Jeff Alstott, Senior Information Scientist, RAND Corporation
Progress in AI has advanced rapidly in recent years, leading to expanded debate among the experts about its potential risks. Although AI has the potential to transform entire industries, it could also pose novel threats to national defense and homeland security. Unless society puts in effective guardrails, broadly capable AI systems could hasten the design and proliferation of bioweapons, cyberweapons, nuclear weapons, progressively more general intelligence and other threats not yet conceived. In addition, domestic oversight, although essential, will not be sufficient alone. We must cooperate with our allies and partners and communicate responsibly with our competitors and adversaries.
Over the decades, we've unfortunately seen multiple instances of non-state actors attempt to use bioweapons that really have strategic scale to them. Fortunately, they have never succeeded. Unfortunately, the barriers are going down.
Most of AI can in principle be handled by the current setup of government, with a few exceptions. One is that if someone is making or deploying an AI that is predictably going to get millions of people killed, there is no part of government that has clear authorities and responsibility for addressing that. So, that needs to be created. There are several places that it would be logical to create it, an independent agency is one, DHS, which this committee works with, is another. There's also a Department of Commerce, particularly the Bureau of Industry and Security. There's also D.O.E., which has a lot of existing relevant authorities that could synergize there.
Wherever it is that the Congress chooses to put it, it needs to have the authorities to be able to say, "this is a problem and we're not going to let that AI go out" and needs to have the responsibility to understand this at a technical level. Thankfully, no part of government has to work alone. First, they have all the rest of government to work with, but also all of American society.