Machine Unlearning Doesn't Do What You Think: Lessons for Generative AI Policy, Research, and Practice

We articulate fundamental mismatches between technical methods for machine unlearning in Generative AI, and documented aspirations for broader impact that these methods could have for law and policy. These aspirations are both numerous and varied, motivated by issues that pertain to privacy, copyrig...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:arXiv.org 2024-12
Hauptverfasser: A Feder Cooper, Choquette-Choo, Christopher A, Bogen, Miranda, Jagielski, Matthew, Filippova, Katja, Liu, Ken Ziyu, Chouldechova, Alexandra, Hayes, Jamie, Huang, Yangsibo, Mireshghallah, Niloofar, Shumailov, Ilia, Triantafillou, Eleni, Kairouz, Peter, Mitchell, Nicole, Liang, Percy, Ho, Daniel E, Choi, Yejin, Koyejo, Sanmi, Delgado, Fernando, Grimmelmann, James, Shmatikov, Vitaly, De Sa, Christopher, Barocas, Solon, Cyphert, Amy, Lemley, Mark, Boyd, Danah, Jennifer Wortman Vaughan, Brundage, Miles, Bau, David, Neel, Seth, Jacobs, Abigail Z, Terzis, Andreas, Wallach, Hanna, Papernot, Nicolas, Lee, Katherine
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title arXiv.org
container_volume
creator A Feder Cooper
Choquette-Choo, Christopher A
Bogen, Miranda
Jagielski, Matthew
Filippova, Katja
Liu, Ken Ziyu
Chouldechova, Alexandra
Hayes, Jamie
Huang, Yangsibo
Mireshghallah, Niloofar
Shumailov, Ilia
Triantafillou, Eleni
Kairouz, Peter
Mitchell, Nicole
Liang, Percy
Ho, Daniel E
Choi, Yejin
Koyejo, Sanmi
Delgado, Fernando
Grimmelmann, James
Shmatikov, Vitaly
De Sa, Christopher
Barocas, Solon
Cyphert, Amy
Lemley, Mark
Boyd, Danah
Jennifer Wortman Vaughan
Brundage, Miles
Bau, David
Neel, Seth
Jacobs, Abigail Z
Terzis, Andreas
Wallach, Hanna
Papernot, Nicolas
Lee, Katherine
description We articulate fundamental mismatches between technical methods for machine unlearning in Generative AI, and documented aspirations for broader impact that these methods could have for law and policy. These aspirations are both numerous and varied, motivated by issues that pertain to privacy, copyright, safety, and more. For example, unlearning is often invoked as a solution for removing the effects of targeted information from a generative-AI model's parameters, e.g., a particular individual's personal data or in-copyright expression of Spiderman that was included in the model's training data. Unlearning is also proposed as a way to prevent a model from generating targeted types of information in its outputs, e.g., generations that closely resemble a particular individual's data or reflect the concept of "Spiderman." Both of these goals--the targeted removal of information from a model and the targeted suppression of information from a model's outputs--present various technical and substantive challenges. We provide a framework for thinking rigorously about these challenges, which enables us to be clear about why unlearning is not a general-purpose solution for circumscribing generative-AI model behavior in service of broader positive impact. We aim for conceptual clarity and to encourage more thoughtful communication among machine learning (ML), law, and policy experts who seek to develop and apply technical methods for compliance with policy objectives.
