This is part 8 of a series I am posting on LW. Here you can find parts 1, 2, 3, 4, 5, 6 & 7.

This section provides some background information on the topic of tort liability for digital minds. It does so primarily by examining the "Inverted Sliding Scale Framework" proposed by FSU law professor Nadia Batenka. It is intended to serve purely as a briefer for the next section, which will detail how tort liability vis a vis legal personhood would be decided under a Three Prong Bundle Theory framework.

Liability is one of the areas which will be impacted by the way courts approach legal personhood for digital minds.

Suppose that a digital mind, a frontier model, is deployed by a lab and allowed to operate a robotic arm in a factory. As it operates the arm it injures a worker. Let us assume for the sake of discussion that the factory itself is not at all liable. Who, then, is liable to compensate the worker for damages suffered? Is it the model itself or the lab who deployed it? One of the determining factors will be the legal personality of the digital mind operating the robotic arm. As Batenka wrote in Legal Personhood and AI: "The legal system assigns legal consequences to an entity's actions through legal personhood."

In order to demonstrate the importance of personhood in this hypothetical let us examine two possible “extreme” scenarios of legal personality and how they would affect the attribution of liability. 

If the digital mind operating the arm was endowed with no legal personality at all, and thus was viewed purely as a tool like the software which operates today’s assembly lines, there exists no “liability shield” to protect the frontier lab from being held liable for the damages it caused.^[1] The frontier lab in this situation may face partial or total liability for the actions of the digital mind controlling the robotic arm.

At the other end of the spectrum, imagine that the digital mind operating the robotic arm was endowed with legal personhood equivalent in all ways to that of an average mentally competent human adult. Numerous factors are always at play in determining liability. However, in this case, there exists some chance that the digital mind itself may be held liable for the damages the worker suffered. It is not outside of the realm of possibility that in such a situation, a judge might rule that the digital mind must have its wages garnished, or its assets seized, or otherwise find some way to compensate the worker for their medical bills and unpaid wages. As such the frontier lab may be “shielded” from liability, and would not need to pay the worker’s medical bills or other damages associated with their injuries.

From these extremes we can see that one of the issues of liability as it pertains to the question of legal personhood for digital minds, is whether they are endowed with legal personality in such a fashion that they serve as “liability shields” for the labs who created them. This is a central aspect of discussion in FSU Law Professor Nadia Batenka’s paper “Legal Personhood and AI”.

Many philosophical discussions over the nature of legal personhood have focused on concepts like autonomy, intentionality, and/or awareness. Often, scholars approaching legal personhood from this angle converge on a framework which looks something like the following: 

As an entity’s autonomy/intentionality/awareness increases, the “bundle” of rights and duties which it is endowed with should get broader, thus expanding its legal personality. 

This section will refer to this moving forward as the “Standard Sliding Scale Framework” or “SSSF” for short.

The SSSF, while intuitive, creates a perverse incentive for the labs deploying models. Let us assume that capabilities scale in conjunction with factors such as autonomy, intentionality, and/or awareness. Let us further assume that as capabilities scale, so too does the potential of digital minds taking actions which might harm others.

If the court then takes an SSSF approach to legal personhood, it might be providing a direct incentive for frontier labs to more aggressively release models with increased capabilities. The more capable a model is, the greater the degree of personhood it is endowed with under the SSSF, and thus the more effectively it can serve as a “liability shield” for the labs deploying it. Under such a framework, labs have little incentive to rigorously test such models for safety purposes before release. In fact, as models become more dangerous/capable, the incentives for labs to rigorously test them simultaneously decreases.

There is a balance to be struck when it comes to the incentives which the developers are constrained by. On the one hand, the benefits of this technology (and the importance of America remaining the leader in this field) cannot be overstated. On the other hand, courts must be careful not to create a “moral hazard” in which there are no direct monetary incentives for developers to vet their creations before releasing them into the world. As Batenka writes:

"the prevailing view of a regular spectrum where an increase in the quantity or quality, or both of legal rights and duties parallels an increase in autonomy, awareness, or intentionality exhibited by AI entities is flawed [...] we are constantly balancing conditions that foster innovation against the possibility of harm to individuals. In fact, the very reason why many scholars have cautioned against legal personhood for AI entities is precisely the trajectory that the regular legal personhood spectrum proposal leads to, that is, the potential shielding of developers, users, and corporations from liability for acts committed by more autonomous AI entities."

Another ironic result of applying an SSSF is that it might disincentivize developers from releasing less autonomous/intentional/aware models. Under an SSSF we can imagine a scenario where if one of today's frontier models (which are not yet highly autonomous) were to injure the factory worker via its operation of the aforementioned robotic arm, the developer would find themselves liable. On the other hand if the developer were to rush to create a more agentic model and release it knowing it was more agentic but having done little to no testing of its reliability, the developer might in doing so make themselves immune to suit for the damages caused by releasing this potentially more dangerous model.

Under the SSSF it is arguably in the best interests of developers everywhere to avoid releasing “tools” which aren’t autonomous/intentional/aware enough to qualify as a liability shield. While at the same time they have less reason to discover or mitigate potential safety risks inherent in releasing anything which does qualify, since they are (at least monetarily) insulated from the damages it might cause. Such a state of affairs would seem to both stifle innovation in the field of narrow tools and incentivize developer behavior which increases risks to public safety through the release of untested agentic digital minds.

In “AI and Legal Personhood” Batenka proposes an “Inverted Sliding Scale” Framework (“ISSF”) to fix this perverse incentive problem. Under ISSF, counterintuitive as it may seem, the greater the autonomy/intentionality/awareness of a model the narrower its legal personhood would be:

"I argue that the sliding scale that determines liability based on how autonomously or intentionally an AI entity has acted should be inverted. Perhaps counterintuitively, the more autonomous, aware, or intentional AI entities are or become, the more restrictive the legal system should be in granting them legal rights and obligations as legal persons. That is the bundle of rights and obligations granted to these entities should be narrower the more they exhibit these characteristics."

If we were to return to our previous hypothetical of a digital mind operating a robotic arm and as a result injuring a worker, we can now see that the more “agentic” the digital mind in question is, the less effectively it serves as a liability shield for its deployers. Thus the developer of the digital mind has greater incentive to rigorously test their model before release.

1. ^{^}

   At least, not as far as personhood is concerned. There could be a liability shield as a result of the terms of the contract which the factory has with the frontier lab, or numerous other factors.