In utramque partem

There's definitely a grain of truth in this. Our legal system is built such that only certain people, registered and appointed for specific roles and given specific powers and duties, can perform specific legal functions. Until recently, only an advocate could represent a client in court. Only a conveyancer can register property transactions in the deeds office. Only a notary public can certify a marriage certificate. Only a registered patent attorney can file a complete South African patent application. And on and on. 

To someone looking in from the outside, this seems like the sort of mess that begs to be torn down and rebuilt around more responsive, streamlined systems. Which may very well be the case in an abstract sense, but is unlikely to happen without a thorough-going review and reorganisation of the legal system. This, in turn, would require an equally sweeping push by the legislature to pass various pieces of legislation to overhaul the entire system. Such a raft of legislation would undoubtedly be hotly contested at every stage by all the affected parties who, remember, are all legal practitioners of one sort of another. So organic, voluntary change will probably be slow. In abscence of such a comprehensive, intentional change, there's a good argument to be made that the various roles, duties, powers etc will stay much as they are, and technology will have to adapt to this fact. For example, attempts to get AI representation in court have so far far foundered on the rocks of procedural issues, technical issues, rules around who may represent clients in disputes, and the fact that the tools touted so far have been pretty bad at their job so far. 

However, given the continuous froth of new companies offering to provide AI-driven services of all sorts, I expect this approach to be tried repeatedly using various stratagems of more or less dubious legality. In the interim, self-representation by litigants relying on AI has become a live topic in legal circles overseas, and is likely to become more and more pertinent in South Africa as well.

II - opaque sources of law

This brings up another issue that may restrain the effects of automation on law: the overlapping, intricate and somewhat opaque nature of law itself. Taking South African law as an example, it is the product of various sources of law (legislation, common law, civil law, international law, customary law and Constitutional law) that all interact in various ways under different scenarios and legal facts. This complexity is enough so that practicing lawyers today only really have a full grasp of their specific area of law. More importantly, a lot of the complexity is either obscure or poorly-documented enough that today's Large Language Model-based AI systems are unlikely to have much data to draw from. As a result, at least for the moment, we can expect LLM-driven forms of legal automation to be better at more broad and general legal questions ("What are the legal requirements for me to become the executor of a will?"), and prone to error when presented with specific questions which hinge on this more opaque information ("How many chickens can I keep on my specific property, and what is required for me to keep them there?")***. Unfortunately, the lay public may not be able to spot where one question fades into the other, and so may have an unrealistic understanding of the quality of the legal advice provided by all-AI legal services.

III - the robust nature of the legal system

The fact that our current legal system is however-old it is does not automatically mean that the structure is rotten and liable to come down at the slightest kick. Instead, it could be argued to mean that it is the way it is because that's just an incredibly robust way of formally resolving disagreements and regulating behaviour in a largely-urban, literate society. Some of our legal principles date back to Justinian (something I've been delighted by in the past), not out of an excess of conservatism, but because these principles broadly work and will continue to do so as long as we have neighbours, fences, dogs and the concept of ownership of property. The machine of law is a grinding, creaking thing. But it is also a system that has survived everything that ingenious people can throw at it in an effort to permanently game or destroy it****. Betting against a system that has survived every new technology from the printing press to the internet might not be as rational as it sounds when first considered.

IV - the centaur fights back

Cory Doctorow, perhaps most popular today for coining the term 'enshittification', has recently come up with the concept of a 'centaur' and 'reverse centaur'. He provides the following snappy definitions: 

"In automation theory, a “centaur” is a person who is assisted by a machine. You’re a human head being carried around on a tireless robot body. Driving a car makes you a centaur, and so does using autocomplete. 

And obviously, a reverse centaur is machine head on a human body, a person who is serving as a squishy meat appendage for an uncaring machine."

Crucially, he notes that, while people often enjoy being a centaur, we hate being turned into reverse centaurs:

"Like an Amazon delivery driver, who sits in a cabin surrounded by AI cameras, that monitor the driver’s eyes and take points off if the driver looks in a proscribed direction, and monitors the driver’s mouth because singing isn’t allowed on the job, and rats the driver out to the boss if they don’t make quota.

The driver is in that van because the van can’t drive itself and can’t get a parcel from the curb to your porch. The driver is a peripheral for a van, and the van drives the driver, at superhuman speed, demanding superhuman endurance. But the driver is human, so the van doesn’t just use the driver. The van uses the driver up.

Obviously, it’s nice to be a centaur, and it’s horrible to be a reverse centaur. There are lots of AI tools that are potentially very centaur-like, but my thesis is that these tools are created and funded for the express purpose of creating reverse-centaurs, which is something none of us want to be."

This gels with my understanding of how automation works in general - the system achieves improved productivity by redesigning around the parts that can't adapt (the machines) at the expense of the parts that can (the people). So the factory assembly line is a pretty miserably place to work in if you are a factory worker, but churns out standardised widgets faster than the same number of workers on the line ever could. 

But I should also add that the pace and trajectory of automation is, rather unfairly, dependent on who the subjects are. The luddites got more 12 000 soldiers thrown at them, along with show trials and the government making machine-breaking a capital offence. Cynically, I suspect that CEOs (an eminently automatable job) will not be treated in the same way when they decline to vacate their positions in favour of robo-executives. The social and economic status of lawyers is variable, but overall my feeling is that the profession has enough power to broadly dictate the course of events in line with their own interests. This probably means an embrace of centaur-ism, and substantial changes to how the profession works. But not an end to the importance of lawyers within the system.

V -in summo

Having presented multiple sides of this discussion, I think the future is still very much wide open. Personally, I'm currently leaning more towards 'lots of big changes over the next few years', although I think that we will adapt the existing structure of law rather than tearing it down wholesale. I'm both dreading and looking forward to how this all plays out, and in the interim would value any further input that my colleagues and people in the know on automation might have.

* And law in general, plus a lot more besides. But I'm really only qualified to talk about IP law.

** We're only using this one at a surface level and carefully ignoring the dripping ooze that sticks to anything birthed from places like 4chan. Speaking of which, does it worry anyone else that all of the LLMs include 4chan and Reddit in their training data? And that this has already had weird unintended results? What more subtle horrors are buried in there?

*** It doesn't help that LLM-based AIs seem incurably prone to 'hallucinations' - from my own experience, it looks like nearly all the advice it provides seems to contain at least one error or unsource-able claim. Getting this down to, say, a 1% error rate would not necessarily save you in a legal situation where small procedural or factual mistakes can doom an otherwise-sound legal argument. 

**** This is more than nearly any multiplayer game can boast, by the way. It turns out that trying to balance a system against a swarm of actors, each self-interestedly motivated, aware of all of the rules of play, and capable of learning from & coordinating with each other is just a really, really hard problem.