format Article
fullrecord <record><control><sourceid>proquest</sourceid><recordid>TN_cdi_proquest_journals_3143055658</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>3143055658</sourcerecordid><originalsourceid>FETCH-proquest_journals_31430556583</originalsourceid><addsrcrecordid>eNqNjMuKAjEQAIOwoKj_0ODBi8KYTFS8yfpaUBBRxNMQYo8Td-hoOiP49zsHP2BPdaiiGqIllRoNp6mUTdFlvidJIscTqbVqiXxnbOEI4UQlmkCObrDwyNSPNeFcmAgXX8Gxjn5nsEVmTwy5D7BGwmCieyHMf2DvS2ffAzgg1x9bDMDQFfbB2OgsdsRXbkrG7odt0Vstj9-b4SP4Z4Ucs7uvAtUqU6NUJVqP9VT9r_oDqo9FKA</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>3143055658</pqid></control><display><type>article</type><title>Machine Unlearning Doesn't Do What You Think: Lessons for Generative AI Policy, Research, and Practice</title><source>Free E- Journals</source><creator>A Feder Cooper ; Choquette-Choo, Christopher A ; Bogen, Miranda ; Jagielski, Matthew ; Filippova, Katja ; Liu, Ken Ziyu ; Chouldechova, Alexandra ; Hayes, Jamie ; Huang, Yangsibo ; Mireshghallah, Niloofar ; Shumailov, Ilia ; Triantafillou, Eleni ; Kairouz, Peter ; Mitchell, Nicole ; Liang, Percy ; Ho, Daniel E ; Choi, Yejin ; Koyejo, Sanmi ; Delgado, Fernando ; Grimmelmann, James ; Shmatikov, Vitaly ; De Sa, Christopher ; Barocas, Solon ; Cyphert, Amy ; Lemley, Mark ; Boyd, Danah ; Jennifer Wortman Vaughan ; Brundage, Miles ; Bau, David ; Neel, Seth ; Jacobs, Abigail Z ; Terzis, Andreas ; Wallach, Hanna ; Papernot, Nicolas ; Lee, Katherine</creator><creatorcontrib>A Feder Cooper ; Choquette-Choo, Christopher A ; Bogen, Miranda ; Jagielski, Matthew ; Filippova, Katja ; Liu, Ken Ziyu ; Chouldechova, Alexandra ; Hayes, Jamie ; Huang, Yangsibo ; Mireshghallah, Niloofar ; Shumailov, Ilia ; Triantafillou, Eleni ; Kairouz, Peter ; Mitchell, Nicole ; Liang, Percy ; Ho, Daniel E ; Choi, Yejin ; Koyejo, Sanmi ; Delgado, Fernando ; Grimmelmann, James ; Shmatikov, Vitaly ; De Sa, Christopher ; Barocas, Solon ; Cyphert, Amy ; Lemley, Mark ; Boyd, Danah ; Jennifer Wortman Vaughan ; Brundage, Miles ; Bau, David ; Neel, Seth ; Jacobs, Abigail Z ; Terzis, Andreas ; Wallach, Hanna ; Papernot, Nicolas ; Lee, Katherine</creatorcontrib><description>We articulate fundamental mismatches between technical methods for machine unlearning in Generative AI, and documented aspirations for broader impact that these methods could have for law and policy. These aspirations are both numerous and varied, motivated by issues that pertain to privacy, copyright, safety, and more. For example, unlearning is often invoked as a solution for removing the effects of targeted information from a generative-AI model's parameters, e.g., a particular individual's personal data or in-copyright expression of Spiderman that was included in the model's training data. Unlearning is also proposed as a way to prevent a model from generating targeted types of information in its outputs, e.g., generations that closely resemble a particular individual's data or reflect the concept of "Spiderman." Both of these goals--the targeted removal of information from a model and the targeted suppression of information from a model's outputs--present various technical and substantive challenges. We provide a framework for thinking rigorously about these challenges, which enables us to be clear about why unlearning is not a general-purpose solution for circumscribing generative-AI model behavior in service of broader positive impact. We aim for conceptual clarity and to encourage more thoughtful communication among machine learning (ML), law, and policy experts who seek to develop and apply technical methods for compliance with policy objectives.</description><identifier>EISSN: 2331-8422</identifier><language>eng</language><publisher>Ithaca: Cornell University Library, arXiv.org</publisher><subject>Generative artificial intelligence ; Machine learning</subject><ispartof>arXiv.org, 2024-12</ispartof><rights>2024. This work is published under http://arxiv.org/licenses/nonexclusive-distrib/1.