The cutting-edge science in artificial intelligence in law is startling, but does the tech currently offered by commercial legal providers match the hype?
It’s a sign of the times: Berwin Leighton Paisner (BLP) now has its own ‘robot’. In September, the firm announced it had teamed up with tech company RAVN Systems in a deal to use its artificial intelligence (AI) platform, known as the Applied Cognitive Engine (RAVN ACE).
RAVN ACE reads, interprets and extracts specific information from documents. It converts unstructured data into structured output in a fraction of the time it takes a human. The firm recently used the platform to work on a 1,000-page contract project alongside a team of four lawyers. According to BLP, the robot finished the work to the appropriate standard within a few hours – after ten days spent configuring the computer program – while the legal team took three months to complete the job.
But despite the fervour that surrounds discussions over AI and automation, it is far from the panacea that it is being presented as. David Halliwell, Pinsent Masons’ director of knowledge, risk and legal services, warns that in the legal technology sphere, it can be a challenge ‘trying to identify the genuinely new offerings from the snake oil’.
‘A lot of people have taken existing products and wrapped them in an AI banner, but there is some credible software out there. It’s more a question of identifying business need,’ he says.
But while the endgame is to use the technology available to offer clients cheaper, quicker and more accurate advice, in reality much of the low-end legal work that the large firms carry out has already had the tech treatment. The technology available to speed up the front-end process of law has for years been impacting in areas such as conveyancing and property, due diligence, corporate and M&A transactions, and even in some areas of litigation.
ROSS – are students bringing robots to a law firm near you?
For those who beat the drum of ‘disruptive innovation’, the story of ROSS Intelligence is almost too perfect: a group of young, IT-literate entrepreneurs studying at a North American university enter a competition to develop IBM Watson technology for commercial applications.
One year on, these poster boys of the tech-led revolution in law have attracted venture funding and interest from dozens of law firms, leaving highly-paid technologists and legal software developers trailing the bandwagon. The reality is slightly different, but the narrative remains potent.
In 2014 a University of Toronto student, Jimoh Ovbiagele, opened an e-mail containing a link to the IBM Watson competition, calling on computer science students across North America to develop entrepreneurial systems harnessing Watson’s capabilities. Ovbiagele enlisted the support of three colleagues – Akash Venkat, Shuai Wang and Pargles Dall’Oglio – and teamed up with a young lawyer, Andrew Arruda, to provide subject matter expertise and help develop their entry, ROSS Intelligence. ROSS narrowly missed out on the $100,000 start-up funding prize. That honour went to a group of students from the University of Texas at Austin, developers of the CallScout prototype app that draws on Watson’s ability to respond to natural language questions to streamline the delivery of social services. However, ROSS has since been granted permission to continue operating within the Watson Ecosystem, attracted accelerator funding from Y Combinator (which typically offers $120,000 in return for a 7% stake in ventures) and signed up as Dentons-backed NextLaw Labs’ first portfolio company.
Much like Apple’s Siri, Google Now or Microsoft’s Cortana, ROSS can respond to questions in plain English. ROSS doesn’t in a conventional sense answer questions it is asked; it scans a large database of cases to provide information that is likely to be relevant to the enquiry. Inaccurate information can be flagged as a false hit, helping the system to run more smoothly next time. ‘Watson is like a brilliant child and we took that child to law school,’ comments chief executive and co-founder Andrew Arruda. ‘We have taught it to understand the law so it can help lawyers do their research. Situated dead-centre of ROSS at all times is a human lawyer. What it does is help the end user answer questions; it is augmented intelligence.’
By allowing huge amounts of data to be mined every few minutes, ROSS, along with a host of similar systems coming to market, could potentially negate the advantage well-resourced firms with teams of paralegals have over their smaller competitors. (It is no surprise that Arruda started his career at a boutique litigation firm.) According to Arruda, ROSS is currently being tested in the bankruptcy practices of 20 top law firms in the US and Canada. The company has also signed up with NextLaw Labs to develop a suite of artificial intelligence (AI)-based research tools.
