Will expert systems disrupt the legal value chain?
In the first installment of this book review series on AI (Post 232), I argued that AI will not reduce employment in the legal sector, and in fact, the extensive deployment of AI tools might well increase total legal employment significantly. In the second installment (Post 237), I reviewed a children’s board book, considered weaponized ostriches, and concluded that AI tools are powerful complements to human lawyers but will not soon replace many – or perhaps any – of them.
In both pieces, the point is that AI – while very cool and very powerful – is also just a labor-saving device like anything else. AI should extend the reach of legal services to a broader audience, and there is little to suggest that AI will reduce employment in the legal sector overall.
This is not to say that AI will leave the legal market’s very settled pecking orders undisturbed. AI turns a service into a product, and that can have powerfully disruptive effects in an industry. In particular, legal expert systems are an aspect of AI that go beyond machine learning; instead of learning from a large data set, a legal expert system captures and replicates the expertise of a human legal expert. This turns a service into a product, and if that product gets wide distribution, it will have a big effect on the industry.
Expert systems turn a service into a product
Chapter 11 of AI for Lawyers, written by Michael Mills, the co-founder and Chief Strategy Officer of Neota Logic, outlines what expert systems are and how they work. They differ from most AI tools today in that the rules they apply are not inferred from large datasets, but instead are crafted by humans with subject-matter expertise. Expert systems rely not on the unknowable statistical magic that characterizes most AI, but on the knowable inputs of experienced humans. The result is a product that cheaply and quickly replicates aspects of the full service those experts could provide if they had infinite energy and availability. In short, a little piece of their brain becomes software, and the software does have infinite energy and availability.
Mills describes expert systems this way:
Expert systems in law deliver self-help – efficient, scalable answers to relatively routine legal questions at high volume and low cost. Can I do this? What are the risks if I do that? What are the procedures I need to follow in order to do X, Y, or Z? In a cost-conscious world, being able to answer those relatively routine questions in an automated way is a compelling story to government agencies, corporate legal departments, and law firms, as well as other legal service providers. AI for Lawyers at148.
There are important questions about how legal expert systems work and don’t work, and the many ways they might be applied now and in the future. These are very exciting possibilities, and I have high hopes that expert systems will create affordable legal services across every significant area of law for the mass market of businesses and individuals. Mills does a nice job of reviewing these possibilities.
Assuming Mills is right – that legal expert systems work, that they will get better all the time, and they will deliver huge value to elite clients and the mass market alike – I want to consider the likely impact of advanced expert systems on the legal industry.
And I want to start with my new Volvo.
The value and limitations of the AI we already use
In the Fall of 2015, my Volvo XC90 somewhat assertively steered me back toward the center of my lane when I touched a dotted white line. I was driving it home from the dealership – its maiden voyage. Looking at the GPS screen to my right and the heads-up display noting the lane correction in real-time, I saw the future of driving.
Volvo had used a series of rules to replicate and productize the expertise of a somewhat competent human driver. As a result, my job was to oversee that moderately competent driver’s work and make adjustments where needed. That is a much easier task than fully driving the car, and it frees the human driver’s brain up to think about bigger problems that might be emerging on the road. It also means you get a lot less tired on long drives, and if there’s one thing we know about human brains it’s that their performance seriously degrades when they get tired. This at least is the main thing I remember from Daniel Kahneman’s Thinking, Fast and Slow, though I may just have been tired when I read the other parts.
So far, my Volvo has never gotten tired – but it does occasionally tell me that I appear to need a coffee break.
Volvo could have built a nearly driverless car using the technologies they already had; they just didn’t. Connecting lane correction, adaptive cruise control (holding a constant distance to the car in front, including automatic braking and acceleration), and GPS navigation would result in the wireframe of a real driver, if not a fully competent one. Tesla has been more aggressive about stitching these things together, of course, with impressive if occasionally controversial results. See, e.g., Bryan Pietsch, “2 Killed in Driverless Tesla Car Crash, Officials Say,” NY Times, Apr 18, 2021.
