Subscribe for perks & to support LI

Your Interests & Preferences: Personalise your reading

Which best describes your role and/or interests?

I work in a law firm
I work for a company / in-house
I'm a litigator at the bar
I'm a law student
Aspiring law student
Other
Save setting
Or click here to show more preferences...

I am interested in the following types of stories (uncheck to hide from frontpage)

Firms / In-House
Deals
Courts
Legal Education

Always show me: (overrides the above)

Exclusives & Editor's Picks

Website Look & Feel

Light Text on Dark Background

Save preferences


Note: Your preferences will be saved in your browser. You can always change your settings by clicking the Your Preferences button at the top of every page.

Reset preferences to defaults?

Has India Legal Inc Missed the Bus on Artificial Intelligence, or is it Just the Right Time to Get on Board?

Nearly every industry is trying to predict the impact of Artificial Intelligence, with some bracing themselves for wholesale reinvention. Legal will not be immune, reports Kian Ganz.

What Artificial Intelligence sees in the world: our front cover picture transformed by Google’s DeepMind algorithmWhat Artificial Intelligence sees in the world: our front cover picture transformed by Google’s DeepMind algorithm

Artificial intelligence (AI) and machine learning (ML) have been making inroads into nearly all walks of life. Much of what’s visible in the mainstream has been restricted to headline-dominating stunts such as Google’s DeepMind systems beating humans at ultra-complex board games like Go or Chess or the IBM Watson system besting champions of the TV game show Jeopardy (way back in 2011). There have been rapid improvements in self-driving car technology by several companies. And on the consumer software side, facial and photo recognition, real-time text, voice and image translation and other useful tools from the major tech giants often seem like magic, or at least eerily, almost-humanly intelligent.

AI can theoretically be applied to many problems that seem intractable for human beings to handle, either by virtue of complexity (such as beating the average human at Go) or because of the volumes involved (such as automatically recognising every person or object in all photos uploaded to the internet).

In the medical profession, AI has huge potential, with some ML algorithms touted to outperform trained doctors in diagnosing certain diseases or conditions purely from patient data or scans. It is unlikely that this will ever make obsolete the human touch and expertise required in medicine, but a computer may at the very least assist doctors both quantitatively and qualitatively – even catching patterns and correlations between symptoms that human doctors may miss.

Nearly every industry is trying to predict the impact of AI, with some bracing themselves for paradigm shifts that could require the reinvention of their businesses wholesale. Legal will not be immune.

In Europe and the US, in particular, there are now more of the major law firms that have dabbled in AI and ML than not, with some systems churning out parts of due-diligence reports near-automatically, replacing hundreds of associate hours.

As is so often the case, however, India’s story is its own.

AI use cases

Dibyojyoti Mainak, the GC at technology and news content start-up InShorts, notes that most of the applications and potential in AI for lawyers have so far been in several areas. “There’s due diligence, which can be easily handled by that type of tech and can be completely eliminated [one day in future]. Most of it is a very mechanical job: if you feed in a set of parameters and set of documents, any reasonably good natural language processing will be able to figure that out.”

A corollary to that are smarter document management solutions, that can automatically classify, categorise, connect to each other and help you find the documents sitting on a law firm’s or company’s servers. And in jurisdictions where discovery or pre-litigation disclosure is a big thing, so-called “predictive coding” has made inroads: human lawyers review and categorise, for example, a 1% sample of documents by hand, with the computer then learning from that to identify the remaining 99% of documents as relevant to the case or not, which can be further improved by interaction between lawyer and machine.

The other big use cases is in legal research. “There’s a lot of opportunity for sure,” predicts one India-based CEO of a legal research company. “I’ve had a lot of interaction is with judges and all very clearly portray the fact that the future is very much there for AI and technology to be used in a big way in the judiciary.

“Today what does a judge do? He spends a lot of time in analysing the case and analysing similar cases that have happened in the past, and can not end up giving a judgment that is contrary to another judge… They spend a lot of time worrying about all that. However, if there were tools that did all that and gives a judge three option to consider, it will make their job a lot easier.”

Amlan Mohanty, tech lawyerAmlan Mohanty, tech lawyer

Along similar lines, Mainak adds: “There is some talk around the idea that you will help out judges or potentially replace them with sort of like AI, in smaller or clear-cut cases where it’s a simple question of calculating exact amount of damages or things like that.”

