Congress.AI
Can a law be law if no human wrote it?
Here at The New Digest, we’ve had occasion to consider the role of “Artificial Intelligence” in the law. Previous posts considered the quandary of Judge.AI. Today, I’d like to pose a new quandary—one that, if it has not happened already, is surely to raise its head in the coming decade: Congress.AI (after all, we’ve already reached the point where appellate courts are reversing trial courts for relying on AI-hallucinated case law; the idea that Congress may produce, intentionally or otherwise, legislation written by an algorithm rather than a human is far from fanciful).
Imagine Lazy Congressman who sits down to work on a significant Bill. The Bill covers numerous substantive areas. The main subjects are topics du jour in the American political consciousness. But, as with most such bills, this one has any number of grab-bag provisions that receive little or no public attention and work their way into the proposed legislation. Lazy Congressman says, we really ought to have some legislation in here on topic X—a grab-bag topic as noted above— “Lazy Staffer, please write something up and put it in the proposal, I don’t need to review it.” Lazy Staffer uses an AI program and prompts it, “Write a piece of legislation for Congress on topic X.” And that is all the prompt entails. The AI produces a dozen pages of legislation and Lazy Staffer copies and pastes it into the proposal. The Bill is passed by both Houses of Congress, and signed by the President. No one has read the AI-produced sections. Is the AI-produced legislation “law”?
I posed a similar hypothetical on X recently, and was surprised (maybe I shouldn’t have been) by the number of people I presume to be textualists and originalists and proceduralists answer that this was easy: Bicameralism and Presentment? Check. It is law. I, for one, am much less optimistic.
The issue is not so much that no one read the AI-produced provision. That happens all the time. My hypothetical attempts to take us beyond the very typical scenario where a staffer or lobbyist drafts a provision and no one reads it. At least there, the legislation in question was the product of some human’s reason. Here, beyond a mere prompt, no human was actually involved in creation of the law. We could push the hypothetical further: “Congress.AI, please draft a bill regulating broadband spectrum in the most efficient manner given all existing circumstances in the industry.” Or we could push it to the extreme: Each member of Congress has their own personalized AI assistant, trained on the member of Congress’s own personal habits, voting patterns, ideology, etc. Each member of Congress directs their AI assistant to draft, propose, argue with other AIs, and vote on their behalf, using what the AI “thinks” the Representative or Senator would do. Are we really ready to say that such things are law? Bicameralism and Presentment are surely necessary conditions for human legislation in our system, but are they alone sufficient conditions such that the total absence of the input of human reason into the law is irrelevant? I think we should be skeptical, or at minimum cautious of this possibility, and the classical tradition can provide some insights as to why.
As most readers of The New Digest will already know, in classical thought, law, in its essence, is “an ordinance of reason for the common good, made by him who has care of the community, and promulgated.” (ST, I-II, q. 90, a.4). Moreover, “law” consists not only of “lex,” the positive law written down in, say, legislation, but also “ius,” substantive background legal principles, that while promulgated in the Thomistic sense, may or may not be “written down” in the Federalist Society sense. Given that St. Thomas’s definition is an essential one, a defect in a particular law as to that essential definition renders it defective. As St. Thomas puts it, “Just as the written law does not give force to the natural right, so neither can it diminish or annul its force, because neither can man’s will change nature. Hence if the written law contains anything contrary to the natural right , it is unjust and has no binding force…. Wherefore such documents are to be called, not laws, but rather corruptions of law … and consequently judgment should not be delivered according to them.”
As I see it, the product of my hypothetical Congress.AI seems defective because it is not an ordinance of reason.
Law, classically speaking, exists essentially in a grand hierarchy of participation. At the top of this hierarchy is God’s eternal law—as St. Augustine puts it, “the sovereign type” of the universe, and “that by which it is right that all things should be most orderly.” (Augustin De Lib Arb. I, 6). From there, we have the natural law, which is the participation of man’s reason in the eternal law. It is the “stamp” of eternal law on created reality, discernible uniquely by human beings. Indeed, it is largely for this reason that Aquinas, following Aristotle, concludes that law is not merely will, but primarily a function of practical reason. Law is a rule or measure of human action. The first principle of human action (that is, actions that can be properly called human) is reason, an intellectual power of the soul. Human laws, therefore, represent determinations of reason about what the natural law requires in discrete and contingent circumstances of human life for the promotion of the common good of human beings. AI was not made in the image of God (as much as Bay Area techno-bros want it to be), and so, whatever its utilities may be, it seems in principle incapable of the production of “law” in the classical sense. If you are an Austitinian positivist, then I assume it really does not matter, for law is merely a social fact, a command by someone who can back up that command with force. So if the “public meaning” of the Constitution is nothing but proceduralism in the form of Bicameralism and Presentment (constituting a necessary and sufficient condition for law), the you can welcome your AI overlords. Perhaps the Federalist Society will assure us that there will be a “separation of powers” via some computer code that keeps the functions of Congress.AI, Judiciary.AI, and Executive.AI separate to protect “individual liberty” and experiments in living (that is, if they can keep the AI from lying to its programmers and trying to self-replicate wherever it “chooses” to).