ROSS, like many advanced knowledge management systems, addressed a basic problem many organisations now face with exponential data growth: subject matter specialists lack the IT skills to manipulate large volumes of data, while IT professionals are not familiar with the significance of its content. As with most successful legal AI systems, ROSS works by focusing on a specific set of documents. The ROSS prototype was trained to read and process Ontario corporate law decisions and statues, though later incarnations have been focused on bankruptcy law documents. These are early days and the system still needs to prove itself in practice. Whether the great story of ROSS’s creation will be remembered in ten years’ time will depend on its ability to meet the needs of the firms it services. But it will also depend on the strength of its competitors. After all, if four students and one junior lawyer really did outmanoeuvre the legal tech incumbents in the space of a few months, all bets are off.
I, law firm
BLP has initially rolled out RAVN ACE across its real estate practice, identifying this as one area with a lot of repetitive work.
The system extracts data such as client names and addresses from Land Registry documents, enters it into a spreadsheet and cross-checks data to ensure there is no duplication, negating the need for junior lawyers to handle time-consuming, repetitive tasks.
‘It has a specific purpose: reading through Land Registry documents and outsourcing contracts to find the same data points in each contract,’ he says.
‘It is disembodied because it’s software, obviously, and it’s AI because, as it goes through each document and interacts with the lawyers at the end of the process, it learns whether what it has done is useful or not and then it will apply that to future tasks.’
Whalley admits that the effectiveness of the computer’s learning process varies with the complexity of the documents being processed, but says it can learn well with fairly limited exposure.
‘This type of tech isn’t going to take off if the fine-tuning effort required from lawyers is too arduous, because they’re not going to spend a lot of time doing that. They will be sceptical about the whole thing.’
But while traditional law firms, such as BLP, can see the benefits of using this technology, for alternative legal service providers such as Riverview Law, emerging technology is an integral part of the business. In January it announced that it had formed a knowledge transfer partnership with the University of Liverpool to work together and use its AI expertise.
Under the agreement, Riverview has access to the university’s computer science research in areas such as AI, text processing, network analysis, computational argumentation and data mining.
Riverview chief executive Karl Chapman says the work done by the university is ‘breathtaking’.
‘It works because it looks at something practical rather than this [IBM] Watson stuff. We’ve got linguistic analysis from [the partnership] already, looking not just at how to extract data but how to interpret it.’
He adds that the long-term goal is to develop a substantive AI platform and the firm will work with the university to develop tools to automate in-house legal work.
In September the firm acquired New Jersey-based knowledge automation business CliXLEX, allowing it to create ‘virtual assistants’ for in-house counsel. The cloud-based platform provides corporate counsel with automated case-management processes for new instructions, so that work can be given to the right person based on previous behaviour.
Chapman sums up Riverview’s ethos: ‘We’re focused on the middle 80% of work that all in-house teams have to do every day of the month, every month of the year, that you can package into long-term contracts – employment law, litigation, commercial contracts and obligation management. We do the things you have to do, but they can be done better on a fixed-fee basis.’
Competition between alternative legal providers is intense. In 2014, fellow New Law pioneer Axiom Law launched its IRIS contracting platform, now being used by BT. IRIS is Axiom’s proprietary, cloud-based technology platform, and involves the capture and translation of text contained in contracts to structured data.
In a submission for this year’s Legal Business Awards, in which Axiom was shortlisted for Legal Technology Firm of the Year, BT’s general counsel Dan Fitz highlighted some of the early benefits of IRIS: ‘[It enables us] to spot key performance indicators, such as turnaround time on commercial contracts, which clauses are most frequently negotiated versus not, and so on, [with a view to] increasing the volume of transactions.’
Axiom executive vice president Sandra Devine says that the firm chose to focus on automation in its legal contracts, which she says can be deployed broadly in corporate work. In June this year, Axiom went live on a deal with Dell to provide it with a technology-backed, managed service, responsible for the capture, management, reporting and analytics of legal and commercial information contained in Dell’s sales agreements globally.
‘At Axiom, we come at automation and technology from a few different directions, but generally starting with the data and the repository,’ says Devine.
‘When the contracts repository is done right, it means structuring the data in a manner that lays the foundation for practical intelligence via data analytics. When repository is done wrong, it means an expensive storage device. With respect to AI, there is certainly a lot of hype on its application to legal. Our focus is not on how to replace legal judgement per se, but rather how to take routine tasks off the plates of lawyers and contract professionals, and how to enable these individuals to render better and more consistent judgement.’