“Globally there’s a lot of news around use of AI and the fraternity,” notes technology lawyer and policy advisor Amlan Mohanty from PLR Chambers. ML could for instance assist the judiciary in handing down consistent sentences that reduce human biases, to ensure two defendants in similar situations do not get different jail terms depending on their background or in what part of a country they’re in. That is “definitely something we need to look at in the Indian context” – and discussion of the shape of possible future applications, as well as the ethics surrounding it, would be helpful at this point, he says.

At heart, something like sentencing where lots of hard data and numbers are potentially available lends itself well to the kinds of big pattern recognition that computers and especially ML is famously good at. But much of legal work is less numerical and more linguistic, which has traditionally made it tougher for computers to be helpful, until ML came along, which can increasingly begin to understand the de facto meaning of words and nuance of language.

Big data diligence

In law firm-related legal work, the hopes are high.

“I’ve seen simulations of some of these things and I’ve seen the claim being made that it’ll make interns to A1 and A0 level associates redundant in a couple of use cases, such as finding a precedent or due diligence,” says Mohanty.

The current reality is not quite there yet, though. Mohanty adds: “In terms of specific apps and services, I have reviewed a couple of them and my sense is that they don’t actually go beyond what rudimentary services do right now do. I think it’s quite a misnomer to call them machine learning or AI: it is often just [traditional approaches] packaged together.”

Oberoi Hotels & Resorts general counsel (GC) Dharmesh Srivastava notes that while AI entering the Indian legal industry would be a “very positive development”, he has also not seen much of it in use by law firms, though a Big Four chartered accountancy firm has been pitching something at the contract management end.

They are both right. One of the biggest problems with AI and ML, not just in the legal industry, lies less in what it can do right now but more often in figuring out what it can actually be used for.

One big law firm partner with a technology company background says: “In India you have very limited analytics and almost zero artificial intelligence right now. We see it happening as almost inevitable, but even three to five years is a fairly aggressive estimate in it becoming mainstream.” Instead, even many big Indian law firms were still at the stage of implementing better timesheet management, performance management and e-learning software and databases.

“As a client I haven’t even seen one law firm using AI,” agrees an India-based general counsel (GC) of a VC fund, expressing scepticism about the current state of the technology. “They may be able to automate some very basic tasks, and it might be in a law firm’s interest to reduce its cost but I don’t know if clients necessarily will see that benefit.”

“From what I understand, [AI] is dealing with pretty basic stuff right now,” he adds.

The first steps

Law firm Cyril Amarchand Mangaldas made headlines a year ago, also in the mainstream press, when it announced that it had been the first in India to sign up to software from Canada-based legal AI company Kira.

At heart, while the technology certainly appears like a powerful invention of machine learning, Kira acts as a tool to assist lawyers to quickly identify clauses in contracts, sort of acting like an old-fashioned Microsoft Word text search but on steroids: rather than simply finding the phrases “change of control” in a document, for instance, Kira bills its software as understanding what a “change of control” provision actually looks like, even if that exact phrase does not appear.

“It’s not providing legal advice, it is making the process of the repetitive and more monotonous work much more efficient and much faster, for big teams and groups of lawyers to review the information,” says Toronto-based Kira spokesperson Sondra Rebenchuk. “And it often finds things that a young and experienced lawyer has missed.”

In-house, particularly globally, has also seen companies look at more adoption of AI in legal departments, though not much has made it to India yet. One India GC of an international bank says: “While we don’t have it in our Mumbai office, we are looking to implement [AI] in our international offices.” She adds that AI was still in its infancy on the legal side, but a committee had been set up in the bank’s global HQ to look at how and where it can be implemented.

A matter of distribution

But arguably, the future has already arrived in India legal, it’s just not evenly distributed yet and in all sectors. Venture fund Sequoia Capital’s India director legal Sandeep Kapoor, based in Bangalore, says that legal technology and AI has been a major passion of his for around five years now and has already begun revolutionising the provision of corporate legal services, from compliance down to automatic contract and term sheet creation. “It’s just a matter of time before it reaches tipping point,” he says.

Sandeep Kapoor, Sequoia India director legalSandeep Kapoor, Sequoia India director legal

Kapoor explains that Sequoia has developed several internal tools for Sequoia Capital India that rely heavily on ML to assist its portfolio companies. One of these is capable of automatically generating large parts of term sheets for potential investments, that get automatically turned into checklists for founders and can generate standard subscription agreements at the click of a button.

Much of this is work that has traditionally been done by law firms. “If the community hears the results, they will stop sitting on the fence,” he exhorts the legal profession. “You guys are losing something if you think you are protecting your relevance and practices by not implementing it, and you will have already missed the bus.”