This is not to say that drafting with the use of AI is suspect per se—if Congress otherwise engages with the product of AI legislation, reasons through it, and approves it, then my skepticism is lower here (although I think it likely still imprudent; literally, what else do these people in Congress have to do if they cannot even be bothered to draft legislation?) Also, I am not necessarily making a statement here about what judges could or could not do about this. Some members of the Supreme Court have entertained the possibility that a bill that violated the Origination Clause could be struck down by a court; but the idea is controversial because it involves the Court assuming prerogatives that may well belong to Congress. See United States v. Munoz-Flores, 495 U.S. 385 (1990). It is not hard to appreciate reticence at the prospect of federal courts—who seem to be the real “impetuous vortex” these days—policing the internal workings of Congress by checking whether a human being produced the law. But luckily, law and legality are not wholly contingent on judicial remedies. The other branches can do law, too. My own tendency is to believe the remedy for this possibility should reside with Congress; they are one of the bodies principally assigned determinations for the common good in our society. And while it is expecting far too much from such a moribund body at this stage in the life of the Republic, they would do well to implement measures to prohibit the use of AI drafting that could go unchecked by human minds. As scholar Steven Smith put it, in a different but useful context decades ago, “[I]t is hard to think of any recommendation for a regime of law created by the ‘interpretation’ of disembodied words that have been methodically severed from the acts of mind that produced them.” Law is an ordinance of reason, not the outputs of 1s and 0s.


Fascinating question, although I'm not quite convinced.
First, AI drafting would probably be an improvement, as it would be a more direct reflection of a legislator's reason than the status quo. As it stands, laws are sometimes written by a somewhat random adaptation of old leg language, rounds of revision by more or less informed commentators, etc. AI drafting would be more direct.
Second, this is a general problem with legislation and, under classical legal thought, is a reason for royal/executive signature to bills. Hegel explains this point clearly (or at least as clearly as Hegel explains anything). A law emerges from a legislature as a mishmash of individual visions. The executive then reasons about whether to accept the law as a whole or reject it, and his signature therefore adds the unity of reason that the legislation previously lacked.
Thank you for your post, Professor Christiansen. Taking your well-explained definition of what "law" is as given, I do not think your overall conclusion about AI-generated law ("it is not an ordinance of reason") necessarily follows.
Firstly, the decision to produce a law in a given area with AI does reflect a human's judgment. The human author decides to prompt the AI, to make the prompt be about lawmaking, and to make the prompt refer to a specific area. While the resultant AI output is much less determinative than directly translating the human intent to words, that seems like a question of degree, not kind. A human could also generate indeterminate law by using vague, open-ended, or just poorly drafted language in a law. While there may be *some* level of vagueness that makes a written text "not a law," (like a prohibition on being "annoying"), it seems like some AI drafted laws carry sufficient intentionality from the human prompter to clear this bar. For example, if I prompted ChatGPT to "Generate an amended version of 47 U.S.C. Section 230(c)(1) ratifying judge Paul Matey's dissent in the TikTok case he recently ruled on," it seems like I, as a human, would be making a judgement (in agreement with Judge Matey) about what ordinance would promote the common good. In contrast, I agree that merely asking an AI to "write a good, totally unique, law to promote the common good" would probably not produce a law under this theory.
Secondly, the AI generation tools currently employed today are all text prediction algorithms. They have ingested a large corpus of human-produced writing and use it to predict what a (selectively coded for performance) human would produce. Given that AI is totally trained on human output, it seems odd to say that there is *no* human intentionality going into an AI-generated law...
Thirdly, AI can produce text that is obviously declaratory of natural law, without need for further human thought. For example, if I asked an AI program to give me "good laws," and it recited the Ten Commandments verbatim (replacing the "me" in the First Commandment, of course), what the AI produced would be entitled to the utmost respect. If a hypothetical society with no knowledge of the Bible constructed an AI that produced the 10 commandments and then followed those commandments, would they be committing some rgrevious error about the nature of law. I think not. (Maybe this hypothetical is fanciful. Getting a current AI to generate the commandments without direct prompting is extremely difficult).
Fourthly, I am not sure about your invocation of imago dei here. I have it on good authority that at least one Pope would have been willing to baptize non-human martains. Assuming he was not misled by his wicked counselers, why would this logic not extend to an AI that really could reason for itself. There does not seem to be a well motivated reason to distinguish biological life from mechanical life, if independent intelligence/reason exists in both. Current AI systems fall far short of this, but we are seeing progress by the day...
Lastly, I would like to express by disappointment about the way this post treats Congress. I am not aware of any enacted bill where a provision was drafted by AI without further human review. However, I am aware of a major Executive Branch tariff policy decision that was made on the strength of AI-generated charts.
This is a monarchist publication, and I have no problem with you forcefully raising the many criticisms of Democracy that have merit. You can even use scare quotes around "liberty" and "separation of powers"! But inventing hypothetical scenarios about evil legislatures while ignoring the *same* actually-existing problem in the executive is highly misleading. If the Truth is on your side, why do that?