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>780,784</link.rule.ids></links><search><creatorcontrib>A Feder Cooper</creatorcontrib><creatorcontrib>Choquette-Choo, Christopher A</creatorcontrib><creatorcontrib>Bogen, Miranda</creatorcontrib><creatorcontrib>Jagielski, Matthew</creatorcontrib><creatorcontrib>Filippova, Katja</creatorcontrib><creatorcontrib>Liu, Ken Ziyu</creatorcontrib><creatorcontrib>Chouldechova, Alexandra</creatorcontrib><creatorcontrib>Hayes, Jamie</creatorcontrib><creatorcontrib>Huang, Yangsibo</creatorcontrib><creatorcontrib>Mireshghallah, Niloofar</creatorcontrib><creatorcontrib>Shumailov, Ilia</creatorcontrib><creatorcontrib>Triantafillou, Eleni</creatorcontrib><creatorcontrib>Kairouz, Peter</creatorcontrib><creatorcontrib>Mitchell, Nicole</creatorcontrib><creatorcontrib>Liang, Percy</creatorcontrib><creatorcontrib>Ho, Daniel E</creatorcontrib><creatorcontrib>Choi, Yejin</creatorcontrib><creatorcontrib>Koyejo, Sanmi</creatorcontrib><creatorcontrib>Delgado, Fernando</creatorcontrib><creatorcontrib>Grimmelmann, James</creatorcontrib><creatorcontrib>Shmatikov, Vitaly</creatorcontrib><creatorcontrib>De Sa, Christopher</creatorcontrib><creatorcontrib>Barocas, Solon</creatorcontrib><creatorcontrib>Cyphert, Amy</creatorcontrib><creatorcontrib>Lemley, Mark</creatorcontrib><creatorcontrib>Boyd, Danah</creatorcontrib><creatorcontrib>Jennifer Wortman Vaughan</creatorcontrib><creatorcontrib>Brundage, Miles</creatorcontrib><creatorcontrib>Bau, David</creatorcontrib><creatorcontrib>Neel, Seth</creatorcontrib><creatorcontrib>Jacobs, Abigail Z</creatorcontrib><creatorcontrib>Terzis, Andreas</creatorcontrib><creatorcontrib>Wallach, Hanna</creatorcontrib><creatorcontrib>Papernot, Nicolas</creatorcontrib><creatorcontrib>Lee, Katherine</creatorcontrib><title>Machine Unlearning Doesn't Do What You Think: Lessons for Generative AI Policy, Research, and Practice</title><title>arXiv.org</title><description>We articulate fundamental mismatches between technical methods for machine unlearning in Generative AI, and documented aspirations for broader impact that these methods could have for law and policy. These aspirations are both numerous and varied, motivated by issues that pertain to privacy, copyright, safety, and more. For example, unlearning is often invoked as a solution for removing the effects of targeted information from a generative-AI model's parameters, e.g., a particular individual's personal data or in-copyright expression of Spiderman that was included in the model's training data. Unlearning is also proposed as a way to prevent a model from generating targeted types of information in its outputs, e.g., generations that closely resemble a particular individual's data or reflect the concept of "Spiderman." Both of these goals--the targeted removal of information from a model and the targeted suppression of information from a model's outputs--present various technical and substantive challenges. We provide a framework for thinking rigorously about these challenges, which enables us to be clear about why unlearning is not a general-purpose solution for circumscribing generative-AI model behavior in service of broader positive impact. We aim for conceptual clarity and to encourage more thoughtful communication among machine learning (ML), law, and policy experts who seek to develop and apply technical methods for compliance with policy objectives.</description><subject>Generative artificial intelligence</subject><subject>Machine learning</subject><issn>2331-8422</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>ABUWG</sourceid><sourceid>AFKRA</sourceid><sourceid>AZQEC</sourceid><sourceid>BENPR</sourceid><sourceid>CCPQU</sourceid><sourceid>DWQXO</sourceid><recordid>eNqNjMuKAjEQAIOwoKj_0ODBi8KYTFS8yfpaUBBRxNMQYo8Td-hoOiP49zsHP2BPdaiiGqIllRoNp6mUTdFlvidJIscTqbVqiXxnbOEI4UQlmkCObrDwyNSPNeFcmAgXX8Gxjn5nsEVmTwy5D7BGwmCieyHMf2DvS2ffAzgg1x9bDMDQFfbB2OgsdsRXbkrG7odt0Vstj9-b4SP4Z4Ucs7uvAtUqU6NUJVqP9VT9r_oDqo9FKA</recordid><startdate>20241209</startdate><enddate>20241209</enddate><creator>A Feder Cooper</creator><creator>Choquette-Choo, Christopher A</creator><creator>Bogen, Miranda</creator><creator>Jagielski, Matthew</creator><creator>Filippova, Katja</creator><creator>Liu, Ken Ziyu</creator><creator>Chouldechova, Alexandra</creator><creator>Hayes, Jamie</creator><creator>Huang, Yangsibo</creator><creator>Mireshghallah, Niloofar</creator><creator>Shumailov, Ilia</creator><creator>Triantafillou, Eleni</creator><creator>Kairouz, Peter</creator><creator>Mitchell, Nicole</creator><creator>Liang, Percy</creator><creator>Ho, Daniel