Tech-led corporate compliance plays have been prevalent in India for a while now. One such provider, Lexplosion, which was founded by group of corporate and in-house lawyers 11 years ago, has begun investing into “very rudimentary forms” of ML around 18 months ago, says CEO Indranil Choudhury.

With each state having myriad rules and regu­lations that can intersect and affect businesses in unpredictable ways, compliance is tricky and expensive work. “Law firms have historically avoided compliance, they see it as more mundane work which is not exciting for their associates and lawyers,” notes Choudhury.

Likewise, smaller companies, which may avoid compliance advice due to high costs of lakhs of Rupees in advisor fees, can potentially now afford to become compliant at a fraction of the cost, increasingly assisted by ML. Lexplosion and other start-ups in the space such as OneDelta, have begun to offer services where a series of simple questions answered by a company are theoretically enough to produce compliance checklists and more. Big Four consultancies too have been pitching automatic compliance technologies, some with an ML bent.

Choudhury says that the company has been working hard on further enhancing its ML products, to enable the system to further predict and guide small and medium-sized businesses at the kinds of compliance requirements that would crop up in future, even before a company would consider thinking about them. “There is a fair amount to be done with compliance [and ML] itself,” he says. “It requires all aspects of Indian law, whether Companies Act or GST Act – everything gets covered through the compliance tool – what is critical is to be able to predict what would be the next set of compliances that would come up if you are in a certain line of business.”

The problem that is Indian digital data

In part, rather than availability of technologies, a bigger problem in India – both on the litigation and corporate side – often lies with the availability of good enough data to train ML properly (see box).

What is the difference between AI and machine learning?

Artificial intelligence (AI) and machine learning (ML) are often used interchangeably, though AI is a slightly broader term that could as much be applied to a chess playing computer programme as to a science-fiction killer robot enslaving humanity.

ML generally refer to a set of techniques of creating artificially intelligent computer systems by having a computer create a virtual map (or a so-called neural network) of a lot of information, and then predicting and classifying unknown information on the basis of probabilities.

For example, if you train an ML algorithm by feeding it very selective data: i.e., a million cat pictures, and then a million dog pictures, and then a million pictures that are neither of dogs nor of cats, in theory you could show the computer a wholly new picture and it may be able to tell you if there is a dog or a cat in the picture. Similarly, you might feed an ML system a thousand love letters, followed by a thousand pieces of hate mail, and the computer may then be able to tell you whether the author of a new piece of text is angry at or in love with you.

These might seem like a simple task for humans, but reliable and flexible automatic image or language recognition is something that has until more recently eluded even the most powerful and clever computers.

But ML is not magic. For one, it can fail in surprising ways: for every 99 times that they correctly identify a cat in a photo, there will be that one time that it will think a cat is actually a potato, and you will have no real idea why it thought so. Also, the systems generally live and die by the amount and the quality of data they eat. Therefore, if you feed garbage into an ML algorithm, you are still likely to get garbage out.

Garbage in this case includes data that is not categorised or classified in any way, though even there, sometimes ML may find surprising patterns in what humans thought was random data.

A start-up hoping to solve the problem of India’s vast quantity of non-digital commercial documentation is called Leegality, which is in the process of rolling out a legal documentation platform to let anyone with an Aadhaar biometric ID card sign a legally binding contract electronically (as opposed to the current more complex process that requires application for a digital signature). They had even solved the tricky problem of how to pay stamp duty on digital contracts, said Leegality’s founder, Shivam Singla.

Once a majority of a company’s documentation is digital, Singla says he expects that AI can really come into its own. “Implementation of all these advanced technologies, for them to be able to affect the process as a whole is going to take at least two or three years. Five years down the line we’ll begin to see some of the processes becoming optimal.”

“The larger issue of course is the whole idea of digitisation of documents in India, we haven’t moved the needle far enough on that,” says Mohanty. “Just look at basic issue of looking at Supreme Court judgments [on the Supreme Court website]. It’s really hard [for a layperson]…

“Indian Kanoon took a leap of faith and managed to pull it off, but it’s hard to find access to court records. But [court] submissions, written submissions, pleadings: none of that is available from India. A lot of these materials don’t exist that are being used for other countries.”

Indian Kanoon – kanoon meaning law in Hindi – is a well-known website not just in legal circles, and it usually turns up as one of the top Google results for the name of any Indian judgment. It was set up in 2008 as a part-time passion project by former Yahoo engineer Sushant Sinha, which has now turned into a full-time occupation for him.

Sinha has near-perfected the craft and art of laboriously automating the scraping of the latest court judgments and orders from India’s Supreme Court, all high courts, two district courts and 11 tribunals, most of which operate utterly incompatible websites, systems and databases. Sinha has downloaded all those and made them freely accessible and searchable on a single portal.

