How are we regulating ChatGPT and other AI tools?

ChatGPT is only two months old, but we’ve spent the time since it debuted debating how powerful it really is — and how we should regulate it. 

The artificial intelligence chatbot is being used by a significant number of people to help them with research(Opens in a new window); message people on dating apps; write code(Opens in a new window); brainstorm ideas for work(Opens in a new window), and more. 

Just because it can be helpful doesn’t mean it can’t also be harmful: Students can use it to write essays for them, and bad actors can use it to create malware. Even without malicious intent from users, it can generate misleading information, reflect biases, generate offensive content, store sensitive information, and — some people fear — degrade everyone’s critical thinking skills due to over-reliance. Then there’s the ever-present (if a bit unfounded) fear that RoBoTs ArE tAkInG oVeR.

And ChatGPT can do all of that without much — if any — oversight from the U.S. government.

It’s not that ChatGPT, or AI chatbots in general, are inherently bad, Nathan E. Sanders, a data scientist affiliated with the Berkman Klein Center at Harvard University, told Mashable. “In the democracy space, there are a lot of great, supportive applications for them that would help our society,” Sanders said. It isn’t that AI or ChatGPT shouldn’t be used, but that we need to ensure it’s being used responsibly. “Ideally, we want to be protecting vulnerable communities. We want to be protecting the interests of minority groups in that process so that the richest, most powerful interests are not the ones who dominate.” 

Regulating something like ChatGPT is important because this kind of AI can show indifference toward individual personal rights like privacy, and bolster systematic biases with regard to race, gender, ethnicity, age, and others. We also don’t know, yet, where risk and liability may reside when using the tool.

“We can harness and regulate AI to create a more utopian society or risk having an unchecked, unregulated AI push us toward a more dystopian future,” Democratic California Rep. Ted Lieu wrote in a New York Times op-ed last week(Opens in a new window). He also introduced a resolution to Congress written entirely by ChatGPT(Opens in a new window) that directs the House of Representatives to support regulating AI. He used the prompt: “You are Congressman Ted Lieu. Write a comprehensive congressional resolution generally expressing support for Congress to focus on AI.”

All of this adds up to a pretty unclear future for regulations on AI chatbots like ChatGPT. Some places are already placing regulations on the tool. Massachusetts State Sen. Barry Finegold penned a bill that would require companies that use AI chatbots, like ChatGPT, to conduct risk assessments and implement security measures to disclose to the government how their algorithms work. The bill would also require these tools to put a watermark on their work in order to prevent plagiarism. 

“This is such a powerful tool that there have to be regulations,” Finegold told Axios(Opens in a new window).

Tweet may have been deleted
(opens in a new tab)
(Opens in a new window)

There are already some regulations on AI(Opens in a new window) in general. The White House released an “AI Bill of Rights(Opens in a new window)” that basically shows how protections that are already law — like civil rights, civil liberties, and privacy — affect AI. The EEOC is taking on AI-based hiring tools (Opens in a new window)for the potential that they could discriminate against protected classes. Illinois requires(Opens in a new window) that employers who rely on AI during the hiring process allow the government to check if the tool has a racial bias. Many states, including Vermont(Opens in a new window), Alabama(Opens in a new window), and Illinois(Opens in a new window), have commissions that work to ensure that AI is being used ethically. Colorado passed a bill(Opens in a new window) that prohibits insurers from using AI that collects data that unfairly discriminates based on protected classes. And, of course, the EU is already ahead of the U.S. with regulations on AI: It passed the Artificial Intelligence Regulation Act(Opens in a new window) last December. None of these regulations are specific to ChatGPT or other AI chatbots.

While there are some state-wide regulations on AI, there isn’t anything specific to chatbots like ChatGPT, neither state-wide nor nationally. The National Institute of Standards and Technology, part of the Department of Commerce, released an AI framework(Opens in a new window) that’s supposed to give companies guidance on using, designing or deploying AI systems, but it’s just that: a voluntary framework. There is no punishment for not sticking to it. Looking forward, the Federal Trade Commission appears to be creating new rules(Opens in a new window) for companies that develop and deploy AI systems.

“Will the federal government somehow issue regulations or pass laws to oversee this stuff? I think that is highly, highly, highly unlikely,” Dan Schwartz, an intellectual property partner with Nixon Peabody, told Mashable. “It is not likely you will see any federal regulation happening soon.” In 2023, Schwartz predicts that the government will be looking into regulating the ownership of what ChatGPT produces. If you ask the tool to create code for you, for instance, do you own that code, or does OpenAI?

That second type of regulation — in the academia space — is likely to be private regulation. Noam Chompsky likens ChatGPT’s contributions to education(Opens in a new window) as “high tech plagiarism,” and when you plagiarize in school, you risk getting kicked out. That is how private regulation might work here, too.

We may run into a pretty big problem while attempting to regulate ChatGPT on the national level: AI systems can combat the very legislative regulatory system that would put them in check.

Sanders, the data scientist, explained in a piece for the New York Times(Opens in a new window) that artificial intelligence like ChatGPT is “replacing humans in the democratic processes — not through voting, but through lobbying.” That’s because ChatGPT could automatically write comments and submit it in regulatory processes; write letters to submit to local newspapers and comment on news articles and post millions of social media posts every day.

Sanders explains to Mashable a concept called “the Red Queen’s Race” in which someone — originally Lewis Caroll’s Alice — exerts extreme effort only to make no forward progress. If you give an AI defensive and offensive capabilities, according to Sanders, you might get locked in a back and forth similar to a Red Queen’s Race, and it could escalate out of control.

Sanders told Mashable the U.S. could potentially run into a problem: AI lobbyists trying to control the very legislation that is attempting to govern them. “It seems to me that’s likely to be a losing battle for the human legislators,” he said.

“My observation would be that the serious legislation that’s been successfully passed for regulating machine learning in general has been painfully slow and insufficient to keep track with the progress in the field,” Sanders said. “And I think it’s easy to imagine that continuing into the future.”

We have to be careful with how we regulate this, Sanders says, because you don’t want to stifle innovation. So, you could, say, put in more roadblocks for people to submit comments to their legislators, like more captchas. But that could risk making it too difficult for regular people to engage in a democratic system.

“What I think is the most useful response is to try and encourage more democratic participation, try and encourage more humans to participate in the legislative process,” Sanders said. “As AI presents challenges for scale and ubiquity, getting more people involved in the process and creating structures that allow legislators to hear from and be more responsive to real people, is a valid solution for combating that kind of threat of scale.”

ChatGPT is in its infancy, and there are already plenty of ethical issues to take into account with its use. Yet, it isn’t impossible to imagine a future in which sophisticated AI chatbots make our lives easier and our work better without risking the spread of misinformation and the downfall of the democratic system. It just might take a while for our government to put any meaningful regulation into action. After all, we’ve seen this(Opens in a new window) play out before.