Show simple item record

dc.contributor.authorBabbar, Varun
dc.contributor.authorBhatt, Umang
dc.contributor.authorWeller, Adrian
dc.date.accessioned2022-05-03T23:30:10Z
dc.date.available2022-05-03T23:30:10Z
dc.date.issued2022-05-03
dc.identifier.urihttps://www.repository.cam.ac.uk/handle/1810/336716
dc.description.abstractResearch on human-AI teams usually provides experts with a single label, which ignores the uncertainty in a model's recommendation. Conformal prediction (CP) is a well established line of research that focuses on building a theoretically grounded, calibrated prediction set, which may contain multiple labels. We explore how such prediction sets impact expert decision-making in human-AI teams. Our evaluation on human subjects finds that set valued predictions positively impact experts. However, we notice that the predictive sets provided by CP can be very large, which leads to unhelpful AI assistants. To mitigate this, we introduce D-CP, a method to perform CP on some examples and defer to experts. We prove that D-CP can reduce the prediction set size of non-deferred examples. We show how D-CP performs in quantitative and in human subject experiments ($n=120$). Our results suggest that CP prediction sets improve human-AI team performance over showing the top-1 prediction alone, and that experts find D-CP prediction sets are more useful than CP prediction sets.
dc.description.sponsorshipThe Alan Turing Institute Leverhulme Trust via CFI DeepMind Mozilla Foundation
dc.publisherInternational Joint Conferences on Artificial Intelligence Organization
dc.rightsAll Rights Reserved
dc.rights.urihttp://www.rioxx.net/licenses/all-rights-reserved
dc.titleOn the Utility of Prediction Sets in Human-AI Teams
dc.typeConference Object
dc.publisher.departmentDepartment of Engineering
dc.date.updated2022-05-02T16:35:14Z
prism.publicationNameProceedings of the Thirty-First International Joint Conference on Artificial Intelligence
dc.identifier.doi10.17863/CAM.84138
dcterms.dateAccepted2022-04-20
rioxxterms.versionofrecord10.24963/ijcai.2022/341
rioxxterms.versionAM
dc.contributor.orcidWeller, Adrian [0000-0003-1915-7158]
pubs.funder-project-idLeverhulme Trust (RC-2015-067)
pubs.funder-project-idEPSRC (EP/V025279/1)
pubs.funder-project-idAlan Turing Institute (TUR-000346)
pubs.conference-nameThirty-First International Joint Conference on Artificial Intelligence {IJCAI-22}
pubs.conference-start-date2022-07-23
cam.orpheus.success2022-08-05: proceedings now published, embargo removed due to conference's policy
cam.orpheus.counter8
cam.depositDate2022-05-02
pubs.conference-finish-date2022-07-29
pubs.licence-identifierapollo-deposit-licence-2-1
pubs.licence-display-nameApollo Repository Deposit Licence Agreement


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record