Correctness of The Armstrong argument

Despite these challenges, legal AI startups clearly see an opening for themselves. To understand why, it might be worth turning to Kira again, in what can be called The Armstrong Argument.

Noah Waisberg of Kira Systems argues that though IBM may have 100x the R&D funding of all the legal startups combined, one could not assume it would succeed in legal contract review.

He sets it up by enumerating the legal problems that technology could solve, posed as questions. For example: Which of these 1 million documents are relevant to determining if anti-competitive behavior occurred in this specific case? Which of these contracts have a change of control or exclusivity clauses? Is it illegal for individuals to have ferrets in the state of California?

Categories of legal companies

Out of seven types of questions, corresponding to categories of legal AI companies, Waisberg writes that Watson was the go-to technology in only the “legal research-y” one. To assume IBM would succeed in other legal domains, he writes, would be akin to arguing that because Lance Armstrong had come off winning seven consecutive Tour De Frances, he would win the 2006 New York Marathon. (Armstrong finished 856th according to wiki).

“Machine learning is a new and glorified name for statistics,” says Harsh Gupta, who has worked with the Centre for Internet and Society and looked at topics like machine bias. The picture Waisberg presents includes areas in such a state of flux that even knowledge of what the optimal statistical techniques are isn’t clear.

Yet, everyone is jumping in. At an earnings call in April 2017, Vishal Sikka, the former CEO of Infosys, was asked by an analyst from Cantor Fitzgerald about its AI platform Mana, (now Nia). “It is one of a kind. No, everyone and their uncle has an AI platform these days,” a report quotes Sikka as saying, “From the times when I studied AI as a student, everyone calls their toolkits AI platforms.”

Sikka then goes on to describe how Mana was used to devise a solution pertaining to Non-Disclosure Agreements for a bank in Asia, reducing the need for 10-15 lawyers, and references Infosys’ buyout of Skytree.

When going through the well-known players in AI—IBM, Deep Mind, Nia—it becomes noticeable that the aim was towards developing generalized intelligence. First, solve for intelligence and then use it to solve everything else.

Customer’s confidentiality

But the indication today is that machine learning solutions work best when integrated into company systems—ideally when the data is within the company’s control. As products, there is an effort to market that may not be worthwhile for the large players that chase contracts in other areas, and the startups already have moved fast in areas such as law.

Client confidentiality is another issue. Kira Systems has its solution “air-gapped” from the clients’ data, having already refined it. Platforms have become commoditized, and others—like Microsoft and Amazon—have all come in.

While talking of legal AI, Surukam’s Sundareshan believes platforms are going to go the ERP way, accounting for 5-10% of contracts but not necessarily yielding a great return on investment. “The value is not going to be in the platform,” he says, “but in the design and implementation of the solution.”

Large law firms in India are aware of the potential of the technology as well as the pitfalls of not embracing it. For instance, at Nishith Desai, technology head Milind PM says he’s evaluated technologies overseas. For the system to digest what you’re doing and act at an intelligent level, he says, it takes eight to nine months. For now, the firm is building its own internal due diligence system.

The change people like Milind already see is in the eroding hierarchy of firms. “It has become a network organization,” says Milind. “Now you can flow around in the workflow.”

Law firms prefer in-house development, concurs Tony Joseph, the CEO of Cognitive Computer Services (CCS) based in Trivandrum.

CCS had been working for several years with an educational company called Sherston based in the UK. They were later approached by Kennedys, a £150-million law firm specializing in insurance, and have been involved in prototype development for them.