The Singapore Law Gazette

Category Errors, Or Why AI is Eating Our Lunch

A fellow lawyer recently told me, “We tried AI for contract review, but it missed obvious issues, so AI is useless for legal work.”

If you don’t know what a category error is, it has a lot more to do with the hypothetical example above than meets the eye.

The first example of a category error was actually in the title. Of course, as a word-heavy profession, we immediately catch the idiom. But if I had meant it literally (as in an actual AI manifesting itself physically in front of me and eating my noodles) it would have been a category error.

Here are more examples from the Stanford Encyclopaedia of Philosophy (SEP):

“The number two is blue”.

“The theory of relativity is eating breakfast”.

“Green ideas sleep furiously”.

Perhaps wary of introducing more category errors by defining it as a category, the SEP article introduces the idea obliquely. These sentences are “infelicitous in a distinctive sort of way”. As lawyers, dealing with categorisation is part and parcel of the work. Here are some examples that we are familiar with, and a lot less likely to commit category errors in.

What’s the difference between unjust enrichment and restitution?

Is this case more alike than different from the other?

When thinking of dispute resolution, do parties think clearly about the categories of disputes that are carved out for arbitration, and for the Court?

What types of service unavailability would be caught by Service Level Agreements?

Our case law is replete with catchword categorisations, and long discussions about doctrinal boundaries. Without actually defining the term – because doing so would be to delve deeply into the boundaries and penumbra of what falls within the category of category errors – one instinctively knows that mistakes in law seldom rise to the level of infelicity seen in the examples above. Without me saying more, you know we take pride in accuracy. As we should.

Except, when it comes to legal tech (or law tech, if you prefer). Abrupt transition? It reflects the abrupt change in posture that I’ve observed in the sector. We can be consummate, observant, curious, professionals in our respective legal domains, and suddenly just turn Luddite. Anyway, let me explain.

Let’s say we wish to avoid making statements of such a nature that would cause someone, whose opinion we care about, say, a client, to feel that you have basically said “the number two is blue”.

Or let’s say we wish to avoid such statements, because they reflect a level of understanding that may very well soon collide hard with reality, in the same way it becomes quickly obvious to the child that green ideas don’t sleep furiously: “so will AI replace me? <laughs in legal certitude> ¹This obfuscates or defuses a concern of general application, with a personal implication about capability: especially effective if you’re a good lawyer to begin with. But also Luddite anachronism – also another form of category error, just like asking for the sundial’s shadow in an age of clocks – is looking increasingly unfashionable in this day and age.”

Do you know, just off the top of your head, statements that are typically and frequently made in the legal sector that might sound this way, perhaps to fellow lawyers, or more relevantly to other sectors? Especially those that study artificial intelligence and computer science. Or sectors that have faced a reckoning earlier than us, even as they associate us with high costs but peace of mind and risk cover.

The irony is this: the real concern here is not whether AI will replace you. It is the category errors that will catch us unawares, and eat our lunch when we do not expect it, because, well, we categorised the uses and potential of technologies wrongly. Can you imagine looking bad and only being able to say “Wow, we got it really wrong”?

Put it another way: in a game of tennis, it’s the unforced errors that can cost you the match. It is not necessarily the other player wins.

I would argue AI won’t replace lawyers. Whether in the physical sense, in the physical courtroom (DoNotPay’s strategy is an article for another day) or office, as ludicrous as that may sound. Or in the more metaphorical sense, in your role as a trusted advisor and advocate. The lay of legal practice’s land will, however, definitely change.

A quick look at legal history, both local and global, reveals an adaptable, evolving profession. If you understand and delve into AI the same way you understand case law and statute, you will have a relatively smoother, even fun time, navigating the waves. You will find the right tools for the right uses, and the right uses for the right tools. Tool-making and tool-use is usually humankind’s thing. The question is whether it is also the legal profession’s thing. That’s where the category errors are rife.

Category errors may be obvious or not. They may be subtle or crude, simple or complicated. You would probably have seen examples of all of these being committed with legal tech around you, quite possibly without knowing it.

Let’s say legal tech adoption is essentially an issue of tool adoption²This article owes a serious debt to the L4 programming language. It is a Domain Specific Language for law by Legalese. What does a DSL do, and why do lawyers and anyone who uses the law benefit from a DSL for law? Check them out here, here and here. I was inspired by what is called a type system. Which is a sort of clarity check on the things that you refer to or define in code. You’ll want to keep tabs on this project. The best tools take a long time to come to fruition. Microsoft Word was not built in a short time, even with the resources of Silicon Valley.. This of course presumes there is a culture that places emphasis on the right tools, and great tools, not just tools that we have been working with, or have been taught. I do not think that this is necessarily true. Then tool adoption encompasses, very briefly, these stages:

1. Understanding how the tool works
2. Understanding when to use the tool
3. Understanding how to use the tool (or what the steps of tool usage are)
4. Practising using the tool
5. Mastering use of the tool

It’s useful to note that snap judgments and intuitions tend to be more an inkling of a thought in the early stages, before it turns into a well-honed intuition, instinct or insight that comes with practice, and eventually mastery.