Aside from BLP, other traditional law firms have also made significant investment in AI and automation technology. This includes Pinsent Masons, which in July acquired a majority stake and all technology rights in Complete Electronic Risk Compliance (Cerico), the cloud-based compliance joint venture which it launched with IT consultancy Campbell Nash two years ago.
Cerico, which is deployed on Microsoft Azure’s cloud platform, automates the processes that businesses require to comply with essential corporate legislation, such as the Bribery Act and the Health and Safety at Work Act. It allows clients to ensure employees and suppliers can undergo regular, rigorous compliance checks in a fast and auditable manner.
Richard Masters, Cerico executive chair and partner at Pinsent Masons, says: ‘Essentially what we’re doing is helping corporates support their code of conduct. Everything that is done creates a solid audit trail, so we can see who took decisions and what the basis of those decisions were. It’s putting a technology solution around a very heavy administrative burden.’
However, he says technology does not remove the requirement for human input. ‘If a supplier is flagged as carrying a higher risk, then someone needs to look at that. We’re not removing or replacing compliance professionals, but we are making their jobs a great deal easier.’
‘What’s coming down the track’ – the conversion of Riverview Law
To the uninitiated, the list of hubs in the legal tech revolution would probably not include the North West of England, but from unlikely beginnings, the Wirral-based Riverview Law has quickly established itself as one of the most vocal advocates for using technology to improve legal processes. In the three years since its launch, Riverview has arguably joined the pioneering US provider Axiom Law as one of the brand names for tech-driven New Law challengers.
But that positioning was something of a shift. Riverview first launched in 2012 in a blaze of publicity with the backing of a group of DLA Piper partners, largely targeting small and medium-sized businesses with an array of fixed-fee products. While much of the sales pitch was a deliberate break with the conventional law firm model, within 18 months Riverview was focusing more heavily on institutional clients.
The backbone of the business is managed legal services for large in-house teams, often handling large contractual, compliance or regulatory tasks. Riverview also covers legal projects and litigation via an allied but independent group of 13 QCs through its Riverview Barristers arm. Unlike a conventional law firm, Riverview makes no effort to replicate practice areas, instead positioning itself as an operational extension of in-house legal teams.
Riverview chief executive Karl Chapman sums up the strategic shuffle: ‘We could see from our previous business backgrounds that if you deploy techniques, processes and technology from other sectors, you would stand a very good chance of succeeding in the legal sector, but it needed to be a managed service model. We found that large companies were very much more receptive than small companies. They were outsourcing certain aspects of their legal work to us that were historically done in-house or by external law firms. Not only were we saving them up to 30%, but the in-house team was freed up to focus on the key strategic elements of legal work.’
Chapman saw the technology moving to a more central part of the Riverview offering. ‘We were growing rapidly, but we were deploying a model that was ten to 20 years old in business terms. People were saying our offering was innovative and we won awards for it, but we knew it wasn’t innovative and disruptive in other industries.’
At the same time, Chapman says, Riverview became increasingly aware of the developments in artificial intelligence (AI) and legal automation that were coming out of US and UK universities and start-ups.
‘We realised, if we are being praised for innovation in the legal services industry by using a model that isn’t new, imagine what we’d accomplish if we used a model and technology that is new. We started to look at the technology and business models that would replace Riverview Law. We wanted to know what was coming down the track.
‘You’ve got a massive technological revolution taking place that you can’t ignore. Existing technologies will make law more efficient, but they will only restructure the existing marketplace. We wanted to know what would happen when next-generation technology hits. A lot of the developments we saw in the US were staggering. From that point we shifted from being a technology-enabled company to a technology-led company.’
Riverview’s focus on new technologies led it to explore a string of bolt-on acquisitions. Most recently the company acquired US-based CliXLEX, a knowledge automation service that can take instructions coming from any department in a business with a legal requirement and filter them into self-service or triage streams. An important aspect of this system, says Chapman, is an ‘integration layer’ that allows it to be combined with a client’s existing IT platforms.
