Intro:
Welcome to the Contract Lens Podcast, brought to you by Malbek. In this podcast, we have conversations with contract management thought leaders and practitioners about everything contracts and its ecosystem. Today's episode focuses on simplifying contract review through technology. We are joined by Kingsley Martin, president and CEO at KMStandards and chief contract scientists at Akorda. Kingsley has been at the forefront of technology innovation in legal practice. He has 30 years of experience in the practice of law, software design and development, and is a fellow at the World Commerce and Contracting Association. So now it's time to relax, grab a glass of wine and let's talk contracts.
Matt:
Hello, Kingsley. Good afternoon today.
Kingsley:
Hi, Matt. Very nice to meet you.
Matt:
Nice meeting you as well. I look forward to recording this podcast with you. Today, we are going to talk about simplifying contract review through automation. How does AI and ML, which is artificial intelligence and machine learning technology, how does that help legal professionals when it comes to contract reviews, contract negotiations, and other contracts and use cases? So the first topic, Kingsley, is AI and ML have become very popular acronyms and the buzzwords lately. It's everywhere. In fact, if you haven't heard of AI or ML, then you haven't been listening, right? In your opinion, can legal professionals really trust it? And can you rely on this technology for contract reviews, or is it still early in its maturity cycle?
Kingsley:
That's a great question, and I think that the answer, it depends on what you're looking for the AI, ML to do. If you're looking for it for answers, then we may consider it fairly immature along it's progression. But if on the other hand, you're looking for insights, then I think it is well positioned today. The best example I can give or analogy is that think of it a bit like an x-ray or an MRI machine. The surgeon is not going to look to the MRI machine to do the surgery. It's not a threat directly to their job, but it will help them essentially perform the surgery better by giving them insights.
Kingsley:
But I think there's another piece of that analogy that also applies. It doesn't just mean it works out of the box. The surgeons need to be trained on what the x-ray and MRI machines are telling them. I think that's part of where we are today is that while we have some very interesting examples of AI and ML, we don't necessarily have trained lawyers that are capable of essentially using it effectively, and that may be part of the next step. Would you agree?
Matt:
Indeed. I think the piece where you said they have to learn it as well is key. I think the expectations, whenever I talk to organizations is, "Hey, AI and ML can it do all of this for me?" Well, sure it can. How accurately? Well, it depends on how well it's trained and how much you have learned how it's working, meaning if it's your own documents, it's one thing. If it's third-party paper, then there's all kinds of patterns in different industry when it comes to legal language.
Kingsley:
If you think of it to just give you the answer, is very different from thinking of it, "Well, can it help me?" Think of it more like an augmentation. Can it help me do my job more efficiently? So our goal is not to essentially make the longest job 100% more efficient, but if you think of it essentially making it significantly more efficient and higher quality, there's a different equation as to how you approach it.
Matt:
Absolutely. In a way as AI evolves, it's learning. It's learning from your usage patterns, and if you have the right technology, it is supposed to improve with time organically or by you training it for your specific use cases. Having the right technology to be able to train it and having it learn from your contracting process is key.
Kingsley:
Yeah, I think there's one other piece to the puzzle too. I mean, I may be in the minority on this one, that I believe if we combine this advanced technology where the essentially efforts to simplify and streamline and standardize contracts, then the machine will essentially be much more effective and more accurate in reading. I mean, I come across from time to time, contracts where a sentence may be actually over 1,200 words long. The machine will not do a very good job of analyzing that, but then on the other hand, nor the human, because it is quite simply open to too many interpretations. So I believe essentially simplification and maybe even going so far as a control lexicon to build contracts that can be accurately read by both humans and machines. So I think it is a good combination of the advanced technology and quite frankly, simplification.
Matt:
I think there is a movement. I look at NDAs, right? A simple example of what should be a very standard document, there are a billion variations of an NDA. So standardization absolutely helps AI technologies perform better because when you have lengthy and complex content and variations, then the training is never ending, right? You have to keep training it for the different patterns that it has to learn.
Kingsley:
Well, if you don't mind just digging a little bit deeper into that, I mean, NDA is a good example. You're right. We are seeing a proliferation of essentially standards. We're now going to have the battle of the standards. But if you actually look at them and essentially use the AI to identify the concepts that are in each of those documents, there's a huge degree of commonality. The words that we might use to express each of those concepts are going to differ from one person or even one standard to the other. But that's where, again, if you... Essentially, if we have good to AI and machine learning combined with essentially a controlled lexicon, then it really doesn't matter how you express it. I mean, we know that that's a confidentiality obligation. We know that that is an obligation to return. Does it have other conditions and qualifications associated with it? Yeah, we can pick those up too. So I think at a concept level, there's already a lot of consistency in there, but it's very hard to essentially see those concepts in today's very complex language.
