AI Law And AI Ethics Are Giving Thumbs Up To AI That Can Sniff Out Unsavory Legal Smells Embodied In Our Laws

That law stinks!

I’m sure that you’ve said something along those lines at one point or another in your life.

Perhaps you felt that a particular law was written so ambiguously that just about anyone and everyone could be construed as being in violation of the law. Or maybe you have your own special interpretation of the law, which it turns out seems to differ from that of the courts and judges and unfortunately puts you on the wrong side of the law.

Sometimes people suggest that a law smells bad. Almost like the old line about being fishy smelling, akin to rotten fish. The gist is that we use the metaphor or analogize laws to our keen sense of smell. A law might be said to have a foul odor. This is not because it somehow emits an actual scent, but instead because we find it easiest to express our disfavor or shall we say distaste for a law by labeling it as running afoul of smelling right or sweet.

Why am I talking about smelly laws?

I will be in today’s discussion taking you through an intriguing and relatively new way to examine and compose laws, partially by employing some clever techniques and also leveraging the latest in Artificial Intelligence (AI) for the law. For my ongoing and extensive coverage of AI & Law and AI Ethics, see the link here and the link here, just to name a few. For background about the rising capabilities of AI-based legal reasoning, see my book at the link here.

Okay, so we might all generally concur that at times there are laws that stink.

For those of you that are software developers, you probably are familiar with the same or certainly a similar refrain in the realm of software coding or software engineering.

Get ready for it.

That code stinks!

This is a familiar outcry by one programmer that is inspecting the programming code of another coder. Upon examining the source code, there might be oddball aspects afoot. The code could contain cryptically written lines of code. There might be state variables that do not seem to make much sense. Magical constants appear in the code as though they have meaning to the original programmer but are decidedly meaningless to someone else examining the code.

Believe it or not, a programmer might make the very same critiques of their own code.

Yes, a programmer might revisit code that they have previously written and realize that it stinks to high heaven. It could be that the programmer was under the gun to get the code out the door and so they rushed to devise the code. Another possibility is that they previously were less savvy about how to write code. Their prior coding is now a begrudging badge of shame rather than a shining emblem of honor.

Switching hats to the same considerations when it comes to the law, lawyers that devised a law might, later on, realize that the law did not come out as they had intended. Perhaps they were under a tight deadline to meet the needs of lawmakers. Words were smashed together and lots of borrowing from other laws made for a bit of a legal Frankenstein per se. The lawyers involved might also have been less seasoned at the time about how to best write laws. A slew of sometimes credible and other times incredible reasons can exist for why our laws are at times poorly written.

Indeed, let’s all agree that in fact, laws are able to be poorly written.

You can certainly see evidence of this in downstream legal activities. After a law gets passed and lands on the official books, there are bound to be all manner of legal wrangling’s associated with the law. Those that are alleged to have transgressed the law are likely to find attorneys that will seek to identify omissions, misstatements, and other maladies in the law. By doing so, they have a chance of compellingly arguing that their client is not subject to the law.

A notable difficulty with the law is that it is written in conventional human language, also known as natural language. You see, there isn’t a precise mathematical notation used for our prevailing laws. You can interpret the laws in a variety of ways because they rely upon the vagaries of words. Words, words, and more words. In the field of computational law, we refer to this as the laws being foundationally based on semantic ambiguity.

You might be familiar for example with the famous legal case of what the word “is” means. Lots of legal haggling has taken place regarding the seemingly simplest of words. For example, you might be aware that the U.S. Supreme Court case of Van Buren v. United States (2021) hinged on the meaning of the word “so” as contained with the sentence fragment “entitled so to obtain” (the debate centered on whether the use of “so” is either superfluous and unworthy of attention, or it entails great importance as to portending that which is specifically and explicitly communicated). In other instances, the placement of a comma or a semicolon has utterly transformed the possibilities of what a law signifies. Etc.

