Shaping 21st Century AI: Controversies and Closure in Media, Policy, and Research

关于“人工智能”（AI）的讨论比比皆是。政治家、专家和初创企业创始人告诉我们，人工智能明天将改变我们的生活、沟通、工作和旅行方式。自动驾驶汽车、疾病检测、错误信息和仇恨言论的自动过滤——人工智能似乎是为了解决我们社会的根本问题。与此同时，提出了实质性的关切，即这些发展可能会加剧社会和经济不平等，加剧决策过程的不透明，并最终质疑人类的自主权。此外，科学领域本身的方向也存在争议，机器学习的主要实践者公开反对设计和训练这些人工智能体的不同方法的长期重要性。这种动态技术发展和基本争议的结合，为人工智能作为一种社会技术现象如何融入我们的社会进行比较、纵向研究提供了完美的场景。尽管我们之前已经看到了对人工智能的兴趣激增，但21世纪的人工智能目前正处于形成阶段——在公共辩论中不稳定，在专家政策和研究社区中也不稳定。虽然机器学习的进展一直以实验的形式出现，但今天这些“实验”发生在我们的日常生活环境中；这是咨询调查、伦理白皮书、国家人工智能战略以及研究资金大规模重新部署等大量出版物中隐含的潜台词。拟议的项目旨在对比四个关键国家（加拿大、法国、英国和德国）的区域和全球化轨迹。在这些国家内部和国家之间，该项目使用适合每种情况的历史、民族志和计算方法，在三个层面上对人工智能“深度学习革命”在10个形成时期（2012年至2021年）的话语和发展进行比较和联系：媒体、政策空间和研究界。实地工作建立在并扩展了在巴黎政治学院媒体实验室开发的“争议制图”。媒体分析将调查主要新闻媒体、小众网站和社交媒体对话中的人工智能辩论。政策分析将绘制和分析每个国家现有的政策倡议、白皮书和法规，并仔细关注其理由。政策分析将采用精选的专家访谈和社交媒体对话来分析政策发展的背景。研究分析将绘制出版物档案和科学界的地图，并在人工智能与社会问题（例如在线宣传或机器学习模型中的偏见）交叉的相关研讨会和会议上进行人种学嵌入实验。此外，该项目将通过举办参与性研讨会，调查和鼓励公众参与的形式，使利益相关者和公众能够辩论和谈判人工智能的途径。每一项分析和干预都不会以独立的方式进行，而是会反射性地相互通知。这种特殊的研究设计使该项目能够追溯21世纪人工智能是如何被反复构建为一个问题和一个解决方案的：在其短暂的历史和未知的未来中，人工智能文化如何在没有完全定义“人工智能”范围的情况下，跨越争议和（明显的）封闭进行谈判。这项国际、比较、多方法的研究明确致力于公众知识和参与，旨在扩大和重新分配相关的专业知识范围，以确保人工智能的到来真正造福于更大的利益。

Talk about "artificial intelligence" (AI) is abundant. Politicians, experts and start-up founders tell us that AI will change how we live, communicate, work and travel tomorrow. Autonomous vehicles, the detection of illnesses, automated filtering of misinformation and hate speech - AI is seemingly set out to fix fundamental problems of our societies. At the same time, substantive concerns are raised that these developments might reinforce social and economic inequality, exacerbate the opacity of decision-making processes, and ultimately question human autonomy. Moreover, the direction of the scientific field itself is up for dispute, with leading practitioners of machine learning publicly disagreeing about the long-term importance of different approaches to designing and training these artificial agents.This conjunction of dynamic technological developments and fundamental controversies sets the perfect scene for a comparative, longitudinal inquiry into how AI as a sociotechnical phenomenon is being integrated into our societies. Although we have seen spikes of interest in AI before, 21st-century AI is currently in its formative stage - unsettled in the public debate, but also in expert policy and research communities. And while progress in machine learning has long come in the form of experiments, today these 'experiments' take place in our everyday lived environment; this is the implicit subtext of the profusion of publications of consultancy surveys, ethics whitepapers, national AI strategies, and large redeployments of research funding.The proposed project seeks to contrast both regional and globalized trajectories in four key countries: Canada, France, the UK, and Germany. Within and across these countries, the project compares and relates to each other-using historical, ethnographic, and computational methods appropriate to each case-the discourse and developments around AI's "deep learning revolution" over 10 formative years (2012 to 2021) in three layers: in the media, in the policy space, and in the research community.The field work builds on and extends the "cartographie de controverses" that has been developed at the Media Lab, Sciences Po. The media analysis will investigate AI debates in major news outlets, niche websites and social media conversation. The policy analysis will map and analyze the existing policy initiatives, whitepapers and regulations in each country, with careful attention to their rationales. Policy analysis will employ select expert interviews and social media conversations to contextualise policy developments.The research analysis will map publication archives and scientific communities as well as experiment with ethnographic embedding in relevant workshops and conferences where AI intersects with social issues (e.g. of online propaganda or of bias in machine learning models). In addition, the project will investigate and instigate formats of public engagement by hosting participatory workshops that enable stakeholders and members of the public to debate and negotiate AI pathways. Each of these analyses and interventions will not be conducted in a stand-alone manner, but will reflexively inform one another.This particular research design allows the project to retrace how 21st-century AI has been repeatedly constructed as both a problem and a solution: how throughout its brief history and unknown future, AI cultures negotiate across controversies and (apparent) closure without the scope of "AI" ever being fully defined.This international, comparative, multi-methodological study with a clear commitment to public knowledge and engagement seeks to extend and redistribute the range of expertise that is relevant to ensure that the coming of AI is truly for the greater good.