Language barriers

The government took a huge step in 2016 by launching the eCourt website, which provides access to all district court orders and judgments across India on a single portal, but much of that data is not actually easily consumable by machines.

Sinha explains that the eCourt website makes documents available in PDF format, rather than text or HTML, which presents an additional hurdle. On top of that, some bizarre technical decisions made by the government in building eCourt, means that even converting those PDFs to data is exceedingly difficult, particularly when talking about vernacular Indian languages such as Gujarati.

“If it was Unicode,” says Sinha, referring to a standard way of storing non-Latin alphabet text to be universally readable by computers, “I would have had no problem integrating all the district courts.”

Instead, eCourt non-English data is often embedded in PDFs or images with proprietary fonts where several characters may make up a Hindi or Gujarati glyph, which need to be laboriously reverse engineered into Unicode in order to stand a chance of a computer actually reading the content.

“I did that with the Gujarati parliamentary debates,” Sinha recounts. “It took me like three months to do the font engineering.”

As it stands, said Sinha, he had managed to digitise around 80% of the English-language judgments available on eCourt from the Bangalore and Delhi district courts, but he was focusing his efforts elsewhere for now.

That problem of inaccessible data goes even deeper: what is available digitally in India, is not nearly enough for ML to really begin to automate some of the most basic tasks lawyers have to perform in their day-to-day work or in a due diligence.

“Here the problem holding up many AI solutions is that records are tied up [in government or private databases], and none of that is customised to include searches for government of Rajasthan land records,” says the law firm partner, giving an example of the barriers you quickly run into.

“Say you’re doing a due diligence of a data room and you run it through: [the ML] is only able to do contracts. When you separate the fancy branding, 80% of the efforts have gone into contracts, but it can’t go into litigation searching or environmental clearances or real estate. None of that is online.”

However, predicts Leegality’s Singla, AI can also potentially come to the rescue of documents and records trapped in bad PDFs in vernacular languages, via optical character recognitition (OCR). “Google and all are coming out with advanced language services,” he says, adding that a few companies in India too are developing and offering sophisticated OCR for companies to digitise old documents.

Institutional barriers

Trying to get the Indian government to improve its websites or data access is nearly impossible, except as a top-down initiative, with individual requests or submissions usually getting mired in the bureaucracy.

Likewise, the implementation of AI in the judicial system could be as much of a human challenge as a technological one.

“The irony is that not too many of the judges are tech savvy,” the CEO of the legal research company says. While most judges profess to have an interest in reducing the overall pendency levels, understand that they need to become more efficient, and understand that more technology such as AI will be needed in the near future, internal decision making remains very very slow-moving. “They were looking at intelligent tools they can use,” relates the CEO about his experience. “but all of these tools would need to be passed by a committee, and the committee may not have tech savvy folks, so they may not necessarily approve it.”

Local solutions

In part because of the challenges, there has not been a lot of innovation yet in India in legal tech and AI, compared to abroad.

Mainak regularly conducts technology patent searches in India and says that not a single patent relating to legal AI has been filed in India to his knowledge: “I am not sure there are any big companies right now which are spending good R&D money in developing AI in the legal field… but it’s possible some is happening and are using it as trade secrets [without filing patents].”

Mainak primarily attributes that lack of investment to several factors: unlike in the US, there is a lack of big companies on the software side targeting legal as a potential market, in part because the Indian corporate legal market is still so small, in India. “Law firms would be the only one willing to side with new technology but if you look at it, the sector is quite small: 2,000 or maybe 3,000 corporate lawyers worth their name. That’s why nobody is looking at this in India.”

“Unless IITians [graduates from the elite Indian engineering colleges] start taking a serious interest in law, or some part of law, etc becomes part of their courses, people are not going to know how to apply their software know-how to the law. I think that’s a major problem.”

Searching for low hanging fruit

That said, there is some early activity taking place. Some of the largest repositories of Indian legal data is held by research providers, including major multinationals and domestic players.

Indian legal research platform Manupatra has begun taking early steps to integrate some AI into some of its existing services, by learning from users’ previous searches on the platform to improve its ranking for relevant results.

Manupatra chief technical officer (CTO) SJ Sudhakar explains: “I really have to use [the right] keywords for a good [traditional] search. AI is a great step forward in that direction.” He says that could even help laypersons get a better idea of laws, without being as hamstrung by their understand of technical legal issues.