The CliXLEX deal also facilitated a second US office launch for Riverview, with New Jersey joining its existing Manhattan branch. The firm’s approach was illustrated at last month’s International Bar Association conference when one panel debate quoted Riverview’s US vice president, Andy Daws, when asked if he thought Riverview could get good lawyers in New Jersey. Daws reported response: ‘Good lawyers we can find anywhere, the hard part is to find legal technologists.’
Chapman argues that the CliXLEX platform is flexible enough to add on new tools emerging from the company’s partnership with the University of Liverpool. The agent applications, research and technology group at the University of Liverpool is a research centre in the science underpinning AI, robotics and advanced automated decision systems. In late September, Riverview also launched a related consultancy to help in-house teams develop their own automated legal processes. Unusually, Riverview also licenses its technology and platforms for use by in-house teams.
Riverview has so far expanded substantially, having grown to around 150 staff, with additional offices in London and Manchester alongside its Wirral headquarters. The business’s last available accounts in 2014 showed its turnover as £5.01m, having more than doubled from £1.94m the previous year.
Chapman sets out the company’s aspirations: ‘Tech is not only getting more complicated, it is getting simpler. Developers are working on coding platforms that are so simple anyone can use them. Within two hours you can design an end-to-end business system. Configuration, not coding, may be the way of the future. When you start putting the tools in the hands of the people at the front of the business you start to see a big change. Customers have had enough [of the old models of law] and the tech is now available to disrupt the market.’
The spot on the curve
Marc-Henri Chamay, global head of e-business and chief executive of Allen & Overy (A&O)’s subscription service, aosphere, agrees that the AI term is over-hyped and the most significant progress to date has been around process automation. A&O has been using automation in some form for the last ten years, he adds.
‘We use it, for example, in our banking practice, where we have automated a number of banking documents to facilitate the task of our lawyers,’ he says.
According to Chamay, the software can, for example, allow lawyers to create a draft of a complex credit agreement with all the relevant clauses and data within a few minutes. ‘The agreement can then be negotiated by the lawyers who have access to a database of standard clauses if amendments need to be made.’
A&O is currently using Business Integrity’s ContractExpress DealBuilder document automation system, which is also used by a number of other firms, including Ashurst, Linklaters, Eversheds, Dentons, Mishcon de Reya and Nabarro. Clifford Chance has used ContractExpress since 2001, when it replaced an inefficient system of many different manual templates.
The software has proved popular with law firms because it does not require programming skills, only a basic knowledge of document mark-up. Chamay adds that aosphere is also experimenting with how it can roll out software created by Neota Logic.
‘It allows us to build applications that can process complex rules and reasoning. For instance, it could be used to build a marketing compliance application,’ he says.
The Neota Logic software asks the user a few questions about the product or service they intend to sell and then flags up any restrictions or documents that are required for the transaction.
‘The application can also be used to automatically create the documents required based on the characteristic of the transaction,’ says Chamay. ‘The rules are coded by content experts using a standard set of tools that sit on the platform.’
‘We are implementing a major document automation project globally and we are at the stage where we are working through the firm’s precedents and the most commonly used client documents, and automating these. This should decrease turnaround time and bring efficiency into the production process,’ says Royle. ‘Time savings and efficiency improvements will be tracked on a document-by-document basis and will depend on a number of factors, including the complexity of the document and how frequently it is produced.’
Royle adds these documents can include anything from a two-page non-disclosure agreement to a 300-page facility agreement. And they can also be pulled into 600 or 700-page packages.
He adds that the firm’s team of legal technologists take a precedent document in Word format and ‘apply a layer of code over the top of the document’. The document is then fed into the ContractExpress system. Lawyers are asked a series of questions and the code triggers amendments to the underlying document, based on the lawyers’ responses, to produce a Word document tailored to the requirements of the transaction.
However, Royle feels that AI has become an ‘abused term’ in the legal profession and more generally.
‘There is the conception of a general purpose AI and we are a long way away from that. When people realise that, they ask if AI is of any use. Actually that is a false dichotomy, because a significant amount of what can be done is around project management, breaking down legal work into discrete components and asking who the task can be done by. Should it be done by a partner, associate or by software or bespoke AI? It is about moving to the most efficient spot on the curve to get the most value.’
Royle adds that Ashurst has also started using the software for due diligence, which uses machine learning – a form of AI that provides computers with the ability to learn and change when exposed to new data without being explicitly programmed – to refine a set of algorithms so that a computer can pull out material that it has been tasked with looking for.