The point is that trying to write a substantively tight law that leaves little or no room for a legal argument about what it means is a pretty high hurdle. The very tool by which the law is composed has inherent semantic vagaries, namely conventional human languages such as English. On top of that difficulty there is the concern too that if you try to write a law to cover every conceivable interpretation, the law itself might become unwieldy.

A law that is shorter in size will usually be easier to implement and be one that the public at large will have an easier time understanding. If we are to have a semblance of the rule of law, by and large people have to be able to grasp what the law consists of. The longer that law gets stretched out, the likelier it is that the law will seem overly complex, excessively complicated, and contended as being unfair due to the confounding lengthiness of the law.

Seems like you can’t win either way, such that a law cannot be overly short and nor can it be excessively long. It has to somehow be the so-called “right size” akin to the Goldilocks and the Three Bears tale (porridge that isn’t too hot and nor too cold, but just right in temperature).

But the length alone of a law doesn’t tell the whole story. You can write a short-sized law that at least attempts stridently to be carefully composed and will somewhat withstand legal contestations. You can write a long-sized law that also strives to be strongly composed and able to stave off legal argumentation. Seeking to devise a law that has zero wiggle room is generally an impossibility. That doesn’t though mean that we should shrug our shoulders and just write our laws in any wild ways.

You could persuasively suggest that there are suitable ways to compose laws and there are less suitable ways to compose laws.

And, referring once again to programming and coding, there are absolutely lots and lots of studies and standards associated with composing computer programs that are better written than others. The field of software engineering provides a tremendous amount of attention to how to write “good code” as in code that is well written. Metrics such as understandability, maintainability, usability, effectiveness, efficiency, and other factors are all brought to bear in trying to gauge whether code is suitably devised or shall we say averts being stinky.

Well, if we can review and assess programming code to discern whether it stinks or smells, you might be wondering whether we could try doing the same thing to our laws.

Yes, we can.

A relatively newer area of the law consists of identifying what is referred to as legal smells.

I’m sure that you can guess what this means. The notion is that you might inspect a law to figure out whether it has anything fishy or stinky in it. Does the examined law unduly contain gotchas and omissions? Does the law go over the top in terms of being extraordinarily vague? Is the law riddled with semantically ambiguous wording that will open a legal storm?

Again, the chances are that by even detecting these legal smells we won’t necessarily be able to wring them entirely out of the law. As mentioned, words are words. Nonetheless, we can seek to make laws that are “rightly worded” in the Goldilocks sense. Trying to close off as many loopholes and legal problems as you can, at the get-go, would be a societally beneficial activity. The goal isn’t to get to zero legal difficulties, and instead to do as much as is feasible to get a law into a less than stinky and more toward smelling sweet level of legal interpretation.

Notice that the smell of a law is not especially concentrated on what the law covers. Whether a law is aimed at dealing with criminal activity or civil activity, the aim of utilizing legal smells is not whether the law itself is just or publicly warranted. The key to using legal smells entails seeking to make sure that the law as written serves the purpose for which it is intended and does not unduly contain legal gotchas that could have been figured out beforehand when the law was being composed.

Of course, if a law has high stakes, we are almost certainly going to be more attentive to whether the law is understandable and able to be legally comprehended. For example, a law that deals with say murder would seem to be worthy of more cautious wording than would a law that entails the day-to-day dealings of parking violations of a car. The repercussions of the law might guide us toward greater and closer scrutiny.

You might immediately react by exhorting that all laws ought to be properly composed. Sure, some laws are more grave than others, but this shouldn’t especially dictate whether laws should be well written or not. Do not let a lousy law on the books simply because it is a law covering perceived lesser issues.

Part of the law-devising calculation is that we usually make use of expensive human labor to write our laws, as in costly lawyers at steep billable hourly rates. This is not to suggest that we should undercut our laws by seeking to shave on those costs. Once again, if you try to cut corners when writing a law, the odds are that the costs will arise anyway downstream. All that you are doing is shifting the costs and to some degree burdening those subject to the law excessively so.

