EAGER: Can artificial intelligence (AI) invent? And if so, so what?

点击翻译按钮获取中文摘要

Sustained economic development depends upon the intensity and productivity of efforts devoted to inventive activity. There is a growing concern that the productivity of human inventive activity (as measured by the number of patent applications per patent author) is declining, and that this decline may reflect the growing volume and complexity of technical knowledge and the increasing effort needed to identify intellectual novelty. Computer scientists and proponents of Artificial Intelligence (AI) suggest that, because of AI’s ability to execute combinatorial search effectively and rapidly, AI might be helpful in the search for new inventions. This proposition begets two research questions: can AI invent and, if so, will AI invention augment or displace human inventors? The project examines the current and potential inventive capabilities of AI by tapping into insights accumulated from the study, across many disciplines, of how humans have invented throughout our history. This deep historical perspective, coupled with detailed empirical investigations on the nature of invention, has generated a robust narrative about the salient characteristics of the processes underlying human invention. The project brings these insights on the nature of human invention together with assessments by leading AI experts about the type of inventive activity AI can perform now and the type of inventive activity AI could be expected to perform in the foreseeable future. These insights and assessments will inform consideration of whether AI can help mitigate the difficulties encountered by researchers as they search for new knowledge and create new technologies.The research effort includes: (1) convening a meeting of experts on human invention (drawn from industry and academia in fields such as computer science, engineering, economics, anthropology, psychology, and other disciplines) about the inventive capabilities of AI to assess AI’s current inventive capacities and consider what would constitute signals that AI’s inventive capacities are increasing; (2) developing a deeper understanding of combinatorial invention using patent records to classify the types of inventive novelty generated by the recombination of existing knowledge; (3) assessing which AI capabilities match the different types of combinatorial invention revealed in patents. The research will be reviewed at the end of each year by a subset of the experts who will advise on priorities for research in the subsequent period. The research and insights gathered about how AI invention can offset the decline in human inventive productivity will be summarized in “white papers” to be accessible to the general public online; full versions of the papers will be submitted for publication in peer-reviewed journals.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.