Kunsmatige-intelligensie-veiligheid

Kunsmatige-intelligensie- (KI-)veiligheid is 'n interdissiplinêre veld wat gemoeid is met die voorkoming van ongelukke, misbruik of ander skadelike gevolge wat uit kunsmatige-intelligensie- (KI-)stelsels kan voortspruit. Dit sluit masjienetiek en KI-belyning in, wat daarop gemik is om KI-stelsels moreel en voordelig te maak, en KI-veiligheid sluit tegniese probleme in, insluitend die monitering van stelsels vir risiko's en om dit hoogs betroubaar te maak. Behalwe vir KI-navorsing, behels dit die ontwikkeling van norme en beleide wat veiligheid bevorder.

Motiverings

KI-navorsers het wyd uiteenlopende menings oor die erns en primêre bronne van risiko wat KI-tegnologie inhou^[1]^[2]^[3] – alhoewel opnames daarop dui dat kundiges hoë-gevolgrisiko’s ernstig opneem. In twee opnames van KI-navorsers was die mediaan-respondent optimisties oor KI in die algemeen, maar het 'n 5%-waarskynlikheid geplaas op 'n "uiters slegte (bv. menslike uitwissing)"-uitkoms van gevorderde KI.^[1] In 'n 2022-opname van die natuurliketaalverwerking (NLP)-gemeenskap, het 37% saamgestem of swak saamgestem dat dit aanneemlik is dat KI-besluite kan lei tot 'n katastrofe wat "ten minste so erg soos 'n algehele kernoorlog is."^[4] Geleerdes bespreek huidige risiko's van kritieke stelselfoute, ^[5] vooroordeel,^[6] en KI-geaktiveerde toesig;^[7] opkomende risiko's van tegnologiese werkloosheid, digitale manipulasie,^[8] en wapenisering;^[9] en spekulatiewe risiko's om beheer te verloor van toekomstige agente vir kunsmatige algemene intelligensie (KAI).^[10]

Sien ook

Verwysings

↑ ^1,0 ^1,1 Grace, Katja; Salvatier, John; Dafoe, Allan; Zhang, Baobao; Evans, Owain (31 Julie 2018). "Viewpoint: When Will AI Exceed Human Performance? Evidence from AI Experts". Journal of Artificial Intelligence Research. 62: 729–754. doi:10.1613/jair.1.11222. ISSN 1076-9757. S2CID 8746462. Geargiveer vanaf die oorspronklike op 10 Februarie 2023. Besoek op 28 November 2022.
↑ Zhang, Baobao; Anderljung, Markus; Kahn, Lauren; Dreksler, Noemi; Horowitz, Michael C.; Dafoe, Allan (5 Mei 2021). "Ethics and Governance of Artificial Intelligence: Evidence from a Survey of Machine Learning Researchers". arXiv:2105.02117. {{cite journal}}: Cite journal requires |journal= (hulp)
↑ Stein-Perlman, Zach; Weinstein-Raun, Benjamin; Grace (4 Augustus 2022). "2022 Expert Survey on Progress in AI". AI Impacts. Geargiveer vanaf die oorspronklike op 23 November 2022. Besoek op 23 November 2022.
↑ Michael, Julian; Holtzman, Ari; Parrish, Alicia; Mueller, Aaron; Wang, Alex; Chen, Angelica; Madaan, Divyam; Nangia, Nikita; Pang, Richard Yuanzhe; Phang, Jason; Bowman, Samuel R. (26 Augustus 2022). "What Do NLP Researchers Believe? Results of the NLP Community Metasurvey". arXiv:2208.12852. {{cite journal}}: Cite journal requires |journal= (hulp)
↑ PhD, De-Arteaga, Maria, Machine Learning in High-Stakes Settings: Risks and Opportunities, 2020-05-13, Carnegie Mellon Universiteit
↑ Mehrabi, Ninareh; Morstatter, Fred; Saxena, Nripsuta; Lerman, Kristina; Galstyan, Aram (2021). "A Survey on Bias and Fairness in Machine Learning". ACM Computing Surveys (in Engels). 54 (6): 1–35. arXiv:1908.09635. doi:10.1145/3457607. ISSN 0360-0300. S2CID 201666566. Geargiveer vanaf die oorspronklike op 23 November 2022. Besoek op 28 November 2022.
↑ Feldstein, Steven (2019). The Global Expansion of AI Surveillance (Report). Carnegie Endowment for International Peace.
↑ Barnes, Beth (2021). "Risks from AI persuasion". Lesswrong. Geargiveer vanaf die oorspronklike op 23 November 2022. Besoek op 23 November 2022.
↑ Brundage, Miles; Avin, Shahar; Clark, Jack; Toner, Helen; Eckersley, Peter; Garfinkel, Ben; Dafoe, Allan; Scharre, Paul; Zeitzoff, Thomas; Filar, Bobby; Anderson, Hyrum; Roff, Heather; Allen, Gregory C; Steinhardt, Jacob; Flynn, Carrick (30 April 2018). "The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation". Apollo-University Of Cambridge Repository, Apollo-University Of Cambridge Repository. Apollo - University of Cambridge Repository. doi:10.17863/cam.22520. S2CID 3385567. Geargiveer vanaf die oorspronklike op 23 November 2022. Besoek op 28 November 2022. {{cite journal}}: Cite journal requires |journal= (hulp)
↑ Carlsmith, Joseph (16 Junie 2022). "Is Power-Seeking AI an Existential Risk?". arXiv:2206.13353. {{cite journal}}: Cite journal requires |journal= (hulp)

