“Thinking of AI as only GenAI leads to the adoption of solutions that don’t work when there are practical, non-GenAI ways to solve real problems.”
Over the past several parts of this series, we have discussed the problems and risks confronting AI, its use by legal, and how those problems may lead to the eruption of the GenAI volcano. The truth is GenAI has been overhyped and oversold. As a result, there is a real risk of overreliance on GenAI by those who don’t understand it and what it does which could lead to disaster.
There’s yet another danger contributing to the potential eruption we haven’t addressed, one that is more fundamental than all the others: a definitional confusion that’s helping drive the overreliance we’ve been worried about.
The Definitional Danger
The legal community has gone from carefully distinguishing GenAI as a category of AI to using the term “AI” as a reference to GenAI itself. As in only GenAI is AI and the anything else isn’t. In fact, AI is a much broader concept and refers to a whole category of tools with different uses, benefits and value apart from GenAI completely.
This confusion, made worse by vendor marketing, fuels overreliance on GenAI tools on the one hand, and under reliance on solid, accurate, and performing non-GenAI tools on the other.
In fact, real AI experts understand conceptually what AI is, what it can do, and the differences and drawbacks of confusing GenAI with AI generally.
AI Expert Insights
One such expert is Baron Reichart Von Wolfshield who goes by the single name Ki. Ki has worked extensively on AI from the late 70s. By the 90s he was building and designing AI programs for the US military, Disney, the architectural community and, yes, for law firms. He routinely consults with some of the world’s largest companies and law firms on AI and AI development. In addition, Ki has a unique way of designing AI programs to solve human problems that involves observation, logic, and simplicity, not smoke and mirrors.
Like all true experts, he has a way of explaining complicated concepts simply and understandably. I know from years of experience as a trial lawyer how rare that is.
Ki’s Insights
Ki makes it simple: AI should be thought of as something that appears to act like an intelligent thing. He uses a mechanical spring to make this point, “The simplest artificial intelligence in the world is a spring. You set it up, push it down, and it’ll push back against you. That’s the core of AI: it is something you can ask to do something later, and it will. That’s AI. It acts like a human.“
Thinking of AI in this kind of broad way illustrates the point that the key is finding the right tool to solve the problem, not adopting tools just because they happen to be in vogue. It’s what he calls the procrustean effect, aka known as trying to fit a square peg in a round hole.
Ki is also quick to rightfully point out that this doesn’t mean you can use AI tools without understanding what they are doing, how they work, and without proof they will do what is claimed. That’s Ki’s beef with LLMs and GenAI: it’s that the hype doesn’t match reality, and most users don’t bother to get it.
That’s why he calls LLMs a “parlor trick”: “Everything with LLMs right at this moment is on par with and a child of autocomplete.”
He also believes the hallucination problems can’t be fixed: “The reason AI lies is the same reason a human lie, because AI is modeling the same neural system of a human. You can’t get an LLM to stop lying any more than you can stop a human from lying.” It’s just part of what LLMs are, and that’s not going to change.
Because of all this, he concludes that the current proven usefulness of LLMs is little more than that of a glorified search engine. So, thinking of AI as only GenAI leads to the adoption of solutions that don’t work when there are practical non-GenAI ways to solve real problems.
Practical Non-GenAI Examples
Ki gave a couple of examples. He actually sat with a lawyer for a day and watched what he was doing. What he found was that the lawyer spent a lot of time trying to figure out where and how to file attachments to the multitude of emails.
Sounds kind of trivial, but I know this guy’s pain. You’re trying to work quickly and make filing decisions among a multitude of files, and a mistake could be costly in terms of lost materials and information. To top it all off, you can’t enter time for looking for a file and be paid for it.
Ki figured out a simple, non-GenAI solution: create a bot that could automatically file the attachment and then tell you where it put it. Simple but saves lawyers and legal professionals a hell of a lot of time and stress. For all the hype of GenAI, it’s not a tool that can do that simple task. Says Ki, his bot “is AI but it’s not an LLM.”
Another example: Ki noticed that a lot of time was spent on calendaring significant events like hearings, depositions, court dates and the like. Having humans do that was, at best, clumsy and error prone since it required a number of steps to be taken to get the item accurately on multiple calendars, let everyone know, and then set up a process to deal with it. He ultimately set up a complete project management system that did all this and more. By recognizing patterns over multiple cases, it could even help predict what might happen and what an opponent might be doing.
The important thing is that, in both situations, he first learned what lawyers and legal professionals really need to do their everyday jobs and what they care about. Then he developed a simple, usable AI solutions.
This is not GenAI but is AI that works, doesn’t hallucinate, doesn’t make errors, and doesn’t need to be verified.
