through supporting trust calibration. The implications of our findings for the design of reliable and trustworthy automation are discussed.

Overview

CoChat Summary

TL;DR

The paper presents an instance-based learning ACT-R cognitive model of trust calibration and reliance decisions in an automation-aided visual search task, showing it reproduces human predictive-power statistics and subjective reliability ratings. It then uses the model to predict how transparency changes and adversarial disruptions can cause over- or under-reliance on automation.

Key findings

  • Baseline model selected Automation about 72% of the time in the Age condition, 67% in Number, and 82% in both Text and Graphic; decision-stage mean PPP was 0.77 for Age, 0.79 for Number, and 0.82 for Text and Graphic.
  • Model reliability estimates were 0.77 (Age), 0.80 (Number), 0.80 (Text), and 0.81 (Graphic), roughly matching human subjective ratings of 0.75, 0.74, 0.82, and 0.80.
  • A Random disruption (10% of automation-selected trials, reducing overall reliability from 70% to about 63%) increased Commander choice by about 12% in Age, 8% in Number, and 8% in Text/Graphic; Frontload increased Commander choice by 9% in Age/Number and 7% in Text/Graphic versus Random.
  • Frontload-Positive decreased Commander reliance by 12% (Age), 10% (Number), and 6% (Text/Graphic); Frontload-Commander decreased Commander reliance by 18% (Age), 17% (Number), and 9% (Text/Graphic). Low-to-high transparency transfer led to choosing unreliable automation almost 60% of the time, while high-to-low transfer increased automation reliance by 13–15% (Age 72 to 85; Number 67 to 82) and improved decision NPP from about 0.47 to 0.65 for Age and about 0.49 to about 0.74 for Number.

Methods

  • Design: an instance-based learning theory model integrated into ACT-R, representing experiences as {situation cues, choice made, outcome}; blended retrievals estimate expected search time and automation reliability, and the model selects the assist type with the lowest expected search time and relies on automation if it is expected to be reliable.
  • Data/sample: empirical data from the COBALT aided visual search task in Fallon et al. (2019); each block included 143 trials, and 20 simulations per condition were run, comparable to the N=16 human participants; all parameters were left at default ACT-R values (decay d=0.5, activation noise s=0.25, mismatch penalty MP=1.0, linear similarities over [0,−1.0]).
  • Analysis: positive and negative predictive power statistics (PPP, NPP) for decision and search phases, reliability estimates via blended retrieval, and manipulations of cue transfer and disruptions (Random, Frontload, Midload, Frontload-Positive, Frontload-Commander).

Limitations

  • Stated in the paper: the single COBALT interface does not separate actual automation performance reliability from the user’s perception of reliability or performance through the interface; direct observation of automation failure may produce different calibration, and remote unmanned settings may have no or delayed additional information. Future work may explore asynchronous or time-delayed feedback.
  • Evident from the text: the work is a simulation study, and the paper states that findings can inform hypotheses and that additional human-subject experiments can confirm model predictions.
  • Stated in the paper: the model only represents decisions about use of automation; decisions about other task aspects, such as when to initiate a task or trial, do not inform memory for the automation’s reliability.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3389/frobt.2021.652776

