Extending Personhood to Artificial Intelligence Systems: An Evaluation

I. Introduction

There is a global scarcity in the field of regulation for artificially intelligent systems. Ryan Calo applied the teachings of cyber law to robotics, as according to him, robotics combined the volatility of data represented by the internet with the ability to cause physical harm. He concluded that, like the law had to evolve in terms of jurisdiction and liability attribution after the advent of the internet, likewise it will have to evolve again to accommodate the essential features of robotics: embodiment (the capacity to act on the world as distinguished from computers in general and therefore the capacity to cause physical harm), emergence (the adjustment law would have to make for the greater automation of tasks currently performed by humans), and social valence² (anthropomorphism of robots).³

For the purpose of this paper, when the author refers to a strong artificial system, they mean an AI system that is capable of making a liability inducing decision without any human interference and not the anthropomorphic AI robot popularised by science fiction. These could be systems capable of impersonation or influencing human decisions based on fabricated data as would be discussed in the later parts of the paper.

Liability can be ascribed to autonomous systems in three manners:

AI systems as innocent agents:

Treating AI systems as mere innocent agents rejects the fact that the technology has developed in a manner to make decisions independent of human interference that cause legal injury leading to liability. This is a rather myopic view to adopt. Any framework designed for the regulation of artificial intelligence systems should be capable of being flexible enough to take the future advances of technology in stride.

Liability of a Developer for Forseeable faults:

Another popular opinion is to attribute the liability arising out of such autonomous decision making to the developer or the user of the system. This logic seems coherent in the light of weak artificial intelligence systems. However, any legal framework must be made keeping in mind that it is flexible enough to adapt to technological advancements that are at least predicted to take place by the experts in the field. Even as of right now, there are artificial intelligence systems that do not need much human intervention after the design and training stages of development. This means that the decision leading to the legal injury whose liability we are trying to attribute has a very high probability of being made by the system alone. In such a situation, attributing the liability to a party that had no control over the decision making seems counterintuitive.

AI systems as the direct subject of law:

Making AI systems direct subjects of law requires change in our understanding of the jurisprudence of legal personality because these systems are unlike any entity which have been granted personhood by law. Till this point in history legal personality has only been granted on the assumption that all decisions made by the entity were actually made by a person or a body of individuals. Because this will not be the case with an automated system, this model requires an in-depth understanding of the theories of personality as they exist in current jurisprudence and then apply them in relation to automated systems.

Ascribing legal personality means ascription of a set of rights and corresponding duties to the autonomous systems. It must be understood that legal personality is a tool for better governance and not a reward for becoming like humans. Legal personhood is historically a matter of convenience and not recognition of an entity being similar to humans. Such a proposition would have to be balanced, for obvious reasons lest it should cause anthropomorphic issues.

Although the most appropriate system to deal with the rise in artificial intelligence systems is still debatable, but the need for a regulatory system cannot be denied. It would be easier to arrive at the point of conflict with a regulatory mechanism than to arrive there in want of one..

The important question here is if the legal theory developed so far is sufficient to cater to the possibility of ascribing legal personhood to strong artificial intelligence systems. This question arises because the nature of artificial intelligence is so obviously unlike any other entity that has been recognised by law as a person. A strong artificial intelligence would be different from them in one subtle (yet possibly the most important) manner, in that the decisions of a strong artificial intelligence system would not be taken by a human being or a group or association of human beings.

In such a situation, it is only valid to question the relevance of legal theory to a situation that has not yet been encountered by it in the entire history of its existence. If the current legal theory is, in fact, insufficient, there can be two possible approaches. First, to form a new legal theory that can accommodate the ascription of personality to strong artificial intelligence systems. Second, if it is insufficient only to the extent that the current forms of legal personality available in existing legal theory do not, per se, apply to strong artificial intelligence systems, then it must be figured out as to what new form of personality should be ascribed to the strong artificial intelligence system without having to form an entirely new legal theory for it.

II. The Basis for Seeking Personhood for Artificial Intelligence Systems

The law as it stands is made by and for humans. Even a cursory study of laws relating to personhood (even such laws that withdrew personhood from humans, like slavery laws), would show that the law holds humanity (the fact of being a human being) above all else, and rewards it with personality. Hence, while the current legal systems would consider an entity of substantial intelligence as conscious, they would still hold back the grant of personhood on the basis of the fact of being non-human.⁴

For even considering a grant of legal personality to a strong artificial intelligence system, one must first define the three most important aspects based on which such personality is sought: autonomy, intelligence and consciousness. Much like the word personality itself, the vast and vague colloquial meaning for these terms would not be enough for a legal regime, should one be theorised for the ascription of personhood to artificial intelligence systems.

Defining Autonomy

Whatever entity one should consider for seeking personhood from law, they must satisfy the following conditions to be considered autonomous:

1. Once the entity is functional, it is, given sufficient resources, in some capacity able to carry out some action without any further outside input.

2. Once the entity is functional, it is able to make a decision to take an act without a direct command from an input.⁵

The action arising out of such decision that was made without a direct command from an input could be called an autonomous action. The first condition ensures that such entity is capable of performing some action and the second condition ensures that the entity is capable of making the decision to perform such action without any external support. Autonomy is a fact. A system may either be autonomous or not. There is usually no in between. But in cases of complex system, as a strong artificial intelligence system is expected to be, there may be some components that are autonomous and some that are not. So one could hazard saying, that one artificial intelligence system could be more or less autonomous than another artificial intelligence system.

Defining Intelligence

Intelligence can be asserted for an entity if it fulfils the following conditions:

1. An entity is able to program itself autonomously.

2. An entity is able to process information and integrate this information in some form of framework.

Intelligence is innately related to the ability of an entity to learn something that it is not expected to know naturally. In case of a human, the ability to learn anything that they did not know how to do upon birth and in case of computer programs learning to perform a task that they were not programmed to perform. The first condition obviously states that to be intelligent an entity must be autonomous. But such autonomy need only be intellectual and not physical. For example, a person who has experienced an accident resulting in third degree burns on his body, may never be physically autonomous, but he would still be intelligent by virtue of his intellectual autonomy.

As explained in the last section, intelligence and autonomy are separate concepts. To be intelligent is to intellectually autonomous, and not just merely capable of independently performing a task. An intelligent system has the ability to create novel ideas.

Defining Consciousness

Consciousness is the existence of subjective experience.⁶ Note that by the definition of intelligence, this implies that any intelligent being is conscious, as an intelligent individual, by Intelligence Condition 2, must be able to integrate information in some framework, the existence of such a framework implies that there is some subjective experience

III. Ascribing personality to autonomous systems

Going back to the theories of legal personality, one must question whether these theories are sufficiently adaptable for us to consider the possibility to ascribe legal personhood to artificial intelligence systems without developing a new theory of legal personality. For this determination, we will study the theories of personality, in the light of the characteristics displayed by the artificial intelligence systems. This test would be simple:

Step 1: What is the characteristic because of which the entity has been recognised as a person in the eyes of law?

Step 2: Does an artificial intelligence system display this characteristic, as well?

Step 3: If the system does not display the said characteristic exactly, does it display it analogously?

If the answer to either question 2 or question 3 is affirmative, it would mean that the system has like capacity as an entity to be granted some form of legal personhood. Since an artificial intelligence system is technically none of these entities, it would eventually be a new category of person (should we come to the conclusion that it needs to be a legal person at all), but without the need to develop a new theory of legal personality.

Comparison with Natural Persons

First, it is extremely evident that any strong artificial intelligence system that may be developed in the future would not be comparable to a natural person in as much that it would be designed and not naturally occurring. But to decide whether the nature of personhood ascribed to a natural person can also be ascribed to an artificial intelligence system, we must not focus on the difference between how each of them come into existence. While it is clear that their origin would never follow the same course, it is also becoming increasingly probable that artificial intelligence systems would start displaying subjectivity in decision making. This is not to say that they will display subjectivity in the same manner and form as a natural person, but these similarities are what we would focus on.

Perhaps the most comprehensive explanation for legal personality is that it is the capacity for legal relations.⁷ Scholars in both legal philosophies as well jurisprudence wanted to figure out the essence or the underlying requisite for the grant of legal personality to a natural person. Gray thought there can be no right, and therefore no legal personality, without a will to exercise the right. "That a right should be given effect," says he, "there must be an exercise of will by the owner of the right.”⁸

It is interesting to note that this capacity to have a ‘will’ is considered to be present in both an unborn person⁹ and a dead person. In both the cases the exercise of the will and therefore, the rights that are granted corresponding to it are limited in nature. The starkest difference between a natural person, an unborn person, and a dead person and an artificial intelligence system is bearing of life. A natural person and a child in the womb are both living as far as biology is concerned. A dead person has, at some point, borne life. An artificial intelligence system has never, and can never bear life. What it can display, however, is an expression of will.

It cannot be stressed enough that the negation of the existence of something (in this case the existence of will or intelligence in an artificial intelligence system) just because they do not exactly match the form we had known it to exist in so far is a strange and myopic view to hold. It would be akin to a chemistry scholar refusing to recognise the trans-isomer of an organic chemical Q as the chemical Q, just because he has only ever encountered the cis-isomer of the chemical compound Q. As mentioned before it is a myopic view to have and can only serve to hurt the development of theories in the field. Intelligence and will must be recognised as intelligence and will even when they occur in computer programs instead of biological neurons.

If we refer back to the test, we find that a natural person, before, during, and after their life is granted some personality due to their ability to exercise will to claim the rights that come associated with the grant of personhood. An artificial intelligence system displays will to perform tasks in a certain specific manner.

Comparison with Hindu Idol

Hindu idols are given a higher position than mere sculptures. The rationale behind this has been discussed in a Privy Council decision,¹⁰ the idols were deemed to have both interest and will. The interest and will of the idol was to be exercised through a trust formed by its shebaits. The idols for all practical purposes are treated as a living person would be. A description of the same is found in a judgement by Justice Mukherji:

“We need not describe here in detail the normal type of continued worship of a consecrated image, the sweeping of the temple, the process of smearing, the removal of the previous day's offerings of (lowers, the presentation of fresh flowers, the respectful oblation of rice with flowers and water, and other like practices. It is sufficient to state that the deity is, in short, conceived as a living being and is treated in the same way as the master of the house would be treated by his humble servant. The daily routine of life is gone through with minute accuracy; the vivified image is regaled with the necessaries and luxuries of life in due succession, oven to the changing of clothes, the offering of cooked and uncooked food, and the retirement to rest.”¹¹

The difference between a Hindu idol and an artificial intelligence system is that the idol does not possess any intelligence or will of its own. Any decision made in its name, is actually being made by its caretakers, which is usually a trust board. An artificial intelligence system, as discussed earlier, is autonomous, intelligent and conscious, in the terms that those characters have been defined. Second, a Hindu idol, by itself is incapable of causing damage or injury (in the legal sense of both the terms, and not the colloquial sense) to another person, due to lack of any intelligence or autonomy of performing an action that stems from itself. This, again, is not true for an artificial intelligence system which displays both autonomy and intelligence. What reason was there, then, to grant personhood to a Hindu idol? It can be seen that the essence of granting personality to a Hindu idol lies in the paragraph cited from Justice Mukherji’s judgement.

For the test, we can see that a Hindu idol is granted legal personality by virtue of the manner in which it is treated by the natural persons who take care of it. It is treated as a living person, who cannot take care of their physical needs, would be treated. It is bathed, clothed, fed, and put to sleep every day. Essentially it is treated as one would treat a perpetual minor.

An analogy could be drawn from this concept of perpetual minority but not an exactly congruent one. Any artificial intelligence system, even the most rudimentary one, has more capacity to interact with and, therefore, cause damage to the persons in its external environment (one must keep in mind that artificial intelligence systems are always designed to make a task easier or at the very least with a goal in mind). It is not completely infeasible to think of an artificial system that is a perpetual beneficiary of a trust, like in case of a Hindu idol.

Comparison with Corporate Personality

There are three major reasons for the concept of granting personhood to corporations to develop. First, personality is granted to corporations to separate the property of the corporation from its members, or asset partitioning.¹² Second, granting a personality to the corporation makes it significantly easier for the corporation to carry on its commercial transactions due to the ease of entering into contracts. Third, it also makes it significantly easier for the state to form regulations for the governance of corporations, while leaving its internal governance to its members. All of this also makes it feasible for a corporation to sue and be sued.

It must be kept in mind that there is still a provision for the lifting of corporate veil in most jurisdictions for offences committed in the name of the corporation. It is still considered that a corporation is not capable of committing an offence and should one be committed in its name, the natural person who committed the crime should be made liable for the same.

The injuries or damage caused by an artificial intelligence system can be due to one of the following scenarios:

a. It could be the result of a poorly written computer program, making the developer of the code liable.

b. It could be the direct result of a command from the user, making the user of the system liable.

c. It could be the result of an autonomous action of the system resulting from its subjective experience during its training, which would give rise to the need of a liability framework for the system.

The third condition is the one that necessitates the study of the possibility of ascribing legal personality to an artificial intelligence system. It would be unfair and difficult to assign liability from an injury or damage caused by an autonomous action of an artificial intelligence system to the developer or the user of the system.

For the test, we will focus on the first reason that a corporation is granted personality, namely, asset partitioning. There seems to be a similar need to separate the system from its developers and users for the proper assignment of liability. We must remember that asset partitioning is the character of the corporate that actually makes the lifting of the corporate veil possible, in the first place. The separation of the corporation from its members is imperative for the proper assignment of liability in case of the commission of an offence.

Although the law faces a challenge provided by the development of non-human intelligence, it seems to be sound enough in its theoretical foundation making the need to develop an entirely new theory to deal with this redundant. One can see that artificial intelligence systems have developed in an analogous manner and can be easily supported by already developed legal theory. What needs to be done is mere adaptation. It is not as easy as it sounds. It would in fact take a lot of effort and modification to change an entirely anthropocentric legal system to include systems designed by humans. But, it is not impossible and there can be many ways to achieve this end.

Since the legal fraternity is never very welcoming of radical or drastic changes in the law, it only makes sense to draw from existing frameworks and formulate a regulatory regime for artificial intelligence systems. Some such possibilities are discussed in some detail in the next section.

The major problem in developing a regulatory framework is that assigning liability to a system which has more intelligence quotient than a non-autonomous machine but less intelligence quotient than a two year old human being is near to impossible, immediately. This however should not hold the legal fraternity back from developing at least some broad principles which can be relied upon and applied if such a matter involving an artificial intelligence system should arise before a proper legislation for their regulation is enforced.

IV. Suggestions

The author would like to suggest a rudimentary framework where every commercially available artificial intelligence system would need to go through a registration process. By commercially available, the author means an artificial intelligence system that is past its training stage and has been handed over to the users through a commercial transaction. It should not matter if the purpose the use uses it for is commercial or private.

The registered system should be allowed to hold intangible property in its own rights. Many artificial intelligence systems regularly produce images and other copyright eligible work, as would be discussed in the next part of this section. This intellectual property that is regularly churned out by the systems should be utilised, and any royalty accrued out of it should be held by the user in the manner of a trust. Such royalty may be utilised in the upkeep, upgrade, de-bugging, and general maintenance of the system. The user, acting as a trustee, would also be permitted to make decisions relating to property and finances, for the benefit of the artificial intelligence system. This trust would be perpetual in nature as (drawing from the Hindu idol analogy) the system would be perpetually treated as a minor, legally.

For weaker systems, the following set of rules should provide adequate regulation, with minor modifications to suit the specific needs. First, the artificial intelligence system must be made subject to all the laws that apply to its human operator. This rule basically stops the system from cyber bullying, stock manipulation, etc.

Second, the system must clearly state that it is not human. This prevents any one interacting with the system from being taken advantage of. This is to prevent incidents like Jill Watson, which was an AI enabled teaching assistant and fooled students into thinking it was human, in 2016 at Georgia Tech.¹³ Also pro-Trump political bots have been accused of influencing the 2016 Presidential Elections in the USA.¹⁴

Third, an artificial intelligence system cannot retain or disclose confidential information without explicit permission from the source. This would require extreme filters to be built into the system to distinguish what constitutes confidential information, but it is largely considered a measure that cannot be negotiated around.

Fourth, every artificial intelligence or machine learning system must be programmed with an impregnable ‘off’ switch. The exact method of programming such a mechanism can be deliberated upon, but this measure cannot be done away with in order to maintain human control over the system.¹⁵

Another approach to making regulations for weaker artificially intelligent or machine learning systems is through derivation from common law of tort. Under this model the owner should be made strictly liable for the actions or lack thereof of the system owned by him. This can be derived from the law of strict liability established in Reyland v. Fletcher.¹⁶ There are three essentials of establishing strict liability, and these need to be modified to suit the need of the hour and the nature of the object.

V. Conclusion

It can be concluded that despite the large number of personalities evolved and granted by the law, the need for a natural person to be at the helm of decisions remains a major factor. Yet, as discussed in the previous sections, strong artificial intelligence systems display autonomy, intelligence, and the capacity for subjective experience. We have established in the previous sections that every kind of personhood is granted against an essential feature(s). Autonomy, intelligence, and the capacity for subjective experience or consciousness, together form a large part of why any personhood is granted in the first place.

The only major argument against the grant of personhood is that such intelligence or autonomy or consciousness is not human. It is not what one would find inside the brain of a human being. Does this mean that it is should not be recognised as intelligence or autonomy, or that it doesn’t need to be regulated in some manner so that developers and users don’t get implicated for actions they had no direct control over? The answer is clearly no.

Kyle Bowyer¹⁷ makes an argument for the attribution of an ‘electronic’ personality as debated by the European Parliament which would make an electronic person some combination of a legal subject and a legal object. Marshal Willick¹⁸ also seems to reach the conclusion that it may become impossible to draw a valid legal distinction between a human and a computer in the future thereby forcing the computers to be recognised as legal persons. Although his use of the term computer seems redundant in the light of contemporary technology, he does seem to be referring to an intelligent system.

Another example of anthropomorphism of intelligent systems is pointed out by James Vincent¹⁹ where he points out that people often view the argument of personhood almost akin to human right. Legal personhood is not necessarily synonymous with or confined to human beings²⁰. It is not handed out to the most deserving candidate as recognition of morphed version of a human right, but to the most legally convenient one.

Easily the most convenient way to attribute liability to an intelligent system is through the common law of torts, specifically the torts of negligence and nuisance. The rule laid down in Reyland v. Fletcher²¹ that “a person who for his own purposes brings on his lands and collects or keeps there anything likely to do mischief if it escapes must keep it in at his peril, and if he does not do so, is prima facie answerable to all the damages which is the natural consequence of its escape” has been applied to motor vehicles²² and electricity²³ successfully. Considering the doctrine is over a hundred years old, and was being applied to the contemporary technology of that time, it seems only logical to extend its application to AI systems, as proposed by Richard Kemp.²⁴ Similarly the common law tort of negligence is invoked when there is a breach of duty. In the words of Lord Macmillan of the House of Lords ‘the categories of negligence are never closed’.²⁵ This simply means the number of situations where a duty arises and therefore a breach of duty can happen cannot be limited to a list. It is therefore not far-fetched to assume to apply the principle to conflicts arising out of decisions made by autonomous systems.

Keeping all the above in mind, it can be argued that due to the incredible leaps made by computer scientists, it is only prudent to develop the jurisprudence required to adequately deal with it sooner rather than later. The law has a long history of lagging woefully behind new technology. While progress is being made every day as computer scientists all over the globe scramble to create the most intelligent system of them all, the law slumbers quietly in the comfort of the pique that artificial intelligence is always ten years away.