E</creator><creator>Choi, Yejin</creator><creator>Koyejo, Sanmi</creator><creator>Delgado, Fernando</creator><creator>Grimmelmann, James</creator><creator>Shmatikov, Vitaly</creator><creator>De Sa, Christopher</creator><creator>Barocas, Solon</creator><creator>Cyphert, Amy</creator><creator>Lemley, Mark</creator><creator>Boyd, Danah</creator><creator>Jennifer Wortman Vaughan</creator><creator>Brundage, Miles</creator><creator>Bau, David</creator><creator>Neel, Seth</creator><creator>Jacobs, Abigail Z</creator><creator>Terzis, Andreas</creator><creator>Wallach, Hanna</creator><creator>Papernot, Nicolas</creator><creator>Lee, Katherine</creator><general>Cornell University Library, arXiv.org</general><scope>8FE</scope><scope>8FG</scope><scope>ABJCF</scope><scope>ABUWG</scope><scope>AFKRA</scope><scope>AZQEC</scope><scope>BENPR</scope><scope>BGLVJ</scope><scope>CCPQU</scope><scope>DWQXO</scope><scope>HCIFZ</scope><scope>L6V</scope><scope>M7S</scope><scope>PIMPY</scope><scope>PQEST</scope><scope>PQQKQ</scope><scope>PQUKI</scope><scope>PRINS</scope><scope>PTHSS</scope></search><sort><creationdate>20241209</creationdate><title>Machine Unlearning Doesn't Do What You Think: Lessons for Generative AI Policy, Research, and Practice</title><author>A Feder Cooper ; Choquette-Choo, Christopher A ; Bogen, Miranda ; Jagielski, Matthew ; Filippova, Katja ; Liu, Ken Ziyu ; Chouldechova, Alexandra ; Hayes, Jamie ; Huang, Yangsibo ; Mireshghallah, Niloofar ; Shumailov, Ilia ; Triantafillou, Eleni ; Kairouz, Peter ; Mitchell, Nicole ; Liang, Percy ; Ho, Daniel E ; Choi, Yejin ; Koyejo, Sanmi ; Delgado, Fernando ; Grimmelmann, James ; Shmatikov, Vitaly ; De Sa, Christopher ; Barocas, Solon ; Cyphert, Amy ; Lemley, Mark ; Boyd, Danah ; Jennifer Wortman Vaughan ; Brundage, Miles ; Bau, David ; Neel, Seth ; Jacobs, Abigail Z ; Terzis, Andreas ; Wallach, Hanna ; Papernot, Nicolas ; Lee, Katherine</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-proquest_journals_31430556583</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Generative artificial intelligence</topic><topic>Machine learning</topic><toplevel>online_resources</toplevel><creatorcontrib>A Feder Cooper</creatorcontrib><creatorcontrib>Choquette-Choo, Christopher A</creatorcontrib><creatorcontrib>Bogen, Miranda</creatorcontrib><creatorcontrib>Jagielski, Matthew</creatorcontrib><creatorcontrib>Filippova, Katja</creatorcontrib><creatorcontrib>Liu, Ken Ziyu</creatorcontrib><creatorcontrib>Chouldechova, Alexandra</creatorcontrib><creatorcontrib>Hayes, Jamie</creatorcontrib><creatorcontrib>Huang, Yangsibo</creatorcontrib><creatorcontrib>Mireshghallah, Niloofar</creatorcontrib><creatorcontrib>Shumailov, Ilia</creatorcontrib><creatorcontrib>Triantafillou, Eleni</creatorcontrib><creatorcontrib>Kairouz, Peter</creatorcontrib><creatorcontrib>Mitchell, Nicole</creatorcontrib><creatorcontrib>Liang, Percy</creatorcontrib><creatorcontrib>Ho, Daniel E</creatorcontrib><creatorcontrib>Choi, Yejin</creatorcontrib><creatorcontrib>Koyejo, Sanmi</creatorcontrib><creatorcontrib>Delgado, Fernando</creatorcontrib><creatorcontrib>Grimmelmann, James</creatorcontrib><creatorcontrib>Shmatikov, Vitaly</creatorcontrib><creatorcontrib>De Sa, Christopher</creatorcontrib><creatorcontrib>Barocas, Solon</creatorcontrib><creatorcontrib>Cyphert, Amy</creatorcontrib><creatorcontrib>Lemley, Mark</creatorcontrib><creatorcontrib>Boyd, Danah</creatorcontrib><creatorcontrib>Jennifer Wortman Vaughan</creatorcontrib><creatorcontrib>Brundage, Miles</creatorcontrib><creatorcontrib>Bau, David</creatorcontrib><creatorcontrib>Neel, Seth</creatorcontrib><creatorcontrib>Jacobs, Abigail Z</creatorcontrib><creatorcontrib>Terzis, Andreas</creatorcontrib><creatorcontrib>Wallach, Hanna</creatorcontrib><creatorcontrib>Papernot, Nicolas</creatorcontrib><creatorcontrib>Lee, Katherine</creatorcontrib><collection>ProQuest SciTech Collection</collection><collection>ProQuest Technology Collection</collection><collection>Materials Science &amp; Engineering Collection</collection><collection>ProQuest Central (Alumni Edition)</collection><collection>ProQuest Central UK/Ireland</collection><collection>ProQuest Central Essentials</collection><collection>ProQuest Central</collection><collection>Technology