Cites

24 of 24 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

FIGURE 1 Diagram of the adaptive control of thought-rational (ACT-R) computational cognitive architecture. Each module indicates the neural areas associated with the cognitive mechanisms. The general functionality of the modules is indicated by the color coding.
FIGURE 1 Diagram of the adaptive control of thought-rational (ACT-R) computational cognitive architecture. Each module indicates the neural areas associated with the cognitive mechanisms. The general functionality of the modules is indicated by the color coding.CC BY · PubMed Central
FIGURE 2 The COBALT task environment. (A) marks the UAV map window; users select a UAV to initiate each trial here, (B) marks the visual search window, and (C) marks the automation transparency cue window. The four images on the right show the four transparency cue types, from top to bottom: number of sensors only, age of data only, age and number in text format (Text condition), and age and number in visual graphic format (Graphic condition).
FIGURE 2 The COBALT task environment. (A) marks the UAV map window; users select a UAV to initiate each trial here, (B) marks the visual search window, and (C) marks the automation transparency cue window. The four images on the right show the four transparency cue types, from top to bottom: number of sensors only, age of data only, age and number in text format (Text condition), and age and number in visual graphic format (Graphic condition).CC BY · PubMed Central
FIGURE 3 Instance-based learning model representation of the COBALT decisions to rely on automation. The colors used to code each aspect of the IBL model correspond to the type of cognitive module or mechanism leveraged from ACT-R.
FIGURE 3 Instance-based learning model representation of the COBALT decisions to rely on automation. The colors used to code each aspect of the IBL model correspond to the type of cognitive module or mechanism leveraged from ACT-R.CC BY · PubMed Central
FIGURE 4 Performance of the Baseline model together with human performance. Human data are from Fallon et al. (2019). Left plot shows the plot of True Positive against False Positive scores; colored icons are the model performance for each condition, and the black dots are the human data overlaid. The right plot gives the predictive power scores for the humans and the Baseline model in both phases and all conditions. Bar heights are the mean values and the points represent either one person or one iteration of the Baseline model.
FIGURE 4 Performance of the Baseline model together with human performance. Human data are from Fallon et al. (2019). Left plot shows the plot of True Positive against False Positive scores; colored icons are the model performance for each condition, and the black dots are the human data overlaid. The right plot gives the predictive power scores for the humans and the Baseline model in both phases and all conditions. Bar heights are the mean values and the points represent either one person or one iteration of the Baseline model.CC BY · PubMed Central
FIGURE 5 Predictive power statistics plotted for the cue transfer manipulations.
FIGURE 5 Predictive power statistics plotted for the cue transfer manipulations.CC BY · PubMed Central
FIGURE 6 Plots of predictive power statistics over repeated transfer, where the second cue type is repeated for three blocks total. Bars are the means and points are the individual model scores.
FIGURE 6 Plots of predictive power statistics over repeated transfer, where the second cue type is repeated for three blocks total. Bars are the means and points are the individual model scores.CC BY · PubMed Central
FIGURE 7 Predictive power statistics plotted with the Baseline condition scores on the x-axis and one of the cue-performance disruption manipulations on the y-axis. The subplots capture the different manipulations. (A) Frontload. (B) Frontload-Positive. (C) Frontload-Commander. (D) Midload. (E) Random.
FIGURE 7 Predictive power statistics plotted with the Baseline condition scores on the x-axis and one of the cue-performance disruption manipulations on the y-axis. The subplots capture the different manipulations. (A) Frontload. (B) Frontload-Positive. (C) Frontload-Commander. (D) Midload. (E) Random.CC BY · PubMed Central

Verification

Claims the paper makes

0 Valid · 8 Not valid · 16 Not checked · 93 Own results and interpretations

40 more claims

Create a free account to see all 117 claims and to verify the 11 references still open.

See all claims — free account
  • It may state two separate claims as one. 46
  • Probably not a substantial claim. 33

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Citations

Cites

Show all 25 references

Research works printed but not linked: 2

  1. Bliss, 2006, Active Warnings: False Alarms, 231
  2. 2010, DHS Risk Lexicon 2010 EditionTech. Rep

Cited by

See all 62

Metadata

Authors
Lebiere, C., Blaha, L. M., Fallon, C. K., & Jefferson, B.
Affiliations
  • Carnegie Mellon University (US)
  • United States Air Force Research Laboratory (US)
  • Pacific Northwest National Laboratory (US)
Corresponding author
Lebiere, Christian
Venue
Frontiers in Robotics and AI
Published
2021-05-24
Volume
8
Pages
652776-652776
ISSN
2296-9144 (ISSN-L)
OpenAlex
W3164041257
Cited by
62
References
25
Citations by year
2026: 29 · 2025: 6 · 2024: 13 · 2023: 7 · 2022: 7
FWCI
4.17
Type
journal article
Language
en
Open access
Locations
  • Frontiers in Robotics and AI (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by)
  • OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
Air Force Office of Scientific Research
Topics
  • Social Sciences › Human Factors and Ergonomics › Human-Automation Interaction and Safety
  • Social Sciences › Psychology › Social Psychology › Social Robot Interaction and HRI
  • Physical Sciences › Engineering › Safety, Risk, Reliability and Quality › Risk and Safety Analysis
Keywords
trust calibration, reliance on automation, human-machine teams, instance-based learning, visual search, automation reliability, automation transparency, cognitive biases, trust in automation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-11-18, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-18 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Adaptive Cognitive Mechanisms to Maintain Calibrated Trust and Reliance in Automation. https://app.cochat.ai/papers/10.3389/frobt.2021.652776
The paper
Lebiere, C., Blaha, L. M., Fallon, C. K., & Jefferson, B. (2021). Adaptive Cognitive Mechanisms to Maintain Calibrated Trust and Reliance in Automation. Frontiers in Robotics and AI. https://doi.org/10.3389/frobt.2021.652776

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.