In any event, the path is clear: in the future, drivers do less and machines do more. One of my “new Volvo” insights was that I no longer cared whether a car would ever drive me entirely; I just never wanted to have to entirely drive myself again.
As we marvel at achievements in self-driving technology, we should keep in mind that 108 years after the introduction of the Ford Model T, even a Tesla doesn’t have the steering intelligence of what the Model T replaced: a horse. Horses do not make catastrophic mistakes – they will not mistake a truck for a clear horizon (a Tesla limitation circa 2019) and plow into it at full speed, or simply run right into an obvious barrier (another Tesla limitation circa 2021). This is because horses have common sense, at least as far as steering is concerned. To re-state a point from the second article in this series (238): the very best AI still does not have the common sense of a toddler, a dog – or a horse. (To be clear, Teslas appear to be vastly safer than horses, and much safer than the typical car – especially when using Autopilot. Their lack of common sense is more than made up for by many sophisticated features.)
Though much dumber than a horse, my Volvo takes a lot of work off my hands in a way I only appreciated when I experienced it. Yet with this experience came a premonition of Volvo’s demise, because I know how global supply chains work. My newly-beloved Volvo was assembled in Gothenburg, but its parts came from a cloud of suppliers the world over. Legendary Swedish safety used to be – well, Swedish. Now the most important safety features had clearly dropped down into the supplier base – Volvo didn’t own the really important safety bits anymore. Neither did Mercedes or BMW or Lexus. The important bits were being sold to all those brands equally by suppliers none of us have heard of.
In 2021 I drove another new Volvo home from the dealership and it steered me right back into the lane like the first one did. It has GPS navigation, adaptive cruise control, and talks to me when I get too close to another car. It’s awesome.
Except my new Volvo is a Mazda. It’s got all the same stuff as my Volvo, produced by the same cloud of suppliers. It works just as well, because once you’ve productized expertise and bottled it into a combination of hardware and software, it replicates precisely every time. My Swedish safety was now Japanese, assembled in Hiroshima instead of Gothenburg.
Auto manufacturers have conceded much of the detailed engineering aspects of their product to suppliers and turned themselves into assembly and marketing shops. The technical competencies required to build the cool high tech stuff that makes our cars work these days often does not reside with the brand you see on the back of the car.
It’s not just the auto industry. As a lifelong cyclist I am sometimes asked, “Is Bianchi a good bike?” My answer is that neither Bianchi nor any other bike brand makes bicycles. They assemble and market them. And they’re assembling them from components made by the same cloud of suppliers, so the real question is whether those components are any good. (For the record, the answer is “yes” if the bike costs $400 or more, and bikes are all about the same at any given price point. So go buy a bike. It will have two wheels and make you happy, and it won’t matter a bit who assembled it.)
The question for Volvo and Bianchi – and Skadden Arps, or I’d be writing for the wrong publication right now – is whether to source or produce the elements of the final product, and if sourcing them from suppliers will ultimately destroy the firm.
Manufacturers have dealt with this question for a few decades, and in most cases, they have regarded the vertical dis-integration of their businesses as more asset than liability because of the flexibility and innovation it yields. This leaves the major brands to compete on different factors, which still have a lot to do with final quality but arguably turn more than they used to on brand positioning and the psychology of the consumer. At this point Volvo is basically a Swedish lifestyle brand that mostly makes cars. For the record, we will probably buy another Volvo in the future. They are spectacular if not altogether reliable vehicles. But we are also pretty enamored with the Mazda, and particularly its price tag.
The question of whether a law firm can control its most valuable inputs is new, since specialized expertise is the thing law firms have that everyone else needs, and specialization is what commands high fees in the profession. (My views on the value of specialization are laid out fully in Post 138.) The natural reaction to the emergence of legal expert systems would seem to be for law firms themselves to encode their expertise into expert systems and therefore take control of the process. There are examples of this, like Simmons + Simmons’ Rocketeer as discussed in Post 232. But I have doubts that law firms will ever get energetically into the software business, as it is a very different business than practicing law with a different risk profile and much more patient capital.