The legal technologist identifies a contract clause to the software that appears in a sample set of 20 to 40 documents and the machine starts building an algorithm to recognise the clause in other documents. The legal technologist then provides the software with a larger set of documents, and supervises and corrects the software when it makes a mistake. The software is then used to identify and extract the relevant occurrences of this clause into a table, drawing from a very large pool of documents forming part of a due diligence exercise for further review. Additional clause reviews are added to the same table and the software creates a draft review of the due diligence documents as a whole.
Royle believes the legal industry is currently witnessing an explosion in legal technology that did not exist five years ago.
‘Some of the tools we are hoping to use did not even exist six months ago. We are currently seeing a drastic acceleration in pace. There have been huge advances in application programming interfaces in the last three or four years, which are enabling one system to talk to another. These advances are a significant factor in the current acceleration.’
In reality, many large firms have already set off down the path of introducing the available new legal technology. As one lawyer puts it: ‘There is a silent arms race going on in the top firms.’
Tony Joyner, managing partner of Herbert Smith Freehills’ Perth office, says he believes there is a ‘game-changer coming up’.
‘It will require the senior management of law firms to take action. We’ve had tech coming into the law firm model in the past, but it was safe tech. E-mail didn’t change what lawyers do, it just made it quicker. Video conferencing didn’t change the way we speak, it just allowed us to do it over distances. But some of the stuff that gets called AI will change the business model.
‘You can’t just buy this tech and plonk it in the office. It requires a discussion about how to integrate it, how to make sure it delivers for you and the client, and so on. I’d be staggered if the board of any sizeable firm isn’t discussing this. Along with joining the gym and eating more vegetables, it’s the one thing on everyone’s to-do list.’ LB
An A to Z of Artificial Intelligence and Law – the legal bluffer’s guide
A is for Artificial Intelligence (AI) and Law, a subfield of AI research that first emerged as a substantive area in the 1980s. Research seeks to apply the techniques and methods of computer science to legal problems.
B is for Blockchain, a decentralised ledger of transactions distributed to a network of users. Blockchain is part of the architecture that underlies cryptocurrencies like Bitcoin and smart-contract platforms like Ethereum. The public nature of the blockchain means a trusted third party (eg a law firm) is not required to draft and validate a contract.
C is for Computational Law, a branch of research concerned with the automation of legal reasoning.
D is for Data Mining, a subfield of AI that seeks to identify patterns and overlapping features in large, unstructured datasets and extract useful information from them.
E is for Expert System, a computer program that can solve problems typically requiring human expertise in a particular discipline (law, medicine, engineering, etc). Although a number of legal expert systems were developed by researchers in the 1980s, their failure to produce reliable outputs led to most of the software being rebadged as ‘decision support systems’ intended to assist, but not replace, a human expert.
F is for Forward Chaining, a process that reasons from facts to conclusions by repeatedly applying if-then rules. Rules-based AI systems work either forwards or backwards. (In the latter case the system would start from the question and work backwards to find more data that could satisfy the enquiry.)
G is for Genetic Algorithms, an AI search heuristic mimicking the flexibility of natural evolution to allow for principles of selection, crossover and mutation in finding the best available solution to a problem.
H is for Heuristic, an experience-based approach to solving problems in which a system deploys a rule that is likely, but not certain, to lead to a desirable outcome. Much of an expert’s reasoning can be reduced to if-then heuristics (eg, if it looks like the merits of a case are weak, appeal to pity). The Stanford Heuristic Programming Project, which began life as the DENDRAL project in 1965, was one of the earliest attempts to build an expert system.
I is for the International Association for Artificial Intelligence and Law, an organisation formed at the first International Conference on AI and Law in 1987. Since 1992, the Association has published the Artificial Intelligence and Law journal.
J is for JURIX, a Netherlands-based foundation that works at the intersection of computer science and the law. Following the first JURIX conference, held in 1988, it has since become one of the most important forums for AI and Law research.
K is for Knowledge Management, a system for collecting, organising and distributing knowledge within an organisation. If knowledge is defined as information translated into meaning, an AI-based knowledge management system is capable of identifying data as information relating to a particular subject.
