Template:M intro technology Better call ChatGPT: Difference between revisions

From The Jolly Contrarian
Jump to navigation Jump to search
Line 72: Line 72:
Before long, this process will have itself sedimented into the administrative sludge that weighs your organisation down. Other processes will depend on it. Surgical removal will be ''hard''.
Before long, this process will have itself sedimented into the administrative sludge that weighs your organisation down. Other processes will depend on it. Surgical removal will be ''hard''.
====LLMs and waste====
====LLMs and waste====
{{drop|L|LMs can’t function}} by, or think for, themselves (''yet''). Their deployment implies not saved legal cost, but “[[Seven wastes of negotiation|waste]]” transferred: what once was spent on [[legal eagle]]s will be diffused among [[software-as-a-service]] providers, the firm’s procurement complex, [[internal audit]], [[operations]] and, yes, dear old [[legal]] who will ''still'' have to handle exceptions, manage and troubleshoot the system, vouch for it, be blamed for it, to periodically certify its legal adequacy to CASS compliance and then, when it turns out not to be, explain why it wasn’t to the operational risk [[steerco]].
{{drop|L|LMs can’t function}} by, or think for, themselves (''yet''). Their deployment implies not saved legal cost, but “[[Seven wastes of negotiation|waste]]” transferred: what once was spent fruitlessly on [[legal eagle]]s will instead be diffused among a phalanx of [[software-as-a-service]] providers, procurement personnel, [[internal audit]] boffins, [[operations]] folk and, yes, the dear old [[legal|legal eagles]] who will ''still'' have to handle exceptions, manage and troubleshoot the system, vouch for it, be blamed for it, periodically certify that it is legally adequate to the [[Chief operating officer|COO]] and then, when it turns out not to be, explain why it wasn’t to the operational risk [[steerco]]. All of this costs money, takes time and distracts the firm’s resources from better things they could be doing. Just because it is harder to evaluate, doesn’t mean it isn’t ''there''.<ref>This is wishful thinking, of course: in a world where accounting projections are the first and last word, that ''is'' all that matters.</ref>


LLMs as finite. They necessarily mimic what has gone before. While, yes [[Alpha Go|AlphaGo]] might engineer a novel strategy in a [[zero-sum game]], it is not so easy on in the non-linear infinitude of life. An LLM that purports to improve on its training material will be distrusted it doesn’t understand, so what good had it got of reimagining?
==== The finite game ====
{{Drop|B|y design, LLMs}} learn and reason exclusively from what has gone before. While, yes, [[Alpha Go|AlphaGo]] might have engineered a novel strategy in a [[zero-sum game]], the non-linear infinitude of life that a contract review process is a different kettle of fish. And this is not, in any case, what one wants in a sorcerer’s apprentice.


The perfect LLM serves up an archetypal sample of what you already have.
That being the case the perfect LLM would be one that served up an archetypal sample of ''what you already have''.

Revision as of 17:53, 8 February 2024

“The era of LLM dominance in legal contract review is upon us”.

Oh, just listen to the tiny, AI violins. Some legal technologists[1] have presented a “groundbreaking comparison between large language models and “traditional legal contract reviewers” — being junior lawyers and legal process outsourcers — benchmarking then against a “ground truth” set by senior lawyers.

It did not go well for the meatware.

The researchers collated and anonymised ten “real-world” procurement contracts — NDAs were deemed a bit easy — and fed them to a selection of junior bugs, LPOs and large language models.[2]

The buried lead: variance increases with experience

An interesting finding, noted but not explored by the paper, was a variance measurement[3] across the categories of human reviewers: the least qualified, the LPOs had an “alpha” variance of 1.0, implying complete agreement among them about the issues (a function, we suppose, of slavish and obedient adherence that is beaten into LPO businesses). This dropped to 0.77 for junior lawyers and further still to 0.71 for senior lawyers.

You read that right: experienced lawyers were least likely to agree what was important in a basic contract.

This says one of two things: either lawyers get worse at reading contracts as they get more experienced — by no means out of the question, and would explain a few things — or there is something not measured in these key performance indicators that sets the veterans apart. That, maybe, linear contract analytics is the proverbial a machine for judging poetry, and isn’t all there is to it.

Hold that thought.

Results: all hail the paralegals?

In any case, for accuracy the LPO paralegals did best, both in spotting issues and in locating them in the contract. (How you can spot an issue but not know where it is we are not told). Junior lawyers ranked about the same as the chatbots. Perhaps to spare their blushes the report does not say how the vets got on.

But it shouldn’t surprise anyone that all the machines were quicker that the humans of whom. LPOs were by far the slowest. There is a cost to obliging humans to behave like robots.

Clear implication: as we can expect LLMs to get better over time,[4] the meatware’s days are numbered.

Now, if you ask an experienced lawyer to craft a set of abstract guidelines that she must hand off to low-cost, rule-following units, but for whose operation she remains accountable, expect her to draw her boundaries conservatively.

There being no “bright lines” wherever there is scope for nuance or call for subtlety, she will stay well inside it, not trusting a dolt — whether naturally or generatively intelligent — to get it right.

This is common sense and little more than prudent triage: well before any practical danger, her amanuenses must report to matron for further instruction. She can then send the machine back into the fray with contextualised instructions, or just handle anything properly tricky herself. This is all good best-practice outsourcing, straight from the McKinsey playbook.

Now, a standard-form contract without at least one howling error is unknown to legal science, so she should expect an assiduous machine reader, so instructed, to be tireless, and quickly tiresome, in rooting out formal discrepancies and bringing them to her attention.