What I am trying to drive toward is that we could use automation of a sort to aid in crafting laws. The crafting of a law is nearly always manually done today. To clarify, yes, of course, laws are written in word processing packages. Yes, laws are often cut and pasted from other online electronic sources.

I am referring to using AI to aid in writing laws.

A handy first place to start consists of devising AI that can seek out legal smells. A lawyer drafting a law can feed the draft into the AI. The AI would then produce an indication of where there appear to be stinky or afoul lines of the law. The human lawyer could then take a look and see if the legal language could be adjusted appropriately.

Rinse, repeat.

During the process of crafting a law, AI could be used repeatedly throughout the numerous changes and versions that are concocted by the human hand. This might be done periodically. Perhaps if a law is being drafted over a period of several months, you might run the AI on a weekly basis to see how things are coming along.

A more real-time approach is also possible. While in the act of composing a law, the AI could be directly and instantly assessing the legal language. You can think of this as how today’s word-processing packages work. I’m sure you know that you can start writing a line in a word processing app and it will try to predict what you are likely to compose next, plus it will catch misspells and misplaced punctuation. Word processing is examining your writing as you are in the midst of composing a story or memo.

The same could be undertaken when composing a law.

There are LegalTech packages today that try to aid in composing laws, though the inclusion of a legal smells AI-based detection is rarer right now. This will gradually grow and it will be a promising market for crafting laws at the international, national, state, and local levels.

Mark my words.

In any case, AI for detecting legal smells is the first step of maturation in this realm.

The notion for now is that the AI acts to alert someone that a law has some form of legal wording of a stinky nature. After that, it is up to a human lawyer to decide how to deal with the legal wording. They might adjust it, or they might leave it as is. There can be many contentious reasons underlying whether the seemingly legal wording is truly suitable or not.

More advanced AI is being devised that would not only detect legal smells but in addition proffer suggestions of how to correct the legal smells. This could be done hand-in-hand with a lawyer, thus the AI is working in a semi-autonomous fashion. The future might be that the AI would make the corrections autonomously and there would not necessarily be a human lawyer in the loop, though this is a hotly debated topic that you might want to read about at the link here.

I’ve somewhat emphasized so far that this AI for legal smells detection and potential correction would be used when crafting laws. We might leverage software development parlance and suggest that this is similar to dealing with code at what is known as compile time. Once code has been compiled, it is then available to be put into production or what might be referred to as run-time.

Consider how AI that is devised to deal with legal smells might be useful even after a law has already been put onto the books. Assume that the law was handcrafted and there wasn’t any AI at the time or that wasn’t used to assess the law before it was enacted.

I had mentioned earlier that if a law has legal interpretation issues this can open the door for a lawyer to mindfully and legally argue that the law doesn’t apply to their client or maybe applies in ways that are say less harsh than otherwise might seem to be the case.

How does the lawyer reach that kind of legal argument?

They inspect the law.

As such, an AI tool for dealing with legal smells can readily be useful even after a law has been formally anointed as a law. A lawyer that is seeking to find ways to potentially undercut or attack the law could use AI to find the gotchas and potential stench for them. The attorney might do the same absent the use of an AI tool, and that’s fine. On the other hand, it could be that the attorney might not notice facets that the AI catches. Having a double-checker to examine a law could be quite advantageous.

This can readily be used in law firms and an entire law practice might tap into an AI-powered legal smells detection system. Seasoned partners might use the AI on a more cursory basis. Associates that are still coming up the legal learning curve might lean more so into AI. For now, the AI would be considered a tool and not a replacement for the legal mindedness of human attorneys (there is all manner of qualms about AI in the law that goes beyond that of being used by a human lawyer, including skidding into the murky waters of UPL or Unauthorized Practice of Law, see my coverage at the link here).

In my AI & Law research lab, we have been developing AI that entails legal smell capabilities.

I’ll share some tidbits with you about legal smells that you might find of notable interest. Also, in a moment, I’d like to mention a foundational study originally done on legal smells that is deserving of a cornerstone credit for sparking the expanding interest in legal smells such as at my AI lab.