Matt:
Well, how do you recommend organizations banks their people with this technology? Because in my opinion, AI is not there to replace human oversight and reviews, but is, like your doctor and the MRI example is spot on, it will expedite and help them do their job quicker, better, faster.
Kingsley:
That's the right way to approach it today. It's like almost like a triage, as if there were a few medical analogies today. That triage might be... The first pass might be a machine. If that contract is written in a manner that the machine has high confidence that it can accurately assess it, then maybe, it does a lot of the work. The next pass might be by a para professional, a legal assistant who may be able... With the machine, we have shown many, many times that they can perform at a higher level of professionalism using the machine, that maybe they'll be able to tackle more of the contract.
Kingsley:
At the end of the day, maybe they're just a handful, two or three questions, that would have to go to the professional, to the highly trained, highly skilled lawyer. So we are able then to, again, not fully automate, but tackle the questions that are capable today with our current technology or being addressed, and leave those handful of really deliciously hard problems for the professionals, which in some ways it's more fun for them anyhow. It's efficient and uses that time much more effectively than wading through pages and pages of texts that the machine may be able to assure them, "No, this is pretty standard. There's no risk involved here."
Matt:
Couldn't agree with you more. I think it's much better than a legal ops or an attorney reading a document from scratch to begin with, which can take many minutes or hours. You'd triage it like you said, that the technology read through it, identify the key terms and the risky provisions, let the paralegal then take a shot at it, and then leave the more complex and the riskier language for the true attorneys that need to then take care of how they want to accept or reject those provisions.
Kingsley:
Absolutely. I think we're beginning to see, and it's still, unfortunately, just a handful of companies building workflow around that process, and I think it's important for the technology providers to essentially think of it in that manner too and to help them with the workflow and triage.
Matt:
Absolutely. From what I have seen, AI, ML in the legal tech space really took off in the last three to four, five years, and it's come a long way. What about the future? Where do you see this going? Is it more of the same, or do you see it solving anything beyond contract reviews?
Kingsley:
I think there is a significant barrier that we need to essentially achieve next, which is... I'll give you my example. A few years ago, I actually pitched an idea to DARPA and NIST. DARPA, the Defense Advanced Research Projects Association, the NIST, National Institute of Science and Technology. I suggested to them a robotics challenge based upon the autonomous car challenge, but in this case, the three rounds would be draft a clause, draft an entire contract, and negotiate a contract. At the time, I knew that I think we could do very well in a blind test in the first two rounds, that essentially asked both lawyers and invite the technologists to draft the clause or a contract, and then the judges would determine by a test, who was it written by? Now, the reason why I felt that we could pass that test is because the machine would not have to draft anything new.
Kingsley:
We have access to a fast repository of previous existing, successfully executed and negotiated contracts and clauses, and we could with the AI pick the best of them. But I think what we would have been challenged to do would be essentially, create new text. A bit like the use of GPT-3 and fake news today, and we tried this by the way, could we write the first few sentences of a contract, an NDA and see if the machine could write the rest using transfer learning? Well, I actually thought the results were pretty remarkable. It certainly looked like an NDA, but if you read it, it was of course, rubbish. It just wouldn't pass, but still as a first pass, it was pretty remarkable. So what I think it goes next is that we will for the next few years, and for probably many years to come, have a hard time, essentially having the machine draft something. But what we can do, which is very relevant for contract review is I've got my starting point.
Kingsley:
I've got my playbook and all of my fallbacks, and I've got all of the hundreds of contracts that I've previously negotiated in the past. I send it over to you for review and you red line it. That contract comes back and we can then essentially look at that red line and compare it both to our fallbacks and all previously executed contracts, and be able to suggest the closest revision of that. Here's the next. I think that next critical piece, can we surgically edit, eject those fallback provisions into that red line contracts? Because it is not acceptable to the parties to simply take your clause that you've edited and replace it with mine. I've got to inject my requirements into the language of your red line.
Matt:
Correct.
Kingsley:
We cannot do that. I think that's the next piece. So the next stage would be automated first line, red line review, provided the machine was confident enough and the risk was low enough that we would feel comfortable doing it.
