生成式人工智能与网络级发现

A report from the NISO Open Discovery Initiative Standing Committee Authors:Ken Varnum(University of Michigan),Rachel Kessler(Clarivate),Judy Zhu(IEEE),Teresa Hazen(University of Arizona),Bobbi Patham(SpringerNature), andJay Holloway(OCLC) August2025 ExecutiveSummary The NISO Open Discovery Initiative (ODI)Standing Committeemost recently updated its RecommendedPractice, “Promoting Transparency in Discovery,” in 2020, several years before large-scale GenerativeArtificial Intelligence (AI) tools emerged into the public’s view in late 2023. Much as web-scale discoverychanged the library discovery environment around 2010, Generative AI has the potential to besimilarlyrevolutionary,affecting the way library users do their research and content providers share their full-text and indexing. To better understand the specific areas of interest and concern among participants in the discoveryecosystem,ODIconducted a surveyofour constituentsfromSeptembertoOctober 2024, followingseveral months of planning. The findings from that survey are included in this report, along with specificareas of work that ODI will be taking on in the months to come.ODIseeksto enhance mechanisms fortransparency in the use of Generative AI among content providers (publishers and traditionalabstracting andindexing services), discovery providers (companies that collect and index content frommultiple content providers), and libraries (who license both the discovery service and content beingindexed). Key findings of our survey include: •Library and Content Provider respondents sharedcommon hopes about Generative AI indiscovery.Theirmutual top three choices were: Improved visibility of content; More accuratecontent recommendations; and Saving staff time.•There was less uniformity among fears, with each cohort responding in ways that were moreclosely aligned with their interests. Interestingly both cohorts ranked “lack of transparencyabout inputs” in their top three, though presumably for different reasons.oLibrary respondents’ top three fears were: 1) Misrepresentation or fabrication ofcontent; 2) Lack of transparency about inputs; and 3) Quality of AI-generatedsummaries.oContent Provider respondents’ top three fears were: 1) Copyright infringement; 2)Misrepresentation or fabrication of content; and 3) Lack of transparency about inputs.•Library respondents generally felt that Generative AI tools would enhance the value of web scale discovery for users and staff, though not overwhelmingly so.•Content Provider respondents expressed hope that Generative AI would, in most cases, improvethe visibility of their own content, while being less positive about the effect on access to it.•Library respondents, in particular, expressed concern about the potential environmental risksposed by the data centers needed to create Generative AI models and process queries. Based on this survey, theNISOODI Standing Committee is planning to follow up inthe following areas: 1.Present these findingsto the community through conference presentations and a NISO webinarandinvite community feedbackon these findings.2.Develop one or more guidesto foster transparency among libraries, content providers, anddiscovery providers around the use of Generative AI in web-scale discovery.3.Continue to engage with stakeholdersto explore the overlapping hopes and concerns ofcontent providers and libraries, to further explore areas of mutual interest. Contents Executive Summary........................................................................................................................................i1.Introduction.............................................................................................................................................12.Methodology...........................................................................................................................................13.Findings...................................................................................................................................................23.1.Screening Questions.....................................................................................................................23.2.Library-Specific Questions............................................................................................................33.3.Content Provider-Specific Questions...........................................................................................84.Discussion and Recommendations.......................................................................................................134.1.Ranking Hopes and Fears...........................................................................................................134.2.Free-text Responses...................................................................................................................134.3.Value Propositions...................................................................................