Collection</collection><collection>ProQuest One Community College</collection><collection>ProQuest Central Korea</collection><collection>SciTech Premium Collection</collection><collection>ProQuest Engineering Collection</collection><collection>Engineering Database</collection><collection>Publicly Available Content Database</collection><collection>ProQuest One Academic Eastern Edition (DO NOT USE)</collection><collection>ProQuest One Academic</collection><collection>ProQuest One Academic UKI Edition</collection><collection>ProQuest Central China</collection><collection>Engineering Collection</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>A Feder Cooper</au><au>Choquette-Choo, Christopher A</au><au>Bogen, Miranda</au><au>Jagielski, Matthew</au><au>Filippova, Katja</au><au>Liu, Ken Ziyu</au><au>Chouldechova, Alexandra</au><au>Hayes, Jamie</au><au>Huang, Yangsibo</au><au>Mireshghallah, Niloofar</au><au>Shumailov, Ilia</au><au>Triantafillou, Eleni</au><au>Kairouz, Peter</au><au>Mitchell, Nicole</au><au>Liang, Percy</au><au>Ho, Daniel E</au><au>Choi, Yejin</au><au>Koyejo, Sanmi</au><au>Delgado, Fernando</au><au>Grimmelmann, James</au><au>Shmatikov, Vitaly</au><au>De Sa, Christopher</au><au>Barocas, Solon</au><au>Cyphert, Amy</au><au>Lemley, Mark</au><au>Boyd, Danah</au><au>Jennifer Wortman Vaughan</au><au>Brundage, Miles</au><au>Bau, David</au><au>Neel, Seth</au><au>Jacobs, Abigail Z</au><au>Terzis, Andreas</au><au>Wallach, Hanna</au><au>Papernot, Nicolas</au><au>Lee, Katherine</au><format>book</format><genre>document</genre><ristype>GEN</ristype><atitle>Machine Unlearning Doesn't Do What You Think: Lessons for Generative AI Policy, Research, and Practice</atitle><jtitle>arXiv.org</jtitle><date>2024-12-09</date><risdate>2024</risdate><eissn>2331-8422</eissn><abstract>We articulate fundamental mismatches between technical methods for machine unlearning in Generative AI, and documented aspirations for broader impact that these methods could have for law and policy. These aspirations are both numerous and varied, motivated by issues that pertain to privacy, copyright, safety, and more. For example, unlearning is often invoked as a solution for removing the effects of targeted information from a generative-AI model's parameters, e.g., a particular individual's personal data or in-copyright expression of Spiderman that was included in the model's training data. Unlearning is also proposed as a way to prevent a model from generating targeted types of information in its outputs, e.g., generations that closely resemble a particular individual's data or reflect the concept of "Spiderman." Both of these goals--the targeted removal of information from a model and the targeted suppression of information from a model's outputs--present various technical and substantive challenges. We provide a framework for thinking rigorously about these challenges, which enables us to be clear about why unlearning is not a general-purpose solution for circumscribing generative-AI model behavior in service of broader positive impact. We aim for conceptual clarity and to encourage more thoughtful communication among machine learning (ML), law, and policy experts who seek to develop and apply technical methods for compliance with policy objectives.</abstract><cop>Ithaca</cop><pub>Cornell University Library, arXiv.org</pub><oa>free_for_read</oa></addata></record>
fulltext fulltext
identifier EISSN: 2331-8422
ispartof arXiv.org, 2024-12
issn 2331-8422
language eng
recordid cdi_proquest_journals_3143055658
source Free E- Journals
subjects Generative artificial intelligence
Machine learning
title Machine Unlearning Doesn't Do What You Think: Lessons for Generative AI Policy, Research, and Practice
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-22T13%3A25%3A45IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest&rft_val_fmt=info:ofi/fmt:kev:mtx:book&rft.genre=document&rft.atitle=Machine%20Unlearning%20Doesn't%20Do%20What%20You%20Think:%20Lessons%20for%20Generative%20AI%20Policy,%20Research,%20and%20Practice&rft.jtitle=arXiv.org&rft.au=A%20Feder%20Cooper&rft.date=2024-12-09&rft.eissn=2331-8422&rft_id=info:doi/&rft_dat=%3Cproquest%3E3143055658%3C/proquest%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=3143055658&rft_id=info:pmid/&rfr_iscdi=true