L is for Legaltech, a biannual tradeshow, held in New York and San Francisco, showcasing the latest developments in legal technology.
M is for Machine Learning, a process through which computers work out how to analyse data, draw inferences or make more accurate decisions without relying on rigid rules-based programming.
N is for Natural Language Processing, a branch of research that explores ways of enabling computers to read and process natural language expressions, whether they occur in a journal article, a blog post, a court docket or a contract.
O is for Open Texture, the philosophical principle that usage rules for certain ideas or concepts cannot be fixed in such a way that their future application can be rigidly defined. Many of the early computational law systems encountered problems with the open texture of legal language, leading researchers to re-examine their approach.
P is for Petabyte, the currency of Big Data. A petabyte is a quadrillion, or 1015, pieces of information. Although data-based management systems are entering the legal market rapidly, the amount of information a law firm handles is relatively small. Even Big Law does not yet truly deal with Big Data.
Q is for Quantum Artificial Intelligence, a project established by NASA in 2013 to explore how quantum computing could help develop advanced machine-learning tools. Computers act as a series of switches that perform calculations on information that can exist in one of two states, 1 or 0. Because quantum particles exist in a ‘superposition state’, holding multiple properties at the same time, a quantum computer would be able to perform multiple calculations on each bit of information, making them potentially much faster than a normal computer.
R is for ReInvent Law Laboratory, a Michigan State University project established by professors Daniel Martin Katz and Renee Newman Knake in 2012 that brings together thinkers in technology, law and business. ReInvent Law has since spread to other cities and inspired students at a number of universities to look beyond the traditional ways of delivering legal advice.
R is also for ROSS Intelligence, a much-touted electronic paralegal system that can respond to plain language questions without relying on keywords as prompts. ROSS was created by a group of students at the University of Toronto for entry in the 2014 IBM Watson Challenge, a North American contest calling on computer science students to develop entrepreneurial systems. While ROSS narrowly missed out on winning first place, it has since attracted accelerator funding from Y Combinator and has been signed up as Dentons-backed NextLaw Labs’ first portfolio company. (See box).
S is for Semantic Networks, a means of representing knowledge by plotting a series of relationships between objects (eg specific contracts) and types (eg all contracts of a particular kind). Semantic networks allow the information of a domain to be structured into shared concepts that both humans and machines can understand.
T is for the Turing Test, a means of identifying whether a machine can think, proposed by pioneering computer scientist Alan Turing in his 1950 paper, ‘Computing Machinery and Intelligence’. To pass the Turing test, a machine must be capable of deceiving a questioner about its identity as frequently as a human in an imitation game in which both are trying to convince the third player that their opponent is a machine.
U is for Universities, encouraging students to apply technology to the solution of legal problems. Noteworthy examples include Georgetown University Law Center’s Iron Tech Lawyer competition, Michigan State University’s ReInvent Law, Stanford University’s CodeX, and Suffolk University’s Institute for Law Practice Technology and Innovation.
V is for Vacuum Tube, a glass tube containing electrode switches that formed the circuitry of early computers. Colossus, the first programmable computer built in 1943 at Bletchley Park, relied on vacuum tubes.
W is for Watson, the IBM supercomputer consisting of 90 servers and six million logic processing rules. Watson can perform simultaneous data mining and complex analytics on large, unstructured datasets to identify patterns and assist in making decisions. IBM is currently exploring ways to commercialise Watson in a number of sectors, including legal. (See box).
X is for CodeX: The Stanford Center for Legal Informatics, which encourages students, researchers and entrepreneurs to collaborate in designing legal technologies. In addition to supporting a number of leading research initiatives, CodeX is working with Thomson Reuters to promote university-led work on next-generation legal technologies.
Y is for Y Combinator (YC), a venture capital ‘technology incubator’ that invests in and provides advice to start-ups, with past successes including Airbnb, Dropbox and Reddit. YC is now showing an interest in legal start-ups and has just launched Ironclad, an automated assistant that manages legal paperwork.
Z is for LegalZoom, one of the leading online self-help platforms, which, along with competitors like Rocket Lawyer, Responsive Law and LegalShield, is helping to bring affordable legal advice to the mid-market and transform the retail legal services market.