Rather than gate access to the technology itself, we should focus on how it’s used

While the conversation about AI’s potential risks and how to regulate use of the technology has yet to reach any conclusion, it’s clear that oversight of the use of AI, especially generative AI, is necessary—not only to prevent catastrophic outcomes but, perhaps more important, to ensure everyone who can benefit from this paradigm-shifting innovation does. 

The key question, we believe, is what to regulate. In his May testimony before Congress, OpenAI CEO Sam Altman proposed no one be permitted to work with AI without first obtaining a license. Mr. Altman’s clear-eyed assessment of his own company’s technology, and his desire to collaborate with government regulators, is a refreshing change from what we’ve seen when social media companies’ heads have come before Congress. But while we wholeheartedly support the government safeguarding society against AI’s risks by establishing safeguards around AI, licensing is not the right way to do it. 

Licensing is almost certain to stifle development in unequal measure to the protection it provides. If we require the people and entities working in AI to stop innovating and  apply for licenses from a (currently nonexistent) federal agency—and await review of hundreds of thousands of applications—then the tremendous innovation we’ve seen in just the last several months will more or less cease. This nearly insurmountable limitation will curb not only building  novel and improved applications, but also developing them quickly and inexpensively enough to benefit ordinary people, for whom the stakes are quite real, as soon as possible.

Licensing would have real costs, for real people

Specific to the field of law, it is impossible to overstate the nearly immediate real-world effect AI products can have on the 80% of legal needs of Americans that go unmet—that’s tens of millions of people who go without legal representation every year.

Using generative AI in legal tech—especially this current generation, which holds unprecedented power and promise—enables lawyers to work more quickly, accurately, and efficiently, for more people. Several months’ or a year’s delay in legal AI development could  mean:

• small businesses give up on recouping unpaid invoices, because their suit is taking too long,
• unjustly imprisoned people spend another 12 months jailed, because the legal aid organization looking into the DNA evidence cannot review their case, or
• victims of a wrongful dismissal decide they can’t afford to file another suit against their former employer.   

Regulation should tackle deliberate and inadvertent misuse

Deliberate misuse might take the form of using AI to create misleading, inflammatory, offensive, or dangerous content, designed to compel people to act based on that false information. Sometimes that’s to spread hate, drive election results, or even just to create chaos.

Inadvertent misuse happens in the context of AI applications where bias in AI models can drive inequality. This might come into play when using AI to help with hiring decisions, policing, or housing, to name just a few.

The resources that might have gone into a licensing application process should instead go toward building professional associations among the AI innovation community, government agencies, and the end users they both serve. This coalition’s focus should be to establish rules to ensure people and companies innovating in this space do so responsibly and securely, keep regulations updated, and enforce them.

Even relatively simple rules requiring transparency of source materials, a certain standard of data security, and specific guidelines about protecting users’ privacy would go a long way, and could be established much sooner than could a licensing program.

We can’t wait for regulation to begin regulating

Government absolutely has the responsibility to act, but we’re unlikely to have both rules and the means to enforce them in place soon. In the meantime, those of us innovating with and applying AI have the responsibility to do this work the right way, beginning by prioritizing transparency and security. 

In all applications, it’s vital to give people authority and control over their information, as is making it easy to recognize, access, and exercise that control. This is even more important when the tech in question is new and not yet fully understood, which is the case with the latest, most advanced large language models (LLMs), such as GPT-4. Even the world’s leading AI authorities don’t completely understand how information that’s fed into an LLM might impact future models and answers, or how it could come out in ways that could reveal or compromise private or confidential information about people or businesses. 

In the practice of law, giving people the means to control their data is particularly important. Lawyers have an ethical obligation to protect privileged and confidential client documents and their content, meaning, among other things, they must ensure none of that information is subject to an inadvertent disclosure. Fundamental to the work of building solutions for lawyers is the expectation that any product billing itself as such keeps their information and their clients’ information secure and private, always.

And ensuring control of data use in the practice of law, when the use in question involves AI, is of utmost importance. Any professional legal application built on AI should prevent all information users enter into the product from being seen or saved by those running the underlying LLM, or from being used to train that LLM. A product that cannot guarantee the security and privacy of all users’—and their clients’—data is ultimately of no real use to a legal professional.

Policymakers have a duty to go beyond photo ops

If the goals are—as we think they should be—to ensure that AI systems:

• are transparent and explainable; 
• function in a robust, secure, and safe way;
• have accountability mechanisms; and, most important, 
• are designed in ways that respect the rule of law, human rights, democratic values, and diversity, then …

… the people in Congress creating the policies—not only representatives but also their researchers and staffers—must partner with the people—not only C-level executives but also AI and machine learning engineers and product designers—applying this tech in real products, who see its promise and limitations, and who have contended with the market dynamics around data access, security, and privacy, for starters.

Policymakers should make it a priority to hear from and work closely with people at smaller companies in every industry where AI is already disrupting the way day-to-day work is done. This moment represents a remarkable opportunity for substantive collaboration on a phenomenon that is already changing entire professions, including but certainly not limited to law, science, and medicine. In fact the emergence of generative AI most certainly will significantly affect just about every part of how we live, work, and interact with each other and with information. 

AI can contribute to economic growth and the closing of the gap between haves and have-nots, but translating high-level principles into workable policies requires a deep understanding of the technology that members of Congress don’t (and aren’t expected to) have on their own, which is what makes deep collaboration vital if this revolutionary technology is to benefit all of society, which we believe it has the capability to do.

Next, read about how to get ahead of AI regulations, with best practices to begin today.