Former Obama administration advisors say the White House regulatory AI principles announced this week are a good start in many ways, but their simplified mandate to avoid overregulation of private business use is a mistake, and the Trump administration could face an uphill battle in appealing to the rest of the world.
Though the current administration has developed a reputation for blaming the Obama administration when things go wrong and for trying to erase Obama-era policy, when it comes to artificial intelligence, the Trump administration has at times struck a remarkably similar tone to its predecessor. This was evident in the AI research and development strategy plan for federal agencies released in summer 2019. In some instances the same people are driving White House AI policy, like Dr. Lynne Parker, White House deputy CTO and assistant director of AI at the White House Office of Science and Technology Policy (OSTP), who also served in the Obama administration.
The recently released list of 10 AI principles is meant to guide U.S. federal agencies as they consider rules to regulate AI. And White House CTO Michael Kratsios said he wants other countries around the world to adopt similar policies.
When asked how an administration with a less-than-ethical track record can be a leader in the ethical use of AI, Terah Lyons said the president’s record could make it tough to convince others around the world to adopt similar principles.
Lyons shaped AI policy in the Office of Science and Technology (OSTP) during the Obama administration and is now executive director of Partnership on AI, a conglomerate of companies and nonprofits promoting AI policy that benefits humanity.
“For better or for worse, the best way to lead in technology policy or otherwise is from the very top, and without moral leadership coming from the president himself, it may be really tough for the United States to claim leadership and to really drive countries around the world to follow,” Lyons said. “So I think that that certainly can’t be lost from this conversation. I really appreciate the challenge of toiling on specific questions related to science policy as many of the folks at the OSTP are working to do, but I think who is sitting in the West Wing really matters.”
Good and bad AI regulation
Lyons said there’s not much novel about the administration’s regulatory AI principles, which largely continue policy promoting values espoused by the previous administration. She added that she’s happy to see safety, privacy, civil liberties, human rights, fairness, and non-discrimination all explicitly mentioned.
She’s also encouraged to note that the principles call for performance-based rulemaking to ensure laws are able to respond to the emergence of scientific advances or evolutions in state-of-the art methods. And since AI is a widely applied suite of technologies, she thinks it’s important for federal agencies to gain a high level of understanding around the expected performance and accountability of AI.
But she’s wary of one of the principle’s main tenants: avoidance of overregulation.
Lyons said fears that limiting AI will stifle growth overlook the fact that certain types of regulation can actually enable innovative companies.
“I think that the administration or anybody saying that imposing a regulatory limit is best for innovation, that is not always going to be the case,” she said. “This is a nuanced set of questions and considerations, and I think it paints a false dichotomy for anybody to say that regulation is wholesale good or is wholesale bad for innovation or industry or for an ecosystem.”
R. David Edelman agrees there’s no single answer to the question of whether regulation is good or bad or too much or too little. Edelman leads policy research at MIT’s internet Policy Research Initiative and worked in the White House National Economic Council during the Obama administration. He said many or most innovative AI business models, from health care applications to autonomous vehicles, must pass regulatory hurdles before they get off the ground.
“Those in the community that I talk to [about regulation] are probably most concerned with underregulation, but I don’t mean that to suggest that they’re calling for new restrictions on AI,” he said. “There’s actually been a call for more — let’s call it regulatory action — to try to update some of the areas where public policies have fallen out of date relative to what’s possible in the AI space.”
Edelman said he has heard particular concerns around heavy-handed export control by the Department of Commerce. Indeed, last week the Department of Commerce moved to place limits on the exportation of AI software.
To examine issues at the intersection of AI and policy — affecting business, transportation, and labor — last year Edelman and others at MIT hosted the AI Policy Congress, a series of conversations between former Obama and Trump cabinet officials and Organization for Economic Co-operation and Development (OECD) leadership.
Among the key takeaways from the AI Policy Congress, he said, was that we’re not going to solve AI’s public trust crisis without giving engineers clear design requirements for maintaining public trust and law. Another key point is that guidelines are just a start.
“One clear area of agreement between the Trump and Obama cabinet members was that international principles are excellent but inadequate to meet the design requirements, that at the end of the day the economy is going to [make it necessary] to figure all this stuff out, [and] that companies will require design systems that actually meet public trust or at the very least are lawful,” he said.
The U.S., China, and world AI leadership
The modern reemergence of AI has given rise to dozens of AI ethics principles, and today there is a tapestry of prescribed principles and best practices for military, government procurement officers, and nations.
Together with the United States, more than 40 countries signed on to OECD AI ethics principles last year.
The principles can be seen as an alternative to the more conservative approach Europe has taken on issues like privacy, Edelman said. But he argued that this needn’t be the case, since Europeans regulators share concerns about the dystopian vision of AI emerging in China and understand the need to manage growth of AI across the economy.
In the months ahead, EU Commissioners are expected to introduce new AI regulation with more protections for individuals. Some initiatives, like an AI ethics guidelines pilot project and recommendations from EU AI experts, were introduced last year. Members of the European AI community have talked about creating a third way to implement AI in business and society that’s different from the approaches taken by both the U.S. and China and more respectful of individual privacy.
Edelman says the White House appears to be trying to form some broad consensus among allies. However, the elephant in the room is China, a nation largely considered a world leader in AI, with plans to become the world leader in the next decade. He believes the United States needs to begin reaching out to China now.
The current standoff between the two superpowers largely reflects tensions from a trade war and the ongoing detention of hundreds of thousands of Uighur Muslims in Western China. As a response to this humanitarian situation, the U.S. blacklisted Chinese AI startups last fall.
But Edelman said bilateral talks between Beijing and Washington need to begin, and soon, because if there’s one area with substantial potential for mistrust and miscalculation — particularly as it relates to national security — it’s AI.
VentureBeat reached out to a White House spokesperson to ask if the United States has held any direct talks with China over common AI principles but received no response. We will update this story if we hear back.
“It’s not to say we are going to agree on a whole lot, but we had decades of negotiations and discussions with the Soviet Union on issues like arms control and nuclear weapons even if we couldn’t agree on a lot then either. So dealing with AI issues and China should be very high on the to-do list of this administration as it thinks about what the next several years are going to look like in this space, and particularly if they want to create and sustain a model of what trustworthy AI that represents our values is and should be,” Edelman said.
Ultimately, Edelman and Lyons agree incorporating the principles the White House put forward will require a great deal of expertise and effort by those in leadership roles, but the real test is whether agencies will take the mandate seriously. That won’t be evident until well after the White House Office of Management and Budget (OMB) sends a final copy of the memo to federal agencies in two months.
“We won’t really see the real impact of this in all likelihood until the landscape analysis has been conducted by the agencies, and we understand what work and activities might take place as a result of it,” Lyons said. “I think that seeing how this plays out is going to be critically important, and that’s only really going to happen as agencies start actually conducting rulemaking.”