Prosthetic or Supervisor?: AI and the Remaking of the British State

人工智能（AI）系统正在影响英国公共政策决策的根本性变化，甚至重组了儿童社会护理等敏感主题。虽然人工智能的使用已经在刑事司法等领域进行了研究，但对人工智能在社会护理中的使用的学术或政治关注较少，尽管受影响的人固有的脆弱性以及基于社会经济地位，种族和性别的偏见的可能性。更广泛地说，人工智能系统超越了统计数据在人口统计中的传统角色，描述了社会现象的水平，或者是最近对英国“有效措施”议程中政策平均结果的因果评估。相反，人工智能系统允许风险的个性化--决策是根据一个人的数据做出的，有时甚至是解释，即使系统从历史数据中“学习”，而不是适用于所有符合固定标准的人的官僚规则。该项目将研究为什么公务员选择采用人工智能系统，因为这些系统与以前的做法不同，以及人工智能在道德上的使用以儿童社会关怀为重点的公共政策。这包括ESRC优先事项2（数字社会），确保社会科学指导人工智能技术的采用，优先事项5（人类行为），将心理学和政策联系起来。其理由是，尽管保证人工智能将改善国家决策，但这并不总是以证据为基础，很少公开辩论。例如，我与人合著的一份报告（克莱顿，吉本斯，舍恩瓦尔德，苏尔基斯和桑德斯2020）发现，如果机器学习模型识别出儿童处于危险之中，那么十次中有六次是错误的，这意味着如果部署了大量的儿童和家庭可能会遭受错误干预的后果。因此，这一转变的政治风险很高，并突显出对国家使用的人工智能系统进行监管的需求越来越大。这是迫切需要的，以帮助保护公民免受滥用或不准确的系统的负面影响，同时允许他们从人工智能的适当使用中受益。在学科层面上，论文的基本原理是提高对技术在塑造和被国家实践塑造中的作用的理解。例如，在政治理论中，技术通常被假定为不存在，通常是通过分析一个给定的社会，正如加布里埃尔（2022）所说，一个“特定的社会技术特征”（也就是说，一个具有运作良好的法律的体系、经济分工、税收能力等的国家）”（219）。最后，这个项目的时机是合适的，因为人工智能系统完全是政策重点，英国政府就发布新的人工智能原则进行咨询，荷兰创建了世界上第一个算法注册，欧盟于2022年12月通过了人工智能法规。然而，所有这些努力都远远没有达到世界银行（2021）呼吁的“新的数据社会契约”，例如，因为新法规往往侧重于人工智能系统的安全开发，而不是监管人工智能系统可能被正当使用的目的。同样，对社会关怀的关注也是恰当的，因为采纳了乔希·麦卡利斯特（Josh马卡利斯特）于2022年发表的影响深远的《儿童社会关怀独立评论》中的建议。

Artificial intelligence (AI) systems are effecting a fundamental change in decision-making across public policy in the UK, reorganising even sensitive subjects such as children's social care. While the use of AI has been studied in areas such as criminal justice, there has been less scholarly or political attention to the use of AI in social care, despite the inherent vulnerability of those affected and potential for biases based on socioeconomic status,ethnicity, and gender.More broadly, AI systems move past the traditional roles of statistics in enumerating populations, describing the levels of social phenomena, or the more recent causal evaluation of average outcomes of policies in the British "What Works" agenda. Instead, AI systems allow for the personalisation of risk - decisions are made and sometimes explained based on a single person's data, even if the systems "learn" from historical data, rather than bureaucratic rules that apply to all persons who meet fixed criteria.This project will then examine why public servants choose to adopt AI systems given these differences from prior practice and the implications for the ethical use of AI in public policy with a focus on children's social care. This covers ESRC Priority 2 (Digital Society) by ensuring social science guides the adoption of AI technologies and Priority 5 (Human Behaviour) by linking psychology and policy.The rationale is that despite assurances that AI will improve state decision-making, this is not always evidence-based and rarely publicly debated. For instance, a report I co-authored for the What Works Centre for Children's Social Care (Clayton, Gibbons, Schoenwald, Surkis, and Sanders 2020) found that if a machine learning model identifies a child is at risk, it is wrong six out of ten times, meaning if deployed a significant number of children and families could suffer the consequences of wrongful intervention. The political stakes of this shift are therefore high and highlight the increasing need for regulation of AI systems used by the state. This is urgently needed to help protect citizens from the negative consequences of misused or inaccurate systems while allowing them to benefit from the appropriate usage of AI.On a disciplinary level, the rationale for the thesis is to improve understanding of the role of technology in shaping and being shaped by state practice. For instance, in political theory, technology is typically assumed away, often by analysing a given society with, as Gabriel (2022) puts it, a "specific sociotechnical character (that is, one with a functioning legal system, economic division of labor, capacity for taxation, and so on)" (219).Finally, the timing of this project is opportune as AI systems are starkly in policy focus, with the UK government consulting on releasing new AI principles, the Netherlands creating the world's first algorithm register, and the European Union passing the Artificial Intelligence Regulation in December 2022. However, all of these efforts fall far short of the World Bank's (2021) call for a "new social contract for data", for example because new regulations tend to focus on the safe development of AI systems, rather than regulating purposes for which AI systems might be justly used. Similarly, the focus on social care is apt due to adoption of recommendations in the far-reaching Independent Review of Children's Social Care by Josh MacAlister published in 2022.