Even though AI is a grab-bag of things, Congress still seeks it out everywhere and anywhere. The record of bills introduced into the 117th Congress this year, searchable on Congress dot gov, is heating up with lots of proposals for how AI should be used, accepted, and considered. The number of bills introduced that contain AI swelled over the past few years from almost nothing to dozens each year. Consider some recent motions. Arizona representative David Schweikert (Republican) proposed the ‘‘Healthy Technology Act of 2021," H.R. 5467, which proposes to have the Federal Food, Drug and Cosmetic act amended to include artificial intelligence and machine learning among the “professionals” who are authorized to prescribe drugs. Why is that “healthy” technology? The very short proposal doesn’t say. A machine learning program would still need to be authorized by each State, the bill stipulates. South Dakota senator Mike Rounds (Republican) proposed the “Department of Defense Artificial Intelligence Act of 2021,” S.2904, which requires the Secretary of Defense to conduct a review of the Department’s “platforms” and to report on a few different things, including a “comprehensive review of skill gaps in the fields of software development, software engineering, knowledge management, data science, and artificial intelligence”; “assess investment by the Department in artificial intelligence innovation”; and review the prospect of “the integration of artificial intelligence into war-games, exercises, and experimentation.” Massachusetts representative Jake Auchincloss (Democrat) proposed the “United States–Israel Artificial Intelligence Center Act,” H.R. 5148, which seeks $10 million in annual funds for the National Science Foundation to establish a center to “leverage the experience, knowledge, and expertise of institutions of higher education and private sector entities in the United States and Israel to develop more robust research and development cooperation” in a variety of ML tasks such as computer vision and data labeling. None of these bills define what AI is. They throw in the term as if it were self-evident. AI as a placeholder term is so vague and open-ended that substantial disagreement exists within computer science of just what can and should be included. It is safe to say that today, AI can be used to mean just about anything that a vendor claims as AI. A more thoughtful kind of AI proposal at least attempts to point to education and rigor in introducing the term. For example, the “Artificial Intelligence for the Military Act of 2021,” S.1776, introduced in May by Ohio senator Rob Portman (Republican), is mainly about developing a curriculum. The bill specifies that the head of the Army and that other military services should “expand the curriculum for military junior leader education to incorporate appropriate training material related to problem definition and curation, a conceptual understanding of the artificial intelligence lifecycle, data collection and management, probabilistic reasoning and data visualization, and data-informed decision-making.” The bill stipulates that “Whenever possible, the new training and education should include the use of existing artificial intelligence-enabled systems and tools.” Such a bill acknowledges the complexity of the topic, rather than using the term to refer to something that can be pulled off the shelf. Marvin Minsky, a pioneer of artificial intelligence along with McCarthy, spoke of what he called “suitcase words” used to describe intelligence, words such as “attention, emotion, perception, consciousness, thinking” — words that people use to group together a bunch of notions they want to use without having to define anything precisely. Artificial intelligence has become the ultimate suitcase word. Stuck into legislation, it can mean anything and everything, and it probably will. This raises the question of what these government initiatives will ultimately yield. Maybe some AI, probably a lot of other stuff that’s just packed into the suitcase.
