此外,法官和风险评估工具在评估再次犯罪风险方面也存在差异。虽然法官量刑考虑的信息一般不受传统证据规则的限制,这些信息可以包括被告的个性和犯罪历史等③在美国,《联邦证据规则》一般不适用于量刑。,但量刑过程本身必须满足自然公正或正当程序的要求④这在美国最高法院审理的Gardner v Florida(430 U.S. 349,359(1977))一案中得到了确认。。因此,法官不太可能根据被告的固有特征,例如他们的父母是否离婚,作出量刑决定[29]。COMPAS所依赖的变量可能会被人类法官认为与再犯罪不相关,这一事实致使法官根据法律作出的决定与算法推断建议的决定不一致。特定案件中机器学习所依赖的数据缺乏透明度,再加上算法本身的不透明,使法官很难改变他们对辅助决策工具的期望,从而确保决策工具的合理使用。
在机器学习应用到政府决策的情况下,不够透明是正当程序权利受到损害的主要原因。在卢米斯案中,威斯康辛州最高法院裁定正当程序得到了保障,因为COMPAS的评分只是法官考虑的众多因素之一[34](p755)。然而,很难确定法官对COMPAS的使用将在自动化的范围内走多远,特别是一些法官可能会仅把它当作裁决的一个很小考量因素,而另一些法官可能担心,基于其他考量而忽略有关危险的“客观”证据将受到公众、政治或上诉法院的批评。因此,很难评估个人决策在多大程度上是基于COMPAS的评分[35](p767-774)。此外,有理由相信分数将获得其本应具有的更大影响力——首席大法官会议(Conference of Chief)等机构对这类系统的赞扬表明,“客观性”的吸引力蒙蔽了司法部门的许多人,使他们看不到这类软件的实际缺陷。
[1]J.Waldron.The Concept and the Rule of Law [J].Georgia Law Review 2008,43(1).See also S.Sedley,Lions under the Throne: Essays on the History of English Public Law[M].Cambridge:CambridgeUniversityPress,2015.
[2]P.P.Craig.Formal and Substantive Conceptions of the Rule of Law: An Analytical Framework [J].Public law 1997.
[3]Report of the International Congress of Jurists,The Rule of Law in a Free Society [R].New Delhi: In⁃ternational Commission of Jurists,1959.
[4]P.Gowder.The Rule of Law in the Real World[M].Cambridge: Cambridge University Press,2016.
[5]L.Bennett Moses and L.de Koker.Open Secrets:Balancing Operational Secrecy and Transparency in the Collection and Use of Data for National Security and Law Enforcement Agencies [J].Melbourne University Law Review,2017,41.
[6]J.Finnis.Natural Law and Natural Rights[M].2nd ed.Oxford: OUP,2011.
[7]W.Eskridge and P.Frickey.Hart and Sacks’s The Legal Process: Basic Problems in the Making and Application of Law[M].Westbury,NY: Foundation Press,1994.
[9]J.Rawls.A Theory of Justice[M].Oxford: OUP,1999.
[10]D.A.Waterman and M.A.Peterson.Models of Legal Decision-making: Research Design and Methods[M].Santa Monica,CA: Rand,1981.
[11]R.E.Susskind.Expert Systems in Law: A Jur⁃isprudential Inquiry[M].Oxford: Clarendon Press,1987.
[12]Commonwealth Ombudsman.“Centrelink’s Automated Debt Raising and Recovery System: A Re⁃port about the Department of Human Services’Online Compliance Intervention System for Debt Raising and Recovery”Investigation Report,2017[EB/OL].https://www.ombudsman.gov.au/__d ... /0022/43528/Report- Centrelinks- automated- debtraisingand-recovery-system-April-2017,2018-11-27.
[13]T.Carney.The New Digital Future for Welfare:Debts without Legal Proofs or Moral Authority? UNSW Law Journal Forum,May 2018[EB/OL].http://www.un⁃swlawjournal.unsw.edu.au/wpcontent/uploads/2018/03/006-Carney.pdf,2018-08-16.
[14]P.Hanks.Administrative Law and Welfare Rights: A 40-Year Story from Green v Daniels to“Robot Debt Recovery”[J].AIAL Forum,2017,89(1).
[15]CCJ/COSCA Criminal Justice Committee.In Support of the Guiding Principles on Using Risk and Needs Assessment Information in the Sentencing Pro⁃cess(Resolution 7,adopted 3 August 2011)[EB/OL].http://ccj.ncsc.org/˜/media/Microsites/Files/CCJ/Reso ⁃lutions/08032011-Support-Guiding-Principles-Using-Risk- Needs- Assessment- Information- Sentencing-Process.ashx,2018-08-15.
[16]State of Wisconsin v Loomis 881 N.W.2d 749(Wis.2016)[E].The United States Supreme Court denied certiorari on 26 June 2017.
[18]Näringsdepartementet, Statens Offentliga Utredningar.“En digital agenda i människans tjänst [A digital agenda in the service of people]”Statens Of⁃fentliga Utredningar [Official Reports of the Swedish Government],Report no SOU 2014:13,2014[EB/OL].https://www.regeringen.se/rattsl ... 2014/03/sou-201413/ ,2018-08-16.
[20]E.Wihlborg,H.Larsson.and K.Hedström.“The Computer Says No!” A Case Study on Automated De⁃cision- Making in Public Authorities[C].2016 49th Hawaii International Conference on System Sciences.
[21]Swedish Tax Agency.“Taxes in Sweden: An English Summary of Tax Statistical Yearbook of Swe⁃den” 2016[EB/OL].https://www.skatteverket.se/down⁃load/18.361dc8c15312eff6fd1f7cd/1467206001885/taxes-in-sweden-skv104-utgava16,2018-09-10.
[24]M.Hvistendahl.Inside China’s Vast New Ex⁃periment in Social Ranking [N/OL].Wired,14 December 2017 at https://www.wired.com/story/age- of- socialcredit/,2018-09-10.
[25]R.Zhong and P.Mozur.Tech Giants Feel the Squeeze as Xi Jinping Tightens His Grip[N/OL].New York Times(online),2 May 2018 at https://www.ny⁃times.com/2018/05/02/technology/chinaxi- jinpingtechnology-innovation.html,2018-09-10.
[26]R.E.Nisbett and T.DeCamp Wilson.Telling More Than We Can Know: Verbal Reports on Mental Processes[J].Psychological Review 1977,84.
[27]J.Burrell.How the Machine“Thinks”: Under⁃standing Opacity in Machine Learning Algorithms [J].Big Data & Society,2016,3.
[28]Senate Community Affairs References Com⁃mittee,Parliament of Australia[R].Design,Scope,Cost-Benefit Analysis,Contracts Awarded and Implementa⁃tion Associated with the Better Management of the So⁃cial Welfare System Initiative(2017).
[33]S.Mistreanu.Life Inside China’s Social Credit Laboratory[N/OL].Foreign Policy(online),3 April 2018 at https://foreignpolicy.com/2018/04/03/life- insidechinas-social-credit-laboratory/,2018-09-10.
[34]State of Wisconsin v Loomis 881 N.W.2d 749[E].Wis.2016.
[35]C.O’Neil.Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy[M].New York,NY: Broadway Books,2016.
[36]M.Kosinski.D.Stillwell and T.Graepel,Private Traits and Attributes are Predictable from Digital Re⁃cords of Human Behavior [C].2013 110 Proceedings of the National Academy of Sciences of the United States of America.