"A computer can read and understand contracts, the lawyers will be able to do strategic work"

This is how Ilan Edmon, co-founder of LawGeex explains, which developed an AI product that knows how to read and understand contracts, and made a competition between it and human lawyers. "In contracts, every word has a meaning and its place in the sentence"

As we recently published in Dr.Roey Tsezana's article, A new study by the Israeli LawGeex examined who is more successful in law: human lawyers, or artificial intelligence?

"The artificial intelligence developed by Israel's LawGeex beat the human experts to the punch, in the particular field on which it was trained and tested. The 20 professional lawyers had to go over five nondisclosure agreements (agreements to maintain confidentiality), and identify problematic clauses that could get the seal into trouble later on. In front of them was placed the artificial intelligence software that passed exactly the same contracts. The artificial intelligence finished with an accuracy score of 94%, and the human lawyers with a score of only 85%. And if that's not enough, while the lawyers needed 92 minutes on average to go through all the contracts, the artificial intelligence needed only 26 seconds."

Ilan Edmon, co-founder of LawGeex and who came from the computer side, says in an interview with the science site and the people and computers site that the idea came from the co-founder of the company, Nouri Bekur, a lawyer by education and profession who worked for a long time at Mitar, one of the largest law firms in Tel Aviv.

"After years, he realized that most of the lawyer's work is quite boring and that it could be done better with a robot. And that's where the idea comes from. He contacted me, I specialize in AI. We built an initial prototype of a system that reads contracts."

What is special about contracts that makes them a comfortable platform for operating artificial intelligence?
"Unlike English, on which natural language processing mechanisms can be applied, the legal language is very different - legal English. It expresses very precise ideas in a smaller variety of words, and every smallest change - even changing the position of a word in a sentence or a comma, can change the meaning of the sentence."

"This is in contrast to other language AI projects where the accuracy does not need to be so high. There is a lot of data to run on and the level of guessing has to be a little more than flipping a coin to know what the trend is. In the case of contracts there is very little data to learn from and the level of accuracy should be close to 100% otherwise you will miss something important. It's a very interesting challenge."

"We have reached a situation where we are able to identify possible risks in a certain specific type of contract with much higher accuracy than very experienced lawyers are able to do."

how do you do it
"We use the algorithms we developed, most of them of the type of neural networks that try to imitate how we learn. They learn very specific things very, very well. The great advantage of computers over people is that a neural network can read all existing contracts - hundreds of thousands of contracts in a few hours and see how to read a contract, while this person takes several years of studies and specialization."

"The biggest advantage is that once the network is trained, it doesn't get tired. She is also consistent because she doesn't need to take a nap, so it is possible for the users of the system to get a uniform result over time without needing so much auxiliary power (from the interns - AB)."

"The interns will do more contentious things and artificial intelligence will replace them mainly with contracts that take up most of the time, but the risk in them is relatively low - NDA agreements, service agreements, licensing agreements, etc., when 80% of the lawyers' time is wasted on these contracts. Every lawyer, even the youngest, can deal with important things instead of going over every word in every contract."

How was the training?
"When we want to teach an artificial intelligence system to draft a contract, in the first step it needs to understand what is written there. Therefore, in the first stage, unsupervised learning - we let the computer read hundreds of thousands of contracts and learn the language."

"The second thing is to find specific points that could be a risk in the contract, because the fact that he has done contracts does not mean that he knows what is a risk and what is not. Therefore we made the following training semi-guided. Human lawyers marked points of interest in the contract. Then the system knows that this is what it is looking for. She knows how to find among everything she has read things that are similar to the things that have been specified and this is the great advantage of neural networks of the preliminary phase of learning the language. Now if the system sees a case similar to what it learned but worded in completely different words, it will still know how to find it."

what is the next step?
"The next step in the development plans is to further deepen the understanding capabilities of the system. Even now we go deep and make sure that there will not be a word in the contract that we do not know what it is talking about and why it is there. But the depth is required so that we can attack the complicated and large contracts. Today the process is that one party drafts a contract and passes it on to the other party who makes amendments to it. Our goal is to reach a type of automatic negotiation - when each company has its own policy and the lawyer's computer has to offer the ideal middle ground".

Is academia involved?
"The technology was developed for the most part outside academia, but we have a close advisor Prof. Yonatan Uman from Bar Ilan University."

Will the lawyers eventually be replaced?
"I do not think so. I think the equivalent of this is the invention of Excel for accountants. Accountants were very frightened when they first encountered the spreadsheet, but it did not take away their work, it became their work tool."

"Artificial intelligence will not replace human intelligence in the foreseeable future, but it can certainly serve as a work tool that will allow lawyers to make decisions more easily, go through more documents in less time, and reach a higher level of work - to catch things that would otherwise fall through the cracks. But to replace them in terms of the strategic and human work they create? It's not something I think can happen in the future. Certainly not in the near future."