Contrary to modernist wisdom — viz., thou shalt not rest until all problems are solved — descending the fractal tunnel of error is, largely, a bad idea. These are the things an experienced lawyer rolls her eyes at, takes a moment sanctimoniously tut about, but then lets go. Usually, she will not register that the “issue” was even there. This is noise: instinctive, formalistic fluff, well past a seasoned professional’s ditch tolerance.

Q: What’s the difference between an LLM and a trainee?

A: You only have to punch information into an LLM once.[5]

This, perhaps, accounts for that mysterious variance among experienced lawyers. Contract review, end of the day, is an art, not a science. Much of a contract is filler better satsficing than optimising much less perfecting. Sometimes you take a point, sometimes you don’t. Some like the comfort of redundant boilerplate, others cannot abide it. Harbouring different traumas and scars from their life experience, different individuals — and institutions — are fearful about different things.

Does it matter that your contract has a counterparts clause? Does it matter that it doesn’t?

A busy-body LLM that catches every blemish and cannot take a view as often creates a problem as a solution. This kind of literalness rubs off, or is beaten out of, junior lawyers as they develop. But mechanical ducks like LLMs have an insatiable thirst for it.

For what we are fighting here is not bad lawyering, nor bad machines nor bad intentions but bad process design. Supporting it with machinery will make things worse. This is the lesson of the sorcerer’s apprentice.

The oblique purposes of formal contracts

There is one peculiarity that a literal approach to contract review cannot address, but we should mention: sometimes a contract’s true significance runs tangentially to its content. The forensic detail is not always the point.[6]

Sometimes the act of finely thrashing the details out actively frustrates the real purpose of the contract, which is to fulfil a social function. As a commitment signal or competence signal.

A basic example: European financial services regulations require institutions to have written contracts with all customers, as a regulatory end in itself. The rules are less prescriptive about what the contracts should say.

A firm must, therefore, have a written contract to do business. To meet that end, any delay in finalising that contract, in the name of “getting it right”, ought to be a source of regret. (You might be surprised how often firms are obliged to negotiate their terms of business, in that it is not “never”.)

In sorts of contract act as a sort of mating ritual: a performative ululation of customary cultural verities signalling that yes, we care about the same things you do, are of the right stuff, the same mind and our “ad idems” are capable of consensus. Again, it matters less what the contract says than that it is there.

If it is that — most NDAs are that — then descending into an LLM-level of subterranean pedantry and exactitude, in the service of “picking up things that even a gun paralegal might not”, is a rum plan. The point is to carry out the ritual, afford it the minimum required pleasantries, but not to labour them.

Volume contracts

Those exceptions aside, where high-volume, low-risk legal processes do not function as courting rituals, the name of the game is not perfect negotiation, but no negotiation.

Negotiation is the problem.

If you find customers regularly negotiate your terms of business, or you get regular snarl-ups on procurement you have bad forms.

Fix them.

This might mean persuading legal to come to Jesus on the width of its idealised liability exclusion, or it just rewriting the form in a nicer font and plainer language — but either way, the answer is not to leave the problem where it is and mechanise it.

Doing that will leave you two enduring problems: first, your portfolio of standard contracts will not be standard; secondly, your bad form is now beset with administrative machinery it will be hard, later, to take away. By appointing unskilled technocrats to manage a broken process — and, likely, other unskilled technocrats to oversee and monitor them — you have institutionalised a bad process.

John Gall’s Systemantics captures this well. Temporary fixes have a habit of becoming permanent. Bureaucrats are butterfly collectors: they do not give up their responsibilities without a fight. Their managers rarely have the stomach for one: it does a job: leave it be.

Before long, this process will have itself sedimented into the administrative sludge that weighs your organisation down. Other processes will depend on it. Surgical removal will be hard.

LLMs and waste

LLMs can’t function by, or think for, themselves (yet). Their deployment implies not saved legal cost, but “waste” transferred: what once was spent fruitlessly on legal eagles will instead be diffused among a phalanx of software-as-a-service providers, procurement personnel, internal audit boffins, operations folk and, yes, the dear old legal eagles who will still have to handle exceptions, manage and troubleshoot the system, vouch for it, be blamed for it, periodically certify that it is legally adequate to the COO and then, when it turns out not to be, explain why it wasn’t to the operational risk steerco. All of this costs money, takes time and distracts the firm’s resources from better things they could be doing. Just because it is harder to evaluate, doesn’t mean it isn’t there.[7]

The finite game

By design, LLMs learn and reason exclusively from what has gone before. While, yes, AlphaGo might have engineered a novel strategy in a zero-sum game, the non-linear infinitude of life that a contract review process is a different kettle of fish. And this is not, in any case, what one wants in a sorcerer’s apprentice.

That being the case the perfect LLM would be one that served up an archetypal sample of what you already have.

  1. At the Onit Inc. “AI Center of Excellence” in Auckland, New Zealand.
  2. It looks to have been those of OpenAI, Google, Anthropic, Amazon and Meta. Poor old Bing didn’t get a look in.
  3. “Cronbach’s alpha” is a statistic that measures internal consistency and reliability, of a different items such as, in this case, the legal agreement reviews. A high “alpha” indicates consistency and general agreement; a low alpha indicates variance or disagreement.
  4. Maybe not, actually, but okay.
  5. This is a nerd’s version of the drummer joke: What’s the difference between a drummer and a drum machine? You only have to punch information into a drum machine once.
  6. This is, broadly, true of all contracts, from execution until formal enforcement — and the overwhelming majority of contracts are never formally enforced.
  7. This is wishful thinking, of course: in a world where accounting projections are the first and last word, that is all that matters.