Implications for Legal
Of course, as we have discussed before, the hallucination problem has enormous implications for legal. In many areas of legal, inaccuracies and hallucinations can’t be tolerated. “That danger is missed,” says Ki, “by those who don’t understand the tool.”
“But there is second, and perhaps more serious risk here,” says Ki. And that is by lumping all AI into the GenAI bucket, more valid and error free AI and automation tools will be ignored. Tools that can make life simpler and better for lawyers. Tools that solve what Ki refers to as “boring” problems. Problems that are stress points for every attorney.
Instead of focusing on these solutions, GenAI providers often try to coat every solution in a GenAI wrapper without considering the real problem, and a simple solution that works. By doing so, they suggest to legal customers that all AI is GenAI and only GenAI can solve legal their problems. The result is that customers are often getting something that’s expensive, doesn’t solve their real problem, and doesn’t work as they thought. At the end of the day, they discard the tools altogether.
There are in fact things that non-GenAI does quite well and quite accurately if you understand what it is doing and analyze the problem correctly on the front end. Often these problems result in work for which lawyers are not trained for but have to do anyway. Ki wants to stamp all these out, leaving lawyers and legal professionals to do what they are good at.
By thinking that AI is GenAI only, the boring repetitive tasks that Ki tackles would be left undone, perpetuating inefficiencies that could be eliminated while instead adopting GenAI systems that create greater inefficiencies instead.
The Over Reliance Problem
There’s also the danger that lawyers and legal professionals will come to believe all the GenAI hype and just rely on it. It’s the “if GenAI tools says it, it must be true” syndrome. Here’s an example of how that could work. Admittedly, if you create the right prompt, a GenAI tool can give you a list of questions to ask in a deposition or even assist you in the deposition itself to spot inconsistencies or correct bad questions.
But the temptation for a busy lawyer, particularly a less experienced one, is to just take that list and doggedly ask every question on it. We have all seen lawyers who make that kind of list on their own and do just that. They end up asking questions that clearly were no longer relevant based on what the witness previously said. They miss nuance and body language that may lead to unexpected and unplanned questions that sometimes can break open a case. They fail to follow up.
I once took the deposition of a class rep. I made a list of questions in advance to ask. At one point in the deposition, I happened to ask what I thought was a throw away question: what claims the witness had made or had been made against him. There was something in the way he looked when he answered. A certain hesitancy that made me dig in on what seemed to be a meaningless line of inquiry. Come to find out, he had filed bankruptcy a few months before. That fact ended the case. Blind adherence to a GenAI deposition list of questions would never have led me to that question.
Lessons For Law Firms
All of this poses particular problems for lawyers, legal professionals, and law firms. They aren’t Ki and most don’t have a Ki working for them.
But there are some practical steps firms can take and some lessons fordealing with AI and GenAI. First and foremost, firms need to realize that there is a difference between AI and GenAI and that there are solutions to problems that don’t involve GenAI at all.
Firms should also understand that there are issues yet with GenAI that haven’t been solved. Issues with respect to things like accuracy and the costs of verification, the infrastructure, and the robustness of the investment and capital.
So before purchasing GenAI products out of FOMO or over relying on their outputs, ask the hard questions. Identify the actual pain points you want to eliminate and then determine whether the tools can really solve your problem or would simpler, non-GenAI tools do a better job.
And for God’s sake, don’t accept what vendors or others are telling you. Remember, that for a variety of reasons we have discussed, the GenAI volcano may be about to erupt as better and more accurate AI solutions surface, and the hype is replaced by reality.
Next time we will look at how a non-GenAI solution may, in fact, even solve some of GenAI’s real problems.
Read our entire “Pompeii” Series:
Like Lawyers In Pompeii: Is Legal Ignoring The Coming AI Infrastructure Crisis? (Part I)
Like Lawyers In Pompeii: Is Legal Ignoring The Coming AI Cost Crisis? (Part II)
Like Lawyers In Pompeii: Is Legal Ignoring The Coming AI Trust Crisis? (Part III)
Like Lawyers In Pompeii: Is Legal Ignoring The Coming AI Financial Crisis? (Part IV)
Stephen Embry is a lawyer, speaker, blogger, and writer. He publishes TechLaw Crossroads, a blog devoted to the examination of the tension between technology, the law, and the practice of law.
Melissa “Rogo” Rogozinski is an operations-driven executive with more than three decades of experience scaling high-growth legal-tech startups and B2B organizations. A trusted partner to CEOs and founders, Rogo aligns systems, product, marketing, sales, and client success into a unified, performance-focused engine that accelerates organizational maturity. Connect with Rogo on LinkedIn.
The post Like Lawyers In Pompeii: Is Legal Ignoring The AI Definitional Crisis? (Part V) appeared first on Above the Law.