Search and natural language search (or natural language research) is the first area where AI and ML can theoretically step in to aid lawyers. Much like major web search engines now rely heavily on machine learning to present the best results to a user, there is scope too in legal research though the hurdles are more complicated than simple text search.

After all, while a judgment consists of text, interpreting the meaning of what a judge said is the most important and most tricky work of a lawyer.

Machines don’t come for free (or with a manual)

Sinha says that currently Indian Kanoon did not use a machine learning model in its searches, although he adds that potentially there was certainly utility in the process. But “machine learning is not a free task”, he adds. “First, someone [human] has to tap [and categorise] that data first.”

“How do you derive the relevance of a paragraph in a judgment?” he asks. Sure, you could find out which paragraphs are quoted most in other judgments, but to find out if a paragraph is still good law, or binding precedent, or whether it has been overruled by another judgment requires human lawyer’s interaction and involvement first, at which point ML could conceivably help to identify similar paragraphs and parameters in other judgments.

And while Manupatra has been toying with this idea in one of its products dubbed “authority check”, as an “automated system that identifies later-citing cases”, it comes with a big disclaimer in a demonstration video that “it is not a citator, and does not include editorial information telling you whether your case is still good law”.

“Whether it’s good law, bad law or overruled – it’s not something that we have tried, or rather not something we are willing to rely on currently,” notes Sudhakar, as Manupatra chief operating officer Priyanka adds: “I don’t know whether I’d be willing to let a machine decide whether it’s good law or bad law, but it seems a possibility in future.”

In combination with that, the company has also rolled out several visualisation tools, that can graphically display what case has been cited by which other judges at which times, and a “judge analytics” tool in which ML aims to “draw an inference” as to the leanings of a judge in particular types of cases, explains Priyanka. That is similar in approach to the tool by Lexis Nexis, which is not yet available in India: Lex Machina has mined US case law to aggregates what kind of cases a judge has handled and makes predictions on that basis.

But indeed, as pointed out by Sinha, neither Manupatra’s AI efforts nor features such as Kira’s super-powered clause search and identification came for free without human interaction: Kira, for instance, has a team of in-house lawyers who manually train the AI in the first place, by identifying clauses in pre-existing agreements so the AI can spot similar ones in future; at Manupatra, cases in the database were first manually categorised under various heads, with algorithms then derived from those databases.

The second problem with AI, says Sinha, is that you require good features and use-cases that would actually be helpful and required by end users, i.e. lawyers. “If you don’t have good features, machine learning is pretty useless – you can not automatically derive features from machine learning.”

And that is part of the problem with some of the early technology: it can seem as though AI engineers are looking for problems that lawyers don’t necessarily require solutions to right now or that are fairly workable without AI tech.

Finding that killer application that would revolutionise the industry is also part of the allure, however. Priyanka agrees: “We have to stay invested in it. It’s more like scratching the surface and there’s so much more one can do. The objective is to reduce the research time of the user and make it more exhaustive and smarter. You gotta be working on it.”

The future: Law without humans?

So, where could it all go? Mohanty explains a time after a class of law students he was teaching had heard a presentation on IBM’s Ross AI system, and the big question was whether they felt threatened by it. “A lot of my students said, is this going to cause job displacement and at what level? Will we have to re-acquire skills and reskill?”

Kira’s Rebenchuk notes, reassuringly: “We don’t think that Kira in any way is a replacement for lawyers. What it does is free up time so professionals can devote more of their expertise and brainpower to do more difficult challenging parts of their jobs.”

And perhaps in India, more than most places, that will remain the role of AI for longer. “Unless associates become very expensive, no law firm will introduce AI processes,” muses one India GC, who works for a US multinational bank.

Perhaps. Although often technology has a tendency to surprise, especially those within industries who don’t see it coming. Survival as a lawyer in this century, may very well be predicated on recognising that.

International Disputes
Spring 2018
Digital Print Issue

Welcome Legally India's Spring 2018 Issue

If you would like to receive future editions, please click here to register your interest.

Our Spring 2018 print and digital edition of Legally India, a joint publication by Global Legal Media and Legally India, has a strong disputes flavour, and examines: AI, global litigation risk, GC wishlists and more than a dozen jurisdictions and practice areas.

New articles will be going live on the microsite in the coming weeks. In the meantime, you can view the full digital edition of the print magazine below.

Click to show 4 comments
at your own risk
(alt+c)
By reading the comments you agree that they are the (often anonymous) personal views and opinions of readers, which may be biased and unreliable, and for which Legally India therefore has no liability. If you believe a comment is inappropriate, please click 'Report to LI' below the comment and we will review it as soon as practicable.

Latest comments