EAGER: AI-DCL: Measuring and Mitigating Animosity toward Artificial Intelligence Systems and Science

人工智能（AI）应用正在取得不可否认的成功并得到广泛采用，但算法偏见的问题也出现了。当人工智能系统做出不公正的决定时，就会出现算法偏见。人工智能系统对大多数人来说是不透明和不熟悉的，这使得这个问题变得更糟。科学家和公众希望释放快速智能计算机系统的力量，但是当他们看到这些系统不公平地歧视或简单地做出无法解释的决定时，他们会感到犹豫，这是可以理解的。这项研究将回答一些基本问题，以帮助理解公众对人工智能的信任和不信任：公众对人工智能系统的感受如何？他们如何看待这些系统的塑造者？当他们获得人工智能系统的经验时，这些态度会改变吗？这项研究的目的是直接衡量公众对人工智能系统和科学家的看法，并检验一个假设，即接触可解释的人工智能会导致更积极的态度。将人工智能系统整合到决策过程中，这在以前是人类判断的唯一领域，这仍然是一个相对较新的现象。公众舆论很可能处于一个动态阶段，衡量公众对人工智能的态度在未来几年的演变至关重要。因此，该项目将通过调查美国人口的代表性样本，编制对人工智能系统和科学家的月度综合信任指标。此外，目前正在花费大量精力使AI系统更具可解释性。这种努力是基于一个未经验证的假设，即对人工智能的负面态度是由于其底层算法的复杂性和不透明性。研究人员将测试人工智能系统的第一手经验的效果，同时通过实验控制透明度。我们的目标是对这一理论进行明确的测试，即增加对可解释人工智能的接触将减少对人工智能的不信任和其他负面态度。通过这项研究，研究人员将测量和测试一种方法，以减轻对人工智能的不信任，并推进目前在社会科学和工程学科中进行的关于人类将如何与强大的新工具联系起来的对话。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Artificial Intelligence (AI) applications are demonstrating undeniable success and widespread adoption, but the problem of algorithmic bias has arisen. Algorithmic bias occurs when AI systems produce unjust decisions. This problem is made worse by the fact that AI systems are opaque and unfamiliar to most people. Scientists and the public want to unlock the power of fast, intelligent computer systems, but they are understandably hesitant when they see these systems unfairly discriminate or simply make unexplained decisions. This research will answer fundamental questions to assist in understanding the public's trust and mistrust of AI: How does the public feel about AI systems? What do they think of the shapers of these systems? And do these attitudes change as they gain experience with AI systems?The aim of this research is to directly measure public opinion regarding artificial intelligence systems and scientists and test the hypothesis that exposure to interpretable AI will lead to more positive attitudes. The integration of AI systems into decision-making processes previously the sole domain of human judgement is still a relatively novel phenomenon. Public opinion is likely in a dynamic phase and measuring how public attitudes toward AI evolve over the next few years is crucially important. Thus, this project will compile monthly composite measures of trust in AI systems and scientists through surveying a representative sample of the US population. Additionally, it is the case that much effort is currently being expended to make AI systems more interpretable. This effort is predicated on the untested assumption that negative attitudes toward AI are due to the complex and opaque nature of its underlying algorithms. The investigators will test the effect of firsthand experience with AI systems while experimentally controlling the level of transparency. The goal is an explicit test of the theory that increased exposure to interpretable AI will decrease distrust and other negative attitudes toward AI. With this research, the investigators will measure and test a method to mitigate mistrust of AI and advance the conversation currently taking place across social science and engineering disciplines regarding how humanity shall relate to a powerful new tool.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.