Matt:
That is a big time saver in the sense that if you've got your, for example, liability or termination provision negotiated a hundred times to a certain language, then why reinvent the wheel every time to negotiate? You have your fallback, you know what you've negotiated before. Let technology help you.
Kingsley:
Absolutely. I mean, for example, even in the most complicated clauses, the incoming red line basically inserts a curve out for gross negligence from a limitation of liability. We have agreed to that. We know from our database, we've agreed to that 25% of the time, and certainly, for clients or opportunities over 5 million, we're going to agree to it.
Matt:
Correct.
Kingsley:
We accept it, and then essentially then inserted into our contract. So I think that where it goes next. This barrier, this next hurdle for us is creating new. We have now the ability to data mine the past, but that's essentially for us actually, truly helping the lawyers and the contract managers. We need to be able to do this intermediate level of drafting inside of our accepted fallbacks into the counter parties' red line.
Matt:
So then with all of that evolution of this technology and power, the last question I'll ask you is legal professionals, are they adopting and open to this technology and change where they're usually skeptical in terms of, "Will it really work for me? It's too good to be true." I think it's something that this industry is going to have to adopt very quickly because it is the future.
Kingsley:
Yeah. I mean, the truth of matter is, and I'm sure that you see this on a daily basis. My profession is built around essentially conservative practice, and the many of my clients did look to me to essentially do a thorough risk assessment and help them navigate through very complicated transactions. But I think what you see here is a divide. Although the divide is this, the minority are still the people and the adoption of new technology, I think there's a tremendous amount of skepticism by those who feel threatened, and plenty of optimism by those who see opportunity. From time to time, I get a chance to present to law students, which is an interesting gauge of where we are and where we're going. Even in law schools, we're seeing just very significant skepticism, almost a front of the, "Can a machine really do what I've had three years to learn about and it can just do it for me?"
Kingsley:
But as we've talked about, it is not essentially doing the job for you. It's giving you insights to do your job better. But the example that I give them to essentially make them think about it is just saying, for example, you as a lawyer go to a high-tech company that does, let's say 10 or a hundred thousand cloud services agreements per year, and you offer to help them negotiate one of those cloud services agreements with a big potential new client. How much could you charge? It's an MSA basically. I don't know. 10,000 or so? Maybe more? If on the other hand, you went to them and said, "I have some technology and capability to analyze all of your past cloud services agreements, and I believe that I can show you how to essentially save. I can accelerate the transaction by two weeks over your existing six weeks, and I can reduce your ongoing administration costs of this contract, to the tune of many million dollars a year of accelerated income and reduce costs."
Kingsley:
How much could you charge that for? So that's really one of the things. As I shifted from traditional practice to technology, I realized that I was moving from the one to the many. If you're doing the one, and you're like the bespoke tailor, the technology may not help you and you'll feel threatened. But if you can see the opportunity to go from the bespoke tailor to say, Zara, then the opportunity is enormous. So it depends on essentially your perspective of what part of this business you apply the technology to. There is a huge amount of opportunity in here.
Matt:
Absolutely. This is amazing insight, Kingsley. I think you are correct. We often see the praise on LinkedIn posts and all that AI is not there to replace legal professionals, but legal professionals that don't use AI will be replaced. I think this is an area that everyone will have to adopt and embrace because this is the future.
Kingsley:
I think that actually, I mean, I remember essentially the medical professions, as these very sophisticated machines came in, they had that same degree of pushback. You can't imagine a surgeon today not taking advantage of them, but there was that initial period. I remember the same even at Westlaw. Moving from books to online, there was a period of two to three years in that case that I preferred to do it the old way until the convenience of the accuracy and the speed just made it absolutely patently clear that we were offering a better way. But there are bumps along the road, and that's part of the challenge. As we also discussed earlier, there's some time investment that the people using this technology will have to put into it, and the busy professionals. So it is understandable to a degree, and I think it's also on us and to make it as easy to adopt as possible.
Matt:
Yeah.
Kingsley:
That's the other thing quite frankly, I'm seeing in the last two to three years. With cloud based systems, I mean, it's now very easy to use and adopt this technology. Its just almost behind the scenes.
Matt:
It's a great insight, Kingsley. I really appreciate your time today on this topic, and I look forward to doing more with you. Thank you so much.
Kingsley:
Well, thank you, Matt. Pleasure. Thank you so much for inviting me to participate in this podcast.
Matt:
Our pleasure. Thank you.
Kingsley:
Thank you.
