What question did this study set out to answer?

To evaluate the trade-off between perceived efficiency gains and the verification burden when applying large language models to workplace risk assessment workflows.

August 15, 2026Proceedings of the Estonian Academy of SciencesOpen Access

The efficiency–verification trade-off in large language model-assisted workplace risk assessment

View Full Paper

Ask AI

Bookmark

Authors

ABAleksandr Bosler TKTarmo KoppelTallinn University of Technology KRKarin ReinholdTallinn University of Technology

Discussion

Member takes

Implication

Evaluation study demonstrates significant time savings but notable AI-human disagreement in student risk assessments, highlighting potential overreliance on language models.

Key Points

To evaluate the trade-off between perceived efficiency gains and the verification burden when applying large language models to workplace risk assessment workflows.
Analyzed 96–121 analytic sessions derived from 111 student assignment files in an educational setting.
Operationalized efficiency via perceived completion time versus an estimated manual baseline, and verification burden as AI–human quantitative score disagreement (1–5 scale for probability and impact).
Evaluated user acceptance with utility and satisfaction Likert scales and modeled satisfaction predictors using robust ordinary least squares (OLS) regression.
LLM assistance reduced median task time from 120 minutes to 20 minutes, representing a median perceived time saving of 0.84.
The median AI–human disagreement proxy was 0.50, with 26.8% of sessions having disagreement ≥1.0, and 40.6% of sessions displaying high satisfaction alongside high disagreement.
Robust OLS regression showed utility as the strongest positive predictor of satisfaction and disagreement as a negative predictor, whereas time savings were not statistically significant.

Cite This Study

Bosler et al. (2026) studied this question.

synapsesocial.com/papers/6a8019bb75c2e31742c85d88 https://doi.org/10.3176/proc.2026.3.10

View Full Paper

Ask AI

Bookmark

August 15, 2026Proceedings of the Estonian Academy of SciencesOpen Access

The efficiency–verification trade-off in large language model-assisted workplace risk assessment

View Full Paper

Ask AI

Bookmark

Authors

ABAleksandr Bosler TKTarmo KoppelTallinn University of Technology KRKarin ReinholdTallinn University of Technology

Discussion

Member takes

Implication

Evaluation study demonstrates significant time savings but notable AI-human disagreement in student risk assessments, highlighting potential overreliance on language models.

Key Points

To evaluate the trade-off between perceived efficiency gains and the verification burden when applying large language models to workplace risk assessment workflows.
Analyzed 96–121 analytic sessions derived from 111 student assignment files in an educational setting.
Operationalized efficiency via perceived completion time versus an estimated manual baseline, and verification burden as AI–human quantitative score disagreement (1–5 scale for probability and impact).
Evaluated user acceptance with utility and satisfaction Likert scales and modeled satisfaction predictors using robust ordinary least squares (OLS) regression.
LLM assistance reduced median task time from 120 minutes to 20 minutes, representing a median perceived time saving of 0.84.
The median AI–human disagreement proxy was 0.50, with 26.8% of sessions having disagreement ≥1.0, and 40.6% of sessions displaying high satisfaction alongside high disagreement.
Robust OLS regression showed utility as the strongest positive predictor of satisfaction and disagreement as a negative predictor, whereas time savings were not statistically significant.

Cite This Study

Bosler et al. (2026) studied this question.

synapsesocial.com/papers/6a8019bb75c2e31742c85d88 https://doi.org/10.3176/proc.2026.3.10

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

View Full Paper

Ask AI

Bookmark