artificial intelligence It is developing faster than policymakers can keep up, and gaps in understanding between government and industry can pose challenges to effective regulation. But that doesn't mean companies should develop and deploy AI at will until governments catch up, he said at the “Leading with AI” conference held at Harvard Business School (HBS) on May 7. Speakers discussing regulation warned. Instead, panelists urged companies to take proactive steps in self-regulation and preparing for government regulation.
“You need to disclose [a company] “It's internal transparency for when things go wrong and internal transparency for when governments start mandating transparency,” said Frankfurter Law Professor Noah Feldman. served as an advisor to the committee). “It's coming. There's never been a regulatory system that hasn't relied on transparency as one of its key tools.”
Sponsored by the HBS Alumni Association and the Harvard Institute for Digital Data Design, a research institute specializing in technology and business research, the conference featured a number of speakers who discussed how companies should address labor, upskilling, marketing and more. He argued that there was a need to fundamentally change the way things were done. The era of AI. “This change is not gradual,” said Fitzhugh management professor Tsedal Neely. “We're not tweaking existing structures or processes; we're introducing fundamental changes, new systems, new processes, new structures,” said Mitchell Weiss, Professor of Management Practice at Menschel College. “HBS has been committed to researching and preparing students for these changes.” Last fall, HBS became “perhaps the only business school to require all students to have GPT Plus,” which provides access to his cutting-edge model of ChatGPT, “and we paid for it.” he said.
Conference speakers said one of these fundamental changes will occur in the relationship between business and technology. As companies make decisions about how to self-regulate and maintain records of their use and development of AI, “we can no longer sit in business silos,” said the director of the Nasdaq U.S. Exchange and his M.B.A. said Anita Lynch ’08. “You can't be a business leader without understanding technology.”
Speakers also predicted what government regulation might look like. In a Bloomberg article last year, Feldman outlined a possible approach. The most extreme option would be to nationalize the AI industry, but few people in the United States support this path. More moderate options include criminal law, statutory civil regulations, and administrative regulations, he continued. In this more moderate direction, it is unlikely that another regulator will be created to oversee AI, he said at the conference, noting that existing “regulators will not simply hand over their powers.” Ta. Instead, he expects existing agencies to take over related oversight areas.
Mr. Feldman also outlined an additional option: industry self-regulation. This option has already been pursued to some extent, with companies including Meta and Google voluntarily pledging to comply with AI safety standards in July 2023, but with “tight gates preventing them from doing things differently.” There is no element,” Lynch said. He added that government regulations will set “minimum standards” in the future, but self-regulation and higher internal standards that companies can start establishing now will remain important.
Lynch pointed to another technology that could help guide regulators as they face the unknown. Lynch said that while Stanford University computer scientist Andrew Ng compares AI to electricity, and that industry is tightly regulated, AI regulations should be flexible. “It's difficult to think of different ways to [electricity] It can be used or abused, but there are basic safety protocols,” she said. Regarding AI, “there are industry-specific uses that are difficult to predict and therefore difficult to regulate. But that doesn’t mean we shouldn’t try, and we don’t want to try.” That doesn't mean there aren't any.”
Such endeavors may require some experimentation, especially in modern times with little or no regulation. Bemis Professor of International Law and Computer Science Jonathan Gitrain, faculty director of the Berkman Klein Center for Internet & Society, has set out a policy to encourage data collection and disclosure. For example, Zittrain asks how the government can encourage the disclosure of “something that is not clearly illegal, but is a little bit dangerous, because they don't have the time to identify and pass every possible protection law.” Ta. He added that if a company makes its actions public, “there may be a cap on damages.” Such a policy changes the company's incentives. “I will come forward and give society a chance to address this issue, and I will be protected.”
In the midst of this experiment, Lynch said it's important to educate policymakers so they have the foundation to make such decisions. “Despite the efforts being made to get more technologists and product people involved in government, we don’t know if they will catch up. So the mechanisms that work despite the lack of supply “We need to come up with something,” she continued. “We're going to have to educate people more.”
Zittrain, who has taught courses on ethics and AI, added that not only the exact form of regulation needs to be determined, but also the purpose. “From a regulatory perspective, you have to keep the destination in mind,” he says. “We don't know if that will happen.” Currently, companies such as OpenAI “tune” their models to human values during the training process to ensure they don't produce offensive or harmful output. . For example, when a user asks ChatGPT how to commit a crime, the bot is trained to respond that it cannot provide that information “in a friendly but ultimately reprimanding voice,” Zittrain said. Ta. But are such guardrails really necessary and do they not violate free speech principles, he asked?
“I realized that I was more concerned about tweaking these models to make them more secure than I was about the tremendous uncertainty about what the models were trying to say in the first place. ” he says. “The adjustment process is completely internal, so we don't know what will happen. [companies] Choose to share what that collaboration looks like. ”
Addressing these questions should be a societal process, speakers said, and not limited to technology companies or Congress. And this process offers an opportunity to fundamentally change the way we think about internet regulation. Now, Zittrain said, our system is like olives in a cocktail. It's a large green oval with a little red in it. In other words, “everything that is not prohibited is permitted.” What would happen, he asked, if we switched his logic around and banned everything that wasn't allowed?
As AI continues to advance, it is important for regulators to keep these larger issues in mind, along with concerns about regulating specific technological advances. “From a regulatory perspective, how do we…[keep] The important thing is to think about the map and move the wheels of the bus as it roars along. ”