Lawyers can likely find useful the whole conception of legal smells in that it is a topic that can bring to light cognitive lawyering elements that are useful to employ. Ergo, even if you aren’t using AI or devising AI, the focus on legal smells can be an insightful mental framework for thinking about the law and for aiding how you practice law.

First, consider this about legal smells:

• Obvious legal smells
• Non-obvious legal smells

Looking for legal smells in law can be done in a rather perfunctory way, such as detecting punctuation that seems out of place or the use of legal catchphrases that are known already as being problematic. Those are what might be labeled as obvious legal smells. The really tricky part and the tougher challenge for AI is to go after the non-obvious legal smells. That’s where the remarkable excitement exists for those attempting to apply AI in this way.

Next, consider this:

• Distinctive legal smells
• Convoluted legal smells

Here’s what we mean by saying that legal smells can be distinctive versus convoluted. A distinctive legal smell essentially stands on its own. You can point to a particular word or sentence fragment and reasonably declare that something is amiss. The more convoluted legal smells might be hidden from view, such as a piece at the first part of the law that when seen as being connected to a later part of the law gives rise to legal concerns.

Into this comes a number of related considerations. Is a legal smell localized, which we refer to as a whiff, or is it pervasive throughout the law, which we call a stench? Is the legal smell a commonly known one or is it a newly discovered form of legal smell? And so on.

I had earlier indicated that there is legal smell detection, plus there is the possibility of legal smell correction.

For AI that also aims to do legal smell correction, there are some additional aspects.

Legal smells can be rated as:

• Acceptable legal smell
• Intolerable legal smell

An acceptable legal smell would be one that if left in the law would probably not give rise to significant legal concerns. In contrast, an intolerable legal smell would be gauged as likely making for lots of legal troubles downstream. There is a bit of controversy about how to make such ratings. If done by a human lawyer, are they setting themselves up for potential legal exposure if they have indicated that a legal smell is acceptable and yet later on it is shown to be intolerable? The same issue arises associated with AI making those kinds of ratings.

On a related basis:

• Easy-fix legal smell
• Complicated-fix legal smell
• Unfixable legal smell

In this case, the question arises as to whether a legal smell is fixable. The simplest and presumably straightforward instances are easy-fix legal smells. The more complicated ones are harder to resolve. The third category for presumed unfixable legal smells is a raucously controversial aspect. You might argue that there is never a circumstance whereby a legal smell cannot be fixed. If you had to rewrite the entire law, presumably you could likely rectify the legal smell. But, recall that we are also faced with the semantic ambiguities of conventional human languages, such that it could be extremely hard to fix an especially rotten legal smell (whether it is truly unfixable is a deeper and extensive legal philosophical conundrum).

Turning to the study that has aided in kicking off the legal smells procession, a research article in the esteemed journal Artificial Intelligence and Law by authors Corinna Coupette, Dirk Hartung, Janis Beckedorf, Maximilian Böther, and Daniel Martin Katz (published June 2022), defines legal smells or aka law smells in this fashion:

• “Definition 1. (Law Smell) A law smell is a surface indication that usually corresponds to a deeper problem in a legal system.”

The characteristics of a law smell are said to be depicted via a law smell profile. A profile indicates what the law smell is, identifies why such a law smell is problematic and encompasses other facets such as how to detect the law smell and how to potentially mitigate it.

