Interview with UCLA Law Professor Eugene Volokh: AI and the Legal Landscape

UCLA Law Professor Eugene Volokh

Recorded on July 20, 2023

The AI Forum recently sat down with UCLA Law Professor Eugene Volokh to discuss the intricate interplay between artificial intelligence (AI) and the legal system. Volokh, a renowned expert in constitutional law, shared his insights on the challenges and opportunities AI presents to the legal profession.

Volokh began by addressing the legal implications of AI-related incidents. He stated, "AI presents a mix of issues. For instance, if I'm hit by an AI-assisted driver, I'd sue the driver and possibly the manufacturer. The same legal rules would apply." However, he also highlighted the complexities of AI in defamation cases, asking, "If I'm libeled by a GPT-4 output, it raises questions like, 'Can I show actual malice or negligence?' We know how to address these questions with humans, but it's different with AI."

On the topic of AI's potential impact on the legal profession, Professor Volokh was cautiously optimistic. "It depends on AI's capabilities. If it plateaus, it might be a tool for drafting. Lawyers would still need to edit AI outputs," he noted, while also warning of the pitfalls, citing instances where "lawyers cite non-existent cases generated by AI."

The professor also envisioned a future where AI could play a pivotal role in arbitration. "Consider a business dispute. You could resolve it traditionally or submit documents to an AI arbitrator. The AI might be faster and cheaper, even if it's not perfect," he mused, recalling an article he wrote titled "Chief Justice Robots," which speculated on the possibility of AI judges.

The conversation also touched upon the contentious issue of copyright in the age of AI. Volokh expressed sympathy for artists and writers who argue for royalties, stating, "I understand their concern. AI tools like Dali can quickly produce art, potentially replacing human artists." However, he also pointed out the potential "fair use" of copyrighted material by AI.

Regarding the recent debate on whether an AI can be recognized as an author under American copyright law, Volokh opined, "I believe a computer program can't be an author. However, someone involved in the process, like the programmer or user, might be." He further elaborated on the nuances of copyright, emphasizing that "the level of creative decisions made can determine copyright, even if combined with uncontrollable factors."

This comprehensive interview with Professor Volokh underscores the evolving landscape of law in the age of AI, highlighting the pressing need for adaptive legal frameworks and thoughtful deliberation on emerging challenges.

___________________________________________________________________________

AI FORUM:

When you think of the challenges of AI and constitutional law and your other areas of expertise, do you think that the AI poses a set of new issues for the law?

Eugene Volokh:

Well, it's a mix of things.  Suppose, for example, that I get hit by an AI assisted driver, not necessarily a self driving car, but just one where some sort of AI technology is helping.   Well, what am I going to do?  I'm going to sue the driver and perhaps also the manufacturer.  The same normal legal rules for a car accident will apply in this suit. At the same time, there are interesting questions that AI uniquely presents.  

In another example, let's say I am libeled by the output of a being that uses GPT-4.  This raises interesting questions, such as:  “Can I show actual malice or negligence?”  “What do those terms mean in the context of an AI product?”  We know how to address the negligence question when a reporter or some other human says something.  The standard would be whether they knew it was false, or likely to be false.  But these traditional legal definitions and categories are not clear at all in the context of the new facts posed by new technologies.

AI FORUM:

Is AI going to revolutionize the legal profession?

Eugene Volokh:

Well, it all depends on how good it ends up being.  Let’s assume, for example,

that AI kind of plateaus because of technological limitations and ends up simply being a decent tool for writing rough drafts of various kinds of things.

We will still need to have lawyers who will edit the output of an AI tool and who will make sure that that the cases it cites are real. We have already seen two incidents in which lawyers have filed motions in court which cited non-existing cases, dreamed up by Chat-GPT.

Due to the time savings of using AI tools, we can assume that firms will not have to staff their projects quite as heavily as some places do today.  Perhaps instead of a partner and four associates, we’ll have a partner and two or three associates.  This will benefit clients, because it's going to sharply decrease their legal bills.  At the same time, this might not be so good for law school graduates looking for jobs.  This all assumes that the technology plateaus, which is not necessarily the case.

AI FORUM:

How about other use cases within the legal profession?