More likely, it seems to me, is that suppliers like Neota Logic dominate the space, capturing the expertise of top lawyers and making their products available to anyone who wants to buy.
The significance of expert systems is not just that they extend expertise to a broader audience. It is also that they can dramatically improve quality as implemented in the kind of legal work we see today, even with elite clients willing to pay top dollar for humans to do the work. Imagine how this works in the context of a big acquisition. A second-year associate reviews a contract in due diligence. He has seen 100 contracts like it before, and really worked with five of them. He sees something odd in the IP transfer provision of the contract at hand, so he escalates to a sixth year associate. She has seen 300 contracts like it and worked closely with ten or twenty. She agrees that the IP provision looks strange, so she escalates to the partner managing the transaction, who sends it to be seen by an IP specialist. This lawyer has seen innumerable IP transfer provisions. Now we have ascended the mountain.
At the feet of Pythia, we learn that the IP transfer provision is indeed unusual, but it’s actually not a problem. We then learn, however, that an adjacent provision – one the second year associate thought (literally) nothing of – is in fact a big problem.
The moral of the story, of course, is that the whole process is irremediably broken – because what else did the second-year associate miss? We can’t encode the knowledge of every specialist into his mind, but we might be able to encode at least some legal specialist knowledge into a piece of software. At that point the second-year associate is driving my new Mazda – overseeing the software, looking for marginal cases, flagging up issues for further review as before. He’s just achieving much greater accuracy – as Elon Musk assures us Tesla drivers are doing when they enable Autopilot.
This brings us to the question of how the elite law firms will respond to legal expert systems. Will law firms allow their expertise to be productized, potentially conceding one of their primary strategic assets?
The human expert will defect
I don’t believe law firms will willingly concede their treasures of specialist experience. But to end the analysis here is to indulge the Type 1 error of legal industry analysis: assuming that firms make decisions as firms. They do not. They decide as confederations of individuals, and sometimes not even that.
In this case, the lawyer-expert cannot be constrained by the law firm and has every incentive to act in her own interests. And she has considerable interests: lawyers have no way to monetize their expertise but to continue offering a service, day after day. The legal expert is Taylor Swift before the invention of audio recording: able to make a living only for as long as she is willing and able to perform. (That is, she’s Jenny Lind.) Eventually, she gets tired. And probably bored. A business owner doesn’t have to keep working after she gets tired and bored; she can sell her business. With the advent of legal expert systems, perhaps now the legal specialist can sell her business, too.
So if legal expert systems become a mature offering of suppliers like Neota Logic, and they become “standard of care” for the execution of complex legal work, what will happen to elite law firms? It is obviously too early to say anything with certainty, but Michael Porter’s venerable Five Forces provides a serviceable framework for the analysis. In Competitive Strategy (1980), Porter laid out five forces for analyzing industries. Using those five, here are some thoughts on the impact of legal expert systems.
1. Rivalry among existing firms.
Right now, elite firms’ differentiators are primarily their areas of renowned expertise. This is a function of the industries they grew up working for and prominent practice groups they have developed. The most extreme effect of highly developed legal expert systems could run along the lines of my Volvo story. Law firms’ differentiating factors could move away from core service features, since suppliers are making sophisticated diagnostic and specialist technologies available across the entire marketplace, including to entirely dissimilar law firms.
If deep and narrow legal specialties became less determinative, what could the new differentiating factors be? I might joke that Skadden Arps will become a Swedish lifestyle brand, except it’s not really a joke. (Well – Skadden is actually a Norwegian surname, so perhaps it’s a Norwegian lifestyle brand.) Brand matters enormously in high-end purchase decisions, and brand matters all the more when the service or product is complex and quality is difficult or impossible to assess. Elite law firms are very much already luxury brands, and there is every reason to think they will continue to burnish the brand and develop more elite lifestyle dimensions to differentiate from less expensive competitors.
