Should the government regulate artificial intelligence? That was the central question of the first White House workshop on the legal and governance implications of AI, held in Seattle on Tuesday. Attendees included technology entrepreneurs, academics, and members of the public.
“We are observing issues around AI and machine learning popping up all over the government,” said Ed Felten, White House deputy chief technology officer. “We are nowhere near the point of broadly regulating AI … but the challenge is how to ensure AI remains safe, controllable, and predictable as it gets smarter.”
In a keynote speech, Oren Etzioni, CEO of the Allen Institute for Artificial Intelligence, noted that we are still in the Dark Ages of machine learning, with AI systems that generally only work well on well-structured problems like board games and highway driving. He championed a collaborative approach where AI can help humans to become safer and more efficient.
Nevertheless, Etzioni considers it far too early to talk about regulating AI: “Deep learning is still 99 percent human work and human ingenuity. ‘My robot did it’ is not an excuse. We have to take responsibility for what our robots, AI, and algorithms do.”
One of the key aims of the workshop, said one of its organizers, University of Washington law professor Ryan Calo, was to help the public understand where the technology is now and where it’s headed. “The idea is not for the government to step in and regulate AI but rather to use its many other levers, like coördination among the agencies and procurement power,” he said.
Several academics supported the idea of an “information fiduciary”: giving people who collect big data and use AI the legal duties of good faith and trustworthiness. For example, technologists might be held responsible if they use poor quality data to train AI systems, or fossilize prejudices based on race, age, or gender into the algorithms they design.
This is an interesting line of thought – perhaps someday to be known as “the robot decided to do it” defense. It certainly points to another emerging area of cyber risk – which oddly enough is one of the oldest concepts in computing – GIGO. Garbage In, Garbage Out.
So if one were to put (or download?) incorrect waypoints that were then used to program a flight which then ended badly… It’s also interesting when you think of the car manufacturers like Volvo who have indicated that they will warranty their autonomous vehicles eliminating the need for the owner to be insured – at least in certain instances.