MétaCan
Menu
Back to cohort
Record W4405220551 · doi:10.1016/j.ijhcs.2024.103434

Human performance effects of combining counterfactual explanations with normative and contrastive explanations in supervised machine learning for automated decision assistance

2024· article· en· W4405220551 on OpenAlex

Why this work is in the frame

A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.

affAt least one author lists a Canadian institution in the pinned OpenAlex snapshot.
fundA Canadian funder is recorded on the work.

Bibliographic record

VenueInternational Journal of Human-Computer Studies · 2024
Typearticle
Languageen
FieldComputer Science
TopicExplainable Artificial Intelligence (XAI)
Canadian institutionsUniversity of Toronto
FundersNatural Sciences and Engineering Research Council of CanadaMitacs
KeywordsCounterfactual thinkingNormativeArtificial intelligenceMachine learningComputer sciencePsychologyCognitive psychologySocial psychologyEpistemologyPhilosophy

Abstract

fetched live from OpenAlex

• Counterfactual explanations gained popularity as a solution for aiding causal understanding and explain the reasons behind machine learning outputs. • Empirical data on the influence of counterfactuals on human decisions is scarce compared to traditional explanation methods. • An experiment analyzed the effects of counterfactual explanations alongside normative and contrastive explanations in condition-based maintenance. • Including counterfactual explanations reduced false alarm rates and potentially decreased decision time and workload. • Caution is advised against overstating the benefits of counterfactuals in digital work environments. Counterfactual explanations have emerged as a popular solution for elucidating the reasons behind machine learning predictions due to their contribution in supporting people's understanding of causality. Despite psychological research suggesting potential burdens associated with counterfactuals, empirical data on the influence of counterfactual explanations on human decisions is limited, especially in comparison with other more traditional explanation methods in machine learning for decision assistance. We present an experiment to examine the human performance effects of counterfactual explanations combined with normative and contrastive explanations in the context of condition-based manteinance. Twenty-four participants provided their diagnosis of the conditions of a hydraulic system with the assistance of a simulated decision aid based on machine learning, under four experimental conditions (baseline with no explanations, normative plus contrastive explanations, normative plus counterfactual explanations, and normative plus contrastive plus counterfactual explanations). The results indicate a lack of significant performance differences between explanation conditions. However, we found a reduction in false alarm rate in the condition with all three explanations, and a potential reduction in decision time and workload in the two conditions that included counterfactual explanations. These findings highlight the potential of counterfactuals to reduce decision time and workload, but they also caution against overestimating their benefits in supporting decision performance within digital work environments.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

Full frame distilled prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.

metaresearch head score (Codex)0.001
metaresearch head score (Gemma)0.000
Version: codex-gemma-dda1882f352aValidation status: machine_predicted_unvalidated
Candidate categoriesnone
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Simulation or modeling · Consensus signal: none
GenreCandidate signal: Empirical · Consensus signal: Empirical
Teacher disagreement score0.736
Threshold uncertainty score0.610

Codex and Gemma teacher scores by category

CategoryCodexGemma
Metaresearch0.0010.000
Meta-epidemiology (narrow)0.0000.000
Meta-epidemiology (broad)0.0000.000
Bibliometrics0.0010.000
Science and technology studies0.0000.000
Scholarly communication0.0000.001
Open science0.0010.000
Research integrity0.0000.000
Insufficient payload (model declined to judge)0.0000.000

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.028
GPT teacher head0.335
Teacher spread0.308 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it