Whatever its basis may be, differentiation will become more important, especially at the top of the market. Past experience with similar legal work will remain a critical element of an elite law firm’s value proposition, but that alone will not be enough to lock up work entirely. In some lucrative and highly technical legal fields, there are now only three or four firms that can even attempt to do the work; in a future of fully-realized legal expert systems, there might be a dozen. Rivalry among firms will, obviously, increase.
2. Threat of new entrants.
The primary barrier to entry into a legal specialty, be it transactional, litigation or counseling, is that a lawyer needs to have done the work before in order to sell those services to a sophisticated client. Because expert systems make sophisticated diagnostics broadly available, they may significantly lower the barrier to entry into specialized legal fields. This may allow lawyers with strong industry experience and directly adjacent skills to enter into a hitherto-opaque specialty with the help of legal expert software. New entrants, of course, put pressure on established specialists and require them more clearly to differentiate their offerings.
3. Bargaining power of buyers.
If the effects described above result in a broader provision of services within each legal specialty, obviously buyers will have more options and therefore more bargaining power. When buyers have more bargaining power, the obvious microeconomic effect is to reduce prices and seller profitability. But another effect is to induce more innovation from sellers, so that we could see very healthy effects upon the behavior of law firms in the method and substance of their services.
4. Bargaining power of suppliers.
The primary suppliers to law firms right now are associates. The best ones ones regarded as best, with or without supporting evidence, see Henderson, “Successful Lawyer Skills and Behaviors,” in Essential Qualities of the Professional Lawyer at 53 (Paul A. Haskins, ed., 2013), have surprisingly durable bargaining power. How might the rise of legal expert systems affect this?
The most elite firms can be relied upon to hire for prestige, so as long as Yale continues to graduate only 200 students per year, the New Haven flock will be able to service their student loans. But perhaps we will see a more explicit adoption of a two-tier system in which prestige associates are on the tenure track and forklift associates become permanent adjuncts. If the necessary work is to operate various diagnostics that have their own embedded expertise, firms may find it expedient to hire a smaller number of prestige associates and supplement their effort heavily with more modestly paid lawyers working under different circumstances and with entirely different career prospects. There is reason to believe these lawyers will operate the legal expert systems as well or better than any Rhodes Scholar could. They’re also a lot less likely to quit.
5. Threat of substitutes.
American law firms haven’t had to think much about the threat of substitutes, and surely the bar will do its best to keep it that way. But the bar’s best will likely not be good enough, since a great deal of work still done by law firms is not purely legal in nature. If law firms were paid only for those activities narrowly qualifying as the “practice of law,” the law firm business would be considerably smaller than it is today.
Legal expert systems – like so many other innovative legal offerings – suggest the possibility of unbundling the legal value chain, in this case substituting for a core law firm service. Sophisticated clients are the most likely to embrace this, using legal expert systems the same way I use TurboTax and buying law firm services à la carte.
For innovators, legal expert systems hold out the promise of productizing significant elements of the law firm service and reaching a much larger client base at a far more attractive price. Because I continue to believe that there is an enormous amount of unmet need in the legal space, I do not expect this to be a devastating development for major law firms. Their work may become even more specialized than it is now, pushing into areas too obscure to warrant building a legal expert system at all. Their work will also fill the gaps between existing diagnostic tools, and connect the dots to effectuate transactions, business decisions, and litigation moves.
Taken together, my brief observations above lead to the following conjecture: Skadden ends up in the Volvo business, which is complex component assembly supported by elite brand marketing, sales, and service. The differentiators may be harder to sustain than when they were the only car with a seat belt, but there’s still a good deal of engineering involved and it’s a business with pretty high margins if they can stay at the top of the market.