We’ve all been there as beginners. We come to something new, then do or think the equivalent of trying to contain water with a sieve. At the same time, some of us have also tasted mastery, and live it every day.

As a profession, we possess veritable weapons of persuasion. There is a danger that we convince ourselves and others, when we are still at the stage of being blissfully (or embarrassedly) unaware trying to contain water with a sieve, to take certain courses of action. It’s entirely possible that mastery in another area also tricks or lulls one into this. Those wrong courses of action are precisely the journeys one can expect to waylay you and allow a better tool-wielder to whizz past. You don’t even need to talk about a sentient tool – yet.

To be clear, and to put it another way, you can be a fan of technology and tool adoption, and still commit category errors.

You can be an ancient time-traveller who’s just discovered the wonder of modern kitchen tools, and who has already become its greatest fan, and still wreak havoc in the kitchen with these tools (especially your purchasing decisions).

Category errors in legal tech are often more subtle: consequences are often insidious or delayed. You just work the longer hours you always have. A mistake never gets discovered. You never get to enjoy an improvement, and the joy it brings. You never got to see past the lack of low-hanging fruit to the smorgasbord of riches in the crown of the tree. A new case comes in. Life is busy. There’s more hype about technology. There was a previous wave. This is just another one. Life goes on.

Keen to perhaps avoid some infelicities, silly mistakes, and future-proof yourself?

Here are some examples of infelicities in legal tech that you may wish to avoid. Some are scenarios, some are statements, some are funny, but all have a kernel of reality to them. As much as possible, I have tried to find a close analog from legal practice:

1. I just want a quick and dirty guide about how to use this groundbreaking and revolutionary technology. I need a quick and dirty guide for this first case, but I don’t expect to invest much more in skills development.
2. It’s best if it fills up only the time that I have available for it, and only that time. (Why, I’ve got things to do and money to earn!) Client matters of great significance should only take up the time that I have for it.
3. Mastery is only for technical experts. Principles of contract law are only relevant to transactional lawyers, and I am not one.
4. I also expect the technology to master my needs. Contracts should nonetheless serve me well.
5. Because this being a new facet of practice is not a thing (in an earlier era). Contracts are new and therefore not a thing.
6. If I can’t see any use-case now, there must be no relevant use-case. I don’t see any legal risks, so there must be no risks.
7. If I can see a use-case now, there must be one. All I see is risks, therefore this matter is nothing but risk.
8. I do not quite understand this technology, but I know there is no use-case. I do not understand this area of law, but you are not liable. (The opposite of 8 is actually intellectually humble.)
9. Since this implementation of AI is contract-oriented, it means AI cannot draft complex pleadings. Since this technology services contract can’t serve your provision of complex derivatives, it must not be very good.
10. Technology is no good. The law is bad.
11. Technology is amazing, no questions asked. The law in its current form is a completely unalloyed good.
12. AI is bad because lawyers keep using hallucinated answers from AI in court. Microsoft Word is bad because I never learnt proper formatting and use, and never saw it as an ethical duty or duty of competency to get good at it.
13. My area of law is special, and not susceptible to technology X. General ethics don’t really apply to my specific area of law.
14. Your area of law is technology, therefore you must be able to fix computers ³I’m only half-joking here. I will try my best to help you if you ask nicely though.. You do contracts so you can do a pre-nup.
15. ChatGPT is GenAI, and GenAI is ChatGPT. The biggest firm in the world is legal practice, and legal practice is the biggest firm in the world.

It’s high-time that we collectively approach our use of tools with the same rigour and attention to detail that we bring to legal doctrine. Generative AI usage is just one aspect of tool usage, so is Microsoft Word, and we need the whole toolbox. The alternative is being left behind by our own commission of category errors.

Here’s a suggested framework for strategising:

Suggested Framework for Tool Adoption

1. Invest time in understanding capabilities and limitations, and invest more time than you think you should;
2. Apply the same professional development standards you use for legal skills;
3. Recognise that tool mastery takes time, like any legal competency;
4. Test tools systematically with real samples, before judging. Don’t prejudge things quickly, your mileage may vary, but you may also be able to go an exceptional mile further than others with a particular tool.