Dialogical guidelines aided by knowledge acquisition: Enhancing the design of explainable interfaces and algorithmic accuracy

    Publikation: Kapitel i bog/rapport/konference-proceedingKonferencebidrag i proceedingsForskningpeer review

    110 Downloads (Pure)

    Abstract

    Understanding expert domain knowledge may inform the design of explainable interfaces that convey comprehensible information by “mirroring” the explanation practice of domain experts. Likewise, scrutinizing expert domain knowledge is pivotal to guarantee data quality and enhance algorithmic accuracy, by zooming in on the types of data and information that constitute relevant and reliable representations in a given domain. Against this backdrop, the paper revitalizes the field of knowledge acquisition and presents easily applicable user-centered and value-oriented dialogical guidelines to unravel domain knowledge with the aim of enhancing the design of explainable interfaces and algorithmic accuracy. While it might seem counter-intuitive to revisit the field of knowledge acquisition in the era of machine learning and deep learning, there are plenty of cases in which AI systems, trained on biased data, have led to epistemological deficiencies with morally harmful consequences. In order to improve the data preparation and modelling stage in the development of ML models, this paper suggests that AI developers could benefit from the pragmatic application of manageable dialogical guidelines aided by knowledge acquisition to cultivate shared understanding between AI developers and domain expert end users.
    OriginalsprogEngelsk
    TitelProceedings of the future technologies conference (FTC) 2020
    RedaktørerKohei Arai, Supriya Kapoor, Rahul Bhatia
    Vol/bind1
    ForlagSpringer
    Publikationsdato2021
    Sider243-257
    ISBN (Trykt)978-3-030-63127-7
    ISBN (Elektronisk)978-3-030-63128-4
    DOI
    StatusUdgivet - 2021
    BegivenhedFTC 2020 - Future Technologies Conference 2020 - San Francisco, USA
    Varighed: 5. nov. 20206. nov. 2020

    Konference

    KonferenceFTC 2020 - Future Technologies Conference 2020
    Land/OmrådeUSA
    BySan Francisco
    Periode05/11/202006/11/2020
    NavnAdvances in Intelligent Systems and Computing
    Vol/bind1288
    ISSN2194-5357

    Bibliografisk note

    Online due to Covid19

    Fingeraftryk

    Dyk ned i forskningsemnerne om 'Dialogical guidelines aided by knowledge acquisition: Enhancing the design of explainable interfaces and algorithmic accuracy'. Sammen danner de et unikt fingeraftryk.

    Citationsformater