Here are some of the law smells they list:

• Duplicated phrase: “Duplicated phrase is the legal equivalent of the software engineering code smell duplicated code. Lawyers smell it when they get the feeling that a text is verbose and repetitive. More formally, a duplicated phrase is a phrase above a specified length that has more than a specified number of occurrences in a legal text.”
• Long element: “Long element is a legal adaptation of the software engineering code smell long function. Lawyers smell it when they get lost in the text they are reading or have trouble discerning those parts of the text that are relevant to them.”
• Ambiguous syntax: Ambiguous syntax is a legal adaptation of the software engineering code smell mysterious name, with pinches of repeated switches and speculative generality. Lawyers smell it, inter alia, when they litigate over the meaning of commas or argue about whether an or is inclusive or exclusive.”
• Large reference tree: “Large reference tree is a legal adaptation of the software engineering code smell message chain. Lawyers smell it when they find themselves following many references to understand the full meaning of the text they were originally interested in.”
• Natural language obsession: Natural language obsession is a legal adaptation of the software engineering code smell primitive obsession. Lawyers smell it, inter alia, when they are interested in the law related to a specific named entity (e.g., a committee of the United States Senate), or when they try to automate their clients’ compliance. More formally, natural language obsession is the representation of typed data as natural language text, often without a standardized format.”
• Other

They emphasize that in addition to law smell detection there is the ambitious goal of being able to do law smell deodorization. The aim is to be able to perform legal refactoring, which borrows terminology familiar to those in software engineering of being able to take programming code and refactor it into something presumably more robust code. The same can be said for the wordsmithing or right-wording of the law.

Software developers typically make use of relatively sophisticated software-integrated development environments (IDEs or SDEs) that allow them to store their programming code, retrieve it, reuse it, and serve as an essential tool when devising and maintaining source code. As per the research paper, the researchers call for more effort toward legal integrated development environments that could be coupled with legal version control capabilities.

You might vaguely know that there are ongoing expressed qualms that the legal field seems to often be the last to adopt high-tech. Not only is this a frequent case in law offices, but the same can also be said for lawmakers and regulators that are charged with composing and modifying our laws. The tech used is often particularly low-tech and not yet at the high-tech level.

In my view, a bit of high-tech could go a long way in not just doing perfunctory chores related to the law, it would also undoubtedly and indubitably help society to know what laws there are, give ready access to our laws, and generally inspire the public to perhaps have greater faith in the laws and bolster wider confidence in the rule of law.

Conclusion

Now that you’ve gotten a aromatic bouquet of legal smells (pun!), let’s revisit a subtle point that I mentioned toward the beginning of this discourse.

You might have briefly noticed that I decried a manifest difficulty associated with our laws is that they are written in conventional human natural languages such as English. I suppose that seems obvious at first glance. Yes, you say, we use natural language to express our laws. What else can we do?

Suppose that we could compose our laws in some form of precision that would ultimately expunge any possibility of legal interpretive consternation. Think of this as being on par with the logical-mathematical-oriented proofs you learned in high school or college. Perhaps we can construct a legal-specific language that would enable not only for us to express our laws in a more definitive manner, but could also potentially use computing to try and aid the computational proving of our laws and the widespread enactment of our laws (see my discussions at the link here and the link here, for example).

Welcome to the Holy Grail as it were for those immersed in AI & Law and the advent of computational law.

All manner of attempts have been and continue to be made to recraft laws on a computability basis. You might have heard that some describe law-as-code, meaning that we ought to be able to take our existing laws and transform them into the equivalent of computer coding. Others clamor that we are already heading toward code-as-law. In short, as we make use of AI and ADM (algorithmic decision-making) throughout society, we are essentially embedding our laws into programming code whether we realize we are doing so or not.

I assure you that the AI & Law arena is going to be a kicker of a challenge and offers exciting puzzles yet to be solved. Lawyers might find this an exciting field to join. Computer scientists likewise. Those interested in the law and those that are fledgling computer scientists are encouraged too.

Returning to the legal smells topic, Shakespeare famously said that a rose by any other name would still smell as sweet. Well-written laws that provide clarity for society are sweetly aiding the rule of law. I’d dare say they are like roses. Poorly written laws that are confounding and create massive downstream disturbances are, frankly, abjectly stinky. They certainly aren’t roses.

Maybe we can use AI in the law to strive toward rose-smelling laws and avoid getting ourselves mired in skunk-scented laws.

Per Shakespeare, may boldness be our friend.