[:1-1] 1,0 ^1,1 Grace, Katja; Salvatier, John; Dafoe, Allan; Zhang, Baobao; Evans, Owain (31 Julie 2018). "Viewpoint: When Will AI Exceed Human Performance? Evidence from AI Experts". Journal of Artificial Intelligence Research. 62: 729–754. doi:10.1613/jair.1.11222. ISSN 1076-9757. S2CID 8746462. Geargiveer vanaf die oorspronklike op 10 Februarie 2023. Besoek op 28 November 2022.

[2] Zhang, Baobao; Anderljung, Markus; Kahn, Lauren; Dreksler, Noemi; Horowitz, Michael C.; Dafoe, Allan (5 Mei 2021). "Ethics and Governance of Artificial Intelligence: Evidence from a Survey of Machine Learning Researchers". arXiv:2105.02117. {{cite journal}}: Cite journal requires |journal= (hulp)

[3] Stein-Perlman, Zach; Weinstein-Raun, Benjamin; Grace (4 Augustus 2022). "2022 Expert Survey on Progress in AI". AI Impacts. Geargiveer vanaf die oorspronklike op 23 November 2022. Besoek op 23 November 2022.

[4] Michael, Julian; Holtzman, Ari; Parrish, Alicia; Mueller, Aaron; Wang, Alex; Chen, Angelica; Madaan, Divyam; Nangia, Nikita; Pang, Richard Yuanzhe; Phang, Jason; Bowman, Samuel R. (26 Augustus 2022). "What Do NLP Researchers Believe? Results of the NLP Community Metasurvey". arXiv:2208.12852. {{cite journal}}: Cite journal requires |journal= (hulp)

[5] PhD, De-Arteaga, Maria, Machine Learning in High-Stakes Settings: Risks and Opportunities, 2020-05-13, Carnegie Mellon Universiteit

[:3-6] Mehrabi, Ninareh; Morstatter, Fred; Saxena, Nripsuta; Lerman, Kristina; Galstyan, Aram (2021). "A Survey on Bias and Fairness in Machine Learning". ACM Computing Surveys (in Engels). 54 (6): 1–35. arXiv:1908.09635. doi:10.1145/3457607. ISSN 0360-0300. S2CID 201666566. Geargiveer vanaf die oorspronklike op 23 November 2022. Besoek op 28 November 2022.

[7] Feldstein, Steven (2019). The Global Expansion of AI Surveillance (Report). Carnegie Endowment for International Peace.

[8] Barnes, Beth (2021). "Risks from AI persuasion". Lesswrong. Geargiveer vanaf die oorspronklike op 23 November 2022. Besoek op 23 November 2022.

[:13-9] Brundage, Miles; Avin, Shahar; Clark, Jack; Toner, Helen; Eckersley, Peter; Garfinkel, Ben; Dafoe, Allan; Scharre, Paul; Zeitzoff, Thomas; Filar, Bobby; Anderson, Hyrum; Roff, Heather; Allen, Gregory C; Steinhardt, Jacob; Flynn, Carrick (30 April 2018). "The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation". Apollo-University Of Cambridge Repository, Apollo-University Of Cambridge Repository. Apollo - University of Cambridge Repository. doi:10.17863/cam.22520. S2CID 3385567. Geargiveer vanaf die oorspronklike op 23 November 2022. Besoek op 28 November 2022. {{cite journal}}: Cite journal requires |journal= (hulp)

[10] Carlsmith, Joseph (16 Junie 2022). "Is Power-Seeking AI an Existential Risk?". arXiv:2206.13353. {{cite journal}}: Cite journal requires |journal= (hulp)

[1]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]