Eugene Volokh:

Okay, let’s explore a specific scenario, though one that assumes major future advances in the technology (and who knows whether the assumption will be borne out). Let's say you and I have a business dispute and somebody comes up and says, “Look, you can do this two ways. You can resolve this dispute using human judges and human jurors, which might take three years and cost a million dollars. Or you can resolve this dispute by taking all the documents and submitting it to my AI, arbitrator for binding arbitration.”

Now, to be sure, the AI may not be perfect.  It might make errors in judgement. But if in this case I can tell you that it's going to take 30 minutes, and it's going to cost you $10,000, perhaps many business people would say that they were perfectly willing to run this risk of error, even if they are informed that the risk of error is 30%. Even if they think that the risk of error with judges and jurors is merely 20%, it will just be so much more efficient that the parties facing arbitration may decide to completely cut out humans. 

This has downstream implications.  Arbitrators might say, “Look, I'm a human arbitrator, I'm just going do the arbitration more quickly and more cheaply by using AI as a first cut. And then again, I'll edit it.”  

But at some point, people will say, “Why are we paying you anything extra? Why are we tolerating any extra delay, especially since there's an extra risk of error coming from your biases as an arbitrator?”  As a result, the deployment of AI may end up transforming things much more substantially than simply cutting down the number of jobs available to lawyers. 

Before Chat-GPT surged, I wrote an article called “Chief Justice Robots.” The question that article posed is, “What should it take for us to accept AI tools as judges? And it was all quite speculative, quite “science fictiony.” It was published in the Duke Law Journal.  It’s still science fiction, but less aggressively futuristic now that it seemed at the time.

AI FORUM:

You probably saw that a very prominent set of hundreds of writers have signed a petition saying that they should get royalties, because AI is being trained on their published works, some fictions of nonfiction, and that AI is going to learn how to emulate their style. Do you have thoughts on that in terms of copyright law? 

Eugene Volokh:

Look, I understand their concern. Let’s make our analysis particularly simple by focusing on the visual arts. My sense is that many graphic designers and artists-- except maybe those who make a living just selling fine art-- are going to be largely replaced when somebody decides to just use an AI tool like Dall-E, which can produce hundreds of different images, basically for free, in minutes.  Yet this process only works because the AI tool has learned from reviewing the output of all of these human artists.  So I can understand why human artists are concerned that they might be losing their livelihood because AI is drawing upon some portion of their material. 

Now, to be fair, human artists, after all, learn their craft by emulating the style of others. That's the educational process, in many ways.  Yet I sympathize with their concern, even though in some cases the de minimis use of copyrighted material might be considered to be a “fair use.”

AI FORUM:

As you know, there was a very narrow ruling about this in the copyright office earlier this year about whether an AI tool can be an author under American copyright law.  What are your thoughts on this?

Eugene Volokh:

Well, I am pretty confident that a computer program cannot be an author under American copyright law. The question is whether some other person who is involved in the process-- whether it's a human being or a corporation, which is a collection of human beings-- can be an author.  One possibility is to say, “Look, I wrote the program that writes all these things or creates all these images. So I'm indirectly the author for copyright purposes.”

It's instructive to consider the Oscar Wilde photograph case, dating back to the relatively early days of photography in the late 1800s.  The court held that if the photographer stages the photograph, dresses Oscar Wilde, poses him in a certain way and points his camera in a certain way, that these creative choices are sufficient to constitute “authorship” under our copyright law.  

Oscar Wilde photo, late 1800s

Editors Note: The Copyright Act of 1865 explicitly extended copyright protection to the nascent medium of photography. However, the idea that "reproductions of the exact features of some natural object" could qualify as original creative works within the meaning of copyright law remained controversial. The Supreme Court noted that the photographer, Sarony, had intentionally crafted his photograph by arranging the background and by dressing, posing, and coaching his model. The Court found that the control Sarony exercised over his subject rendered him th e "author" of an "original work of art" meriting full copyright protections.

I think there's a pretty decent case for treating AI output as similarly copyright-protected to the extent it was guided by specific and detailed prompts, notwithstanding some of the recent statements by the Copyright Office (and I'm not sure how broad those statements really are). If one provides enough creative expression in the prompt that instructs the AI tool, which then gets turned into the final output, this might entitle the prompter to a claim of authorship under our copyright laws.  Time will tell.

Transcribed by https://otter.ai