This is part 9 of a series I am posting on LW. Here you can find parts 1, 2, 3, 4, 5, 6, 7, & 8.

This section compares Three Prong Bundle Theory with Batenka's Inverted Sliding Scale Framework as covered in section 8, when used as lenses by which to apply legal personhood vis a vis tort liability.

When we imagine how the TPBT approach would compare to Batenka’s Inverted Sliding Scale Framework in practice, we can imagine some situations where the end result would look quite similar. An upgrade which changed an entity from a “tool” to a “legal person” for example, might involve a similar downgrade of potential “rights” for an entity under both frameworks. 

Consider a self-driving car which uses narrow but high quality machine vision software to pilot the vehicle. Under both frameworks it would be considered a tool, as it possesses neither intentionality/autonomy (the metrics Batenka cites) or the capacity to understand rights/duties (the metrics of traditional bundle theory). Imagine then that the car’s software was upgraded to a more generalist digital mind, one capable of piloting the vehicle but also capable of autonomous actions and/or understanding concepts such as rights and duties. Under the ISSF, “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”. Thus in this situation there might actually be a loss of the right to drive.^[1] Similarly under TPBT the moment that the software behind a vehicle gained the capacity to understand concepts like the “right to drive” it would need to demonstrate sufficient capacity to understand/hold to the associated duties and its capacity to have consequences enforced upon it. Absent an ability to do this, it might lose its right to pilot the vehicle, the same way it would under the ISSF. 

Another situation in which both frameworks would treat an entity similarly is that of an entity which is;

• High in autonomy/intentionality,
• Passes the rights and duties prongs of the TPBT,
  But
• Is not vulnerable to courts imposing consequences.

Imagine for example a next generation LLM hosted on a distributed cloud computing network, one which has both a high degree of autonomy/intentionality and is capable of understanding and holding to duties and voluntarily exercising rights. The ISSF and TPBT framework would both be very restrictive to its claims to rights based on legal personality. This would be for different reasons (the ISSF because of its increased autonomy/intentionality, the TPBT because of the lack of capacity to feasibly impose consequences against it), but the end result would be similar. 

This example, however, also demonstrates one key difference between the TPBT and the ISSF, namely the potential for change in legal personality which coincides with improvements in technology. Under TPBT, if technology were invented enabling the enforcement of consequences even on digital minds hosted on distributed compute, said digital minds have a stronger claim to legal personhood/personality. Under the ISSF, this is not so.

TPBT and ISSF generate approximately the same results in practice (if for different reasons) in their handling of low vulnerability but high autonomy/intentionality/capacity digital minds. Under ISSF if a digital mind has high autonomy/intentionality, its potential claim to rights vis a vis its legal personality is substantially restricted. Under TPBT the outcome for such a digital mind would be similar (or perhaps identical), though only because such a mind at least to begin with would not be vulnerable to court/law enforcement imposed consequences. Again, unlike with ISSF, as enforcement technology changes this entity’s legal personality could “broaden” under TPBT.

For most possible digital minds, however, outcomes under ISSF and TPBT differ drastically. Unlike ISSF the TPBT framework does not restrict more autonomous/intentional minds by default, as such in virtually any hypothetical where such a mind would be vulnerable to consequences, one would see greater access to “broad” bundles for said minds under TPBT. On the opposite end of the spectrum, low autonomy/intentionality digital minds which were nonetheless invulnerable to court imposed consequences, would have much less success claiming legal personhood under TPBT compared to ISSF.

Let us briefly discuss the “developer incentives” issue which Batenka focused much of her analysis on. The main thrust of Batenka’s argument regarding the ISSF vis a vis incentives can be paraphrased as;

“If the legal system endows highly autonomous/intentional digital minds with legal personhood to such a degree that they can function as effective liability shields for their developers, then the legal system is incentivizing the deployment of said minds, possibly in a dangerous and untested fashion. If on the other hand the legal system creates the ISSF where more autonomous/intentional digital minds are less effective as liability shields, then developers are strongly incentivized to very thoroughly test any such minds before deployment. Since the latter is the outcome we want (is most aligned with the public interest) we should do the latter.”

When we scrutinize TPBT through this lens, incentives vis a vis liability shields as a result of legal personhood, it is clear that TPBT incentivizes developers in a different fashion. Let us operate from the same assumption that Batenka makes, that a mind being able to serve as a liability shield (as a result of its legal personality) would serve as an incentive for developers to deploy said mind and possibly lead to more aggressive/untested/risky deployment.

What then, are developers now incentivized to do, in order to achieve their desired liability shield, under TPBT? The answer is, develop technologies which guarantee their digital minds are in fact:

1. Capable of passing the first two prongs of the TPBT (rights and duties),
   and
2. Provably vulnerable to court/law enforcement imposed consequences (the third prong).

Compared to the ISSF then, the TPBT provides less of an incentive to develop and deploy narrow “tool” type digital minds. On the other hand, it provides a greater incentive to develop technologies capable of restraining or destroying digital minds. As such the adoption of TPBT in the courts might lead to more investment and developer man hours spent on technology used for the purposes of "control" or "destruction after breach of containment".

1. ^{^}

    Batenka does not provide specifics to the degree needed to say this for certain, but it is a reasonable inference from her framework as described.