I watched a very informative webinar, “Generative AI: Ethical, Legal, and Technical Questions,” co-sponsored by the University of Santa Clara’s High Tech Law Institute and the Markkula Center for Applied Ethics. One of the panelists, Professor Tyler Ochoa, indicated three areas in generative AI that are currently being litigated:
Is the use of copyrighted works (i.e., scraping the internet) to train AIs considered infringement?
Professor Ochoa cited the Google case, Authors Guild, Inc. v. Google, Inc., where Google scanned millions of published works to make them searchable by the public, to which a group of authors and publishers responded by suing Google for copyright infringement. In its decision, the Court indicated that under the Fair Use Doctrine, Google could make short excerpts available to the public without infringing on copyrights. Professor Ochoa noted that if the Courts use this case as precedent, they are likely to find that using copyrighted works to train AIs is also not infringing.
Does output from generative AI that is similar to copyrighted works infringe on those works?
It will be interesting to see how the Courts decide on this point.
Can output from generative AI be copyrighted?
Professor Ochoa noted that the Copyright Office recently issued guidance on this issue in Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence. The Office noted,
In the Office’s view, it is well-established that copyright can protect only material that is the product of human creativity. Most fundamentally, the term “author,” which is used in both the Constitution and the Copyright Act, excludes non-humans.
Professor Ochoa noted that the determinative issue here is the amount of human involvement in the process. If the user inputs a query and the AI spits out a product, that product is not copyrightable. However, if the output is generated through an iterative process, in which a human is working with an AI, then the eventual output may be copyrightable.
As indicated, these three areas are all currently being litigated, so more clarity will be forthcoming once the courts rule.
The subject of generative AI and copyright law is a particular example of a more general process that people in the US experience when a new activity is introduced (generally through the development of some new technology): When a new area of activity arises, it will cause a flurry of cases in which many different individuals or groups will feel aggrieved in similar ways. In the case of generative AI, there are going to be a whole lot of copyright holders whose rights will be threatened by other people’s use of generative AI that was either trained on the copyright holders’ works or that generated output that is “too similar” to the copyright holders’ works.
In these cases, the best route for society as a whole is for a clear, deterministic, and publicly announced decision or new law to be made as to how to resolve the issue. This type of action will provide the most efficient and effective means of guiding future activity in the area.
Unfortunately, too often that’s not what happens. In too many cases, multitudes of similar disputes are not resolved in the most efficient way possible. Rather, too often we’re faced with the following.
1. The laws as they are written are often unclear
There is a general trade-off of time and money resources between
(i) Drafting laws that apply to general principals, and
(ii) Drafting laws that are able to anticipate and address all the issues that might arise.
Presumably, legal and regulatory bodies reache a balance by making laws that tilt more towards general principles, and then letting the courts clarify controversies that arise over the application of the laws to specific circumstances.
2. Gaining clarity requires litigation
When a dispute arises over the interpretation of law, the greatest amount of clarity as to how the law applies to that particular situation is achieved through litigation, that is, by going to trial, getting a verdict, and establishing court precedent.
Such precedent can then be used by others who face similar circumstances. In other words, when two parties go to court to settle a matter, the decision reached in that case helps others who find themselves in similar situations. There are thus externalities, or spillovers, associated with litigation, in which the social benefits of litigation are greater than the private benefits they provide to the original parties involved in the adjudicated dispute.
3. Litigation is expensive
Fully adjudicating disputes has become prohibitively expensive for all but those cases that have the most at stake. As a result, most parties choose the less expensive option of settling their disputes out-of-court. However, such settlements do not provide much clarity as to how the law should be interpreted by others in similar circumstances.
4. Resources are wasted in disputing the same issues over and over
While it might not pay the original two parties involved in a dispute to litigate the case, it would be beneficial to society as a whole to have the issue litigated so as to gain the consequential clarity. That is, if you take all the small disputes involving the same legal issue and add up the costs spent trying to settle all the disagreements, the total costs would exceed the costs it would take to litigate the matter and provide general clarity.
(I can’t believe I’m actually saying there’s too little litigation in the US!)
So how do you solve the problem?
Part of the difficulty in addressing this type of issue is that the first time a dispute arises involving the interpretation of a law as it applies to some new issue, we (society) don’t necessarily know that that issue will end up leading to the same type of dispute over and over again.
On the other hand, there are certain issues for which it is (or becomes) clear (at least to some) that many people will eventually need the same type of clarification, as with the three current areas being litigated for generative AI.
Generally, when there is too little of an activity taking place from a social point of view, one of the first proposals for increasing the incidence of the activity is for the government to subsidize it. In this case, it would mean having the government pay part of the legal costs associated with litigating an issue the first time it brought to court, if it is believed that the same issue will arise again and again. In this case, however, there are a couple of problems with this type of solution:
Where will the funds come from that are used to subsidize the litigation? Ideally, everyone who benefits from the clarification of a particular law would be the ones to pay into pot. This clearly will not work in this case, as most people who need the clarification would probably not step forward and contribute, if they know they can free ride off the payments of others.
Who decides which cases should be subsidized, and how much funding should the government make available? These two issues alone are probably big enough to sink this particular proposal.
Another possible solution would be for a special judge or panel to be assigned to make general clarifications to laws that “the experts” agree need further amplification. Perhaps when an issue arises that is believed will end up causing a lot of other related disputes, a petition could be submitted to the designated judge or panel asking that the law at issue be clarified. The judge or panel could then post the issue to the community, and if and when enough other members of the community agree that the law at issue needs clarification, then the judge or board could act.
We seem to be entering the realm of "Blade Runner" and I'm not jazzed.