A forthcoming breakthrough in smart algorithm systems will certainly revolutionise the entire economy, much as internet access has become universal. This revolution will not just impact opportunities for finding work in professions such as translator or driver, but will also completely transform the operations of the justice system. This was demonstrated in a recent competition to predict the results of court proceedings between a group of lawyers and an algorithm created by an English startup.
We have written before about the approaching changes in the legal professions under the influence of the spread of artificial intelligence. The legal profession is becoming ever more aware of the opportunities for use of algorithms (lawbots) for such time-consuming activities as identifying facts or analysing extensive documentation. Recent events have nonetheless shown that we are ever closer to a technological revolution in this area (aided by such events as the forthcoming Global Legal Hackathon), while smart algorithms, with their inbuilt efficiencies, can already beat entire teams of human lawyers.
Virtual judges—the CaseCruncher Alpha algorithm
The first real clash in history between lawyers and their algorithmic counterparts took place in the UK at the end of 2017. A challenge was set by Cambridge students who had founded a startup called CaseCrunch and created a programme called CaseCruncher Alpha. The application predicts the results of legal disputes and calculates the related risks.
Over a hundred London lawyers specialising in work for large corporations took up this challenge with AI. Their task was to predict the results of legal disputes in over 750 actual cases involving insurance claims. Most of the disputes involved actual complaints lodged with the British Financial Ombudsman regarding the misselling of payment protection insurance instruments. The lawyers and the computer programme had the task of predicting which of the complaints would be denied and which would be upheld by the Financial Ombudsman. The lawyers correctly predicted the results in 66.3% of the cases, while the application achieved a success rate of 86.6%.
Such a gap in the results was a great surprise even for the algorithm’s creators. It is necessary to remember, however, that this programme was designed for a certain kind of case, where the problem areas mostly involve formal issues, analysis of factual states, and churning through an extensive base of existing case law. PPI cases were thus an ideal test bed for AI mechanisms, which rely on analysing great quantities of initial data to find patterns that can apply to new cases.
The test results might have been diametrically different had they been conducted in different conditions, e.g. if the legal problem were not so precisely defined or there were more lawyers taking part in the experiment who specialised in cases involving improper use of PPI instruments. But the experiment did show that predictions of the coming role of AI in the justice system were not just abstract meditations by futurologists.
Although the algorithm was only intended to predict the final decision made by the ombudsman, the operating model could be applied directly to analogous cases. A field where similar programmes could be used, and the founders of CaseCrunch could next direct their attentions, is to resolve cases that occur in large numbers and are easy to analyse. This would significantly ease the pressure on traditional judicial decision-making, as judges could focus on the issues requiring their personal attention. It can be assumed that such solutions will pass the test above all in common-law systems, which are based on extensive precedent and laborious analysis of lines of adjudication even in more straightforward cases. But for them to function properly and accurately resolve disputes, they will need to be fed an extensive and accurately described base of data enabling models to be developed through machine learning.
It is nonetