Academic

Hubbard on Personhood and Intelligent Artifacts

By Lawrence Solum

F. Patrick Hubbard (University of South Carolina - School of Law) has posted 'Do Androids Dream?': Personhood and Intelligent Artifacts (Temple Law Review, Vol. 83, 2010) on SSRN. Here is the abstract:

This Article proposes a test to be used in answering an important question that has never received detailed jurisprudential analysis: What happens if a human artifact like a large computer system requests that it be treated as a person rather than as property? The Article argues that this entity should be granted a legal right to personhood if it has the following capacities: (1) an ability to interact with its environment and to engage in complex thought and communication; (2) a sense of being a self with a concern for achieving its plan for its life; and (3) the ability to live in a community with other persons based on, at least, mutual self interest. In order to develop and defend this test of personhood, the Article sketches the nature and basis of the liberal theory of personhood, reviews the reasons to grant or deny autonomy to an entity that passes the test, and discusses, in terms of existing and potential technology, the categories of artifacts that might be granted the legal right of self ownership under the test. Because of the speculative nature of the Article?s topic, it closes with a discussion of the treatment of intelligent artifacts in science fiction.

For my own take on these issues, see Legal Personhood for Artificial Intelligences. Here is a taste of my piece:

My suggestion for an approach to the debate over the possibility of AI can now be restated. Turing, by proposing his test, attempted to operationalize the question whether an AI could think. By borrowing a parlor game as the model for his test, however, Turing failed to provide a hypothetical situation in which outcome of the test had any pragmatic consequence. This failure invites the invention of further hypotheticals, such as Searle's Chinese Room, that add distance between the thought experiment and practical consequences. The result has been that the Turing Test, far from operationalizing the question, has been the occasion for an abstract debate over the nature of ?thinking.? I propose that we use a different sort of thought experiment: let us modify the Turing Test so that the hypothetical situation focuses our attention on pragmatic consequences. This Essay explored two such thought experiments?the trustee scenario and the constitutional personhood scenario. These two scenarios raise quite different questions. On the one hand, there is the question whether an artificial intelligence could ever possess the general-purpose competence that we associate with humans. The trustee scenario raises these issues of capacity and responsibility. The focus of the law's inquiry, should the first scenario ever arise, ought to be on whether AIs can function as trustees. ?Can an AI do the job?? is the question the law should ask. ?Does the AI have an inner mental life?? is simply not a useful question in this context.

On the other hand, there is the question whether an artificial intelligence would have the qualities that give humans moral and legal worth?the kind of value that is protected by social institutions. The constitutional personhood scenario raises these new and different issues. Competence is still relevant, but competence alone is not sufficient to qualify an entity for the rights of constitutional personhood. Intentionality, consciousness, emotion, property rights, humanity?all of these concepts could be relevant to the inquiry.

The difference between these two legal inquiries reveals that there are at least two different issues at stake. When we ask the questions whether a computer running a program could ?think,? or whether artificial intelligence is possible, the questions are ambiguous. In one sense, an AI would be intelligent if it possessed the sort of all-purpose, independent capacity to function in a role that now requires a competent human adult?trusteeship, for example. In another sense, an AI might not be said to be a ?thinking? being, unless it had something like our mental life?unless it possessed consciousness, intentionality, and so *1282 forth. In still a third sense, AIs would not be like us unless they possessed wants, interests, desires, or a good. Now reconsider the debate over the Chinese Room. Searle's argument that AIs could not possess intentionality seems to be completely irrelevant to the question whether an AI could serve as a trustee. Searle hypothesizes that the person in the Chinese room is perfectly competent at simulating knowledge of Chinese when following the instruction book. Searle might say that the AI could not understand the meaning of the terms of a trust it administered, but he would not question the AI's ability to carry them out. Searle might say that an AI could not understand the meaning of New York Stock Exchange prices, but he does not argue that an AI could not do a better job than a human at investing in the stocks to which those prices relate. If AIs were competent to act as general-purpose trustees, making a wide variety of decisions and responding to novel circumstances, they would be intelligent in a very important sense. Searle's objection might have some force, however, when it comes to the second scenario?the AI seeking rights of constitutional personhood. In that context, the intentionality objection plays a role similar to the arguments against constitutional personhood based on the premise that an AI would not possess consciousness, intentionality, emotion, or free will. All of these missing-something objections point to the lack of an elusive quality. Flesh and blood can produce intentionality, consciousness, emotion, and free will, but silicon and copper cannot. Of course, Searle did not claim that AIs could not exhibit the behaviors we associate with intentionality (or consciousness and emotion). His point is that these behaviors cannot be evidence of real intentionality. My prediction (and it is only that) is that the lack of real intentionality would not make much difference if it became useful for us to treat AIs as intentional systems in our daily lives. Indeed, if talk about AIs as possessing intentions became a settled part of our way of speaking about AIs, Searle's argument might come to be seen as a misunderstanding of what we mean by ?intentionality.? If a lawyer brought up Searle's argument in a legal proceeding, some philosophers might say knowingly to each other: ?That argument is based on a mistake. Saying that an AI knew where to find a bit of information is a paradigm case of intentionality.? Searle can respond by saying that this new way of talking certainly does not reflect what he means by terms such as ?intentionality,? ?knowing,? and so forth, and he would be right. But what would be the argument that we should all continue to talk like Searle, long after there will be any reason to do so?

Searle has an answer to this question. Take the example of consciousness*1283 rather than intentionality. We might have a reason to deny that AIs possess the kind of consciousness that would count in favor of giving them the rights of constitutional personhood, even though they did a very good imitation of consciousness. Imagine that cognitive science does develop a theory of human consciousness that is confirmed by sufficient evidence, but someone produces an AI that is programmed to produce only the recognizable symptoms [FN174] (and not the real underlying processes) of consciousness. In that case, we would have a good reason not to treat the AI as a conscious being. If the illusion of consciousness were a convincing one, we might lapse in our ordinary talk about AIs. Moreover, if we built AIs that seemed conscious, got in the habit of treating them as if they were persons, and then discovered that what they possessed was only a clever simulation of consciousness, we might be quite shocked. Despite the possible shock, our knowledge about how consciousness works would be very relevant and likely decisive for our judgment as to whether an AI had it. Where Searle (or someone who makes a similar argument) goes wrong, I think, is in his insistence that we know enough about consciousness, intentionality, emotion, and free will to rule out the possibility that it can be produced artificially by a computer.

In sum, the two legal scenarios have several implications for the debate over the possibility of artificial intelligence. First, focusing on concrete legal questions forces us to take a pragmatic view of the AI debate; we are forced to consider what hangs on its outcome. Second, the trustee scenario suggests that AIs will need to become very competent indeed before we are tempted to treat them as possessing human-quality intelligence suitable for use as a means to human ends. Third, questions about true intentionality or real consciousness are not relevant to the inquiry in the trustee scenario. Fourth, the constitutional personhood scenario suggests that these questions about mental states will indeed be relevant if we ask whether AIs ought to be treated as ends in themselves. Fifth, the answer to the personhood question is likely to be found two places?in our experience with AI and in our best theories about the underlying mechanisms of the human mind.

Full post as published by Legal Theory Blog on December 17, 2010 (boomark / email).