Etienne Hubert
Facebook AI Research, Paris, France.
Department of Philosophy, Ecole Normale Supérieure, Paris, France.
AI Ethics. 2021;1(4):553-562. doi: 10.1007/s43681-021-00072-1. Epub 2021 Jun 28.
Trust has become a first-order concept in AI, urging experts to call for measures ensuring AI is 'trustworthy'. The danger of untrustworthy AI often culminates with Deepfake, perceived as unprecedented threat for democracies and online trust, through its potential to back sophisticated disinformation campaigns. Little work has, however, been dedicated to the examination of the concept of trust, what undermines the arguments supporting such initiatives. By investigating the concept of trust and its evolutions, this paper ultimately defends a non-intuitive position: Deepfake is not only incapable of contributing to such an end, but also offers a unique opportunity to transition towards a framework of social trust better suited for the challenges entailed by the digital age. Discussing the dilemmas traditional societies had to overcome to establish social trust and the evolution of their solution across modernity, I come to reject rational choice theories to model trust and to distinguish an 'instrumental rationality' and a 'social rationality'. This allows me to refute the argument which holds Deepfake to be a threat to online trust. In contrast, I argue that Deepfake may even support a transition from instrumental to social rationality, better suited for making decisions in the digital age.
信任已成为人工智能领域的一个首要概念，促使专家们呼吁采取措施确保人工智能“值得信赖”。不可信的人工智能的危险往往在深度伪造中达到顶点，深度伪造被视为对民主制度和网络信任的前所未有的威胁，因为它有可能支持复杂的虚假信息运动。然而，很少有工作致力于审视信任的概念，以及是什么削弱了支持此类倡议的论据。通过研究信任的概念及其演变，本文最终捍卫了一个非直观的立场：深度伪造不仅无法促成这一目标，反而提供了一个独特的机会，可借此向更适合数字时代挑战的社会信任框架过渡。在讨论传统社会为建立社会信任而必须克服的困境以及其解决方案在现代社会的演变过程中，我摒弃了用理性选择理论来构建信任模型的做法，并区分了“工具理性”和“社会理性”。这使我能够反驳那种认为深度伪造会对网络信任构成威胁的观点。相反，我认为深度伪造甚至可能支持从工具理性向社会理性的转变，这更适合在数字时代做出决策。
