Fifteen Eighty Four

Exposure to the current media culture in almost any sphere – news, entertainment, business, politics, or technology – involves near constant contact with reporting on, and often barely concealed promotion of, artificial intelligence (AI) as a new and rapidly emerging area of computer engineering.  One facet of this coverage is the characterisation of modern AI as a genuinely original producer of verbal and written expression. 

 Far more than a producer of purely physical sounds that resemble human speech or the playback of previously recorded instances of human expression, current descriptions of AI characterise it as capable of authentic authorship.  Even if this characterisation can be challenged on a technical level by pointing out that AI is thus far wholly dependent upon the acquisition of already existing ideas and expression for its seemingly original output, it can be said that the same circumstance exists for the overwhelming majority of human expression as well.  In any case, apart from the details of its technical design and operation, it appears ever more clear that, as a cultural matter, AI is being increasingly perceived and accepted as a legitimate, bona fide, and independent producer of speech in our society.

As J.L. Austin and the theory of speech acts in linguistic philosophy have elucidated, speech is action, and thus, social status as an original and independent speaker also means status as an original and independent actor in our culture.  Status as an autonomous actor means that the entity in question is perceived as capable of self-producing and self-generating actions that may have formal legal consequences.  There is no question that human action in the form of speech alone can result in very significant legal consequences for a human speaker.  The tort of defamation is one obvious example.  The emerging status of AI as an authentic and independent producer of speech means that our various legal systems will inevitably need to confront and work through the legal consequences of this new kind of expressive conduct and these new kinds of speakers.

In the United States, one attribute of status as a bona fide speaker is that the federal constitution provides formal protection against legal sanction for expressive activity.  The First Amendment of that constitution provides, in relevant part, that “Congress shall make no law… abridging the freedom of speech, or of the press.”  The interpretation and practical application of this constitutional prohibition have occupied the American legal system for more than 230 years, and continue to do so.  Now, a new challenge to the ongoing work of interpretation, most particularly of the First Amendment, will be the interpretation and application of constitutional free speech protections to the expressive product of artificial intelligence.

No matter how this new challenge is confronted, or how the work develops, one important thing that is made clear by the very nature of the enterprise is that there exists no approach to constitutional interpretation – no theory of constitutional exegesis – that will provide to decision makers operating in this realm objective and value-neutral resolutions of the many difficult issues that will arise.  The development of a constitutional free speech jurisprudence for AI speakers will not be achieved by recourse to historical analysis of the original intent of the framers and adopters of the U.S. Constitution and its First Amendment.  How could it possibly be adequately understood from that perspective in this instance?  How could the externally expressed and currently accessible views (and, necessarily, only those views) of those who were involved in the adoption of the U.S. Constitution and its First Amendment be definitive in determining the appropriate scope and understanding of federal constitutional protection of a type of speech that was entirely unknown, and probably unimaginable, by those persons at that time?

One standard response to this challenge by some who claim that such an originalist methodology is capable of yielding an objective, modern-viewpoint—independent, discretion-free body of constitutional law is to offer an analysis whereby certain broader general principles and maxims are said to accurately capture the views of the original adopters. These general principles are then transported and applied to current issues so that specific resolutions of these issues are said to be the one logically, deductively determined outcome most consistent with the original views of the founders.

In the case of a great many modern constitutional controversies, including the most significant and impactful, the idea that what is still left and accessible of a historical record now more than two centuries old is capable of supporting the identification of a definitive position that is discernable without the exercise of significant current judgment and value-preference by those doing the discernment is very difficult to embrace and to accept, especially by those not otherwise inclined by their personal social commitments and values to prefer the result of such a claimed analysis.  In the case of something like the question of what formal constitutional protections should be afforded to speech authored by artificial intelligence, a direct originalist approach, no matter how dubious in other circumstances, is literally unavailable, as the issue could not possibly have been contemplated or considered by the original adopters of the Constitution more than 200 years ago.

This leaves recourse to an approach, as described above, that claims to identify by induction broad, generic principles said to be embraced by the original adopters that are then deductively applied to the current matter to yield a supposedly objective interpretive resolution.  As difficult as it is to accept the possibility of a value-neutral, preference-free reading of a very old and necessarily incomplete historical record when the issue is one that may have been directly considered by the original adopters, it is all the more difficult to accept that a process of subtle and abstract induction and deduction can be developed without the pervasive influence of the personal value-preferences and commitments of whomever is engaged in performing that analysis.  The presentation of such an approach, and its asserted result, as being value-neutral, independent of social preferences, and objective when it almost certainly is not, results in a profound lack of transparency on the part of those entrusted with the responsibility of authoritatively making those judgments.

While the existence and availability of an objective means of constitutional interpretation that yields the same independent results regardless of the social values and preferences of the person or persons doing the analysis and the decision making may be highly desirable from a larger political perspective, the fact of its desirability – even its arguably essential role – in the respect deserved by and accorded to constitutional interpretations by decisionmakers in the judicial branch does not mean that such an approach is in fact, in actual reality, available, or even possible.

What then might be a more transparent and more conceptually defensible approach to constructing an appropriate and workable body of free speech jurisprudence for speech authored by artificial intelligence?  Perhaps the best current answer has been one that openly acknowledges the inescapably value-laden nature of the enterprise and encourages formal decision-makers to present transparently and explicitly the social and legal values that were preferred and to explain the ultimate judgments made in terms of the balancing of various likely social costs and benefits.  Such an approach both promotes current transparency in the exercise of formal societal power by judicial actors and also more effectively empowers future courts to understand the basis of prior decisions when changed circumstances and conditions (such as the development of AI-authored speech) require revisiting and possible reconsideration of these same issues.            

In the case of AI authored expression, an important part of this interpretive approach will involve a reconsideration and a renewed analysis of the traditional social rationales thought to justify constitutional level protection of speech and expression – including support for a vibrant marketplace of ideas, support for constitutional self-governance, individual actualization, and safety-value catharsis – as those rationales and anticipated social benefits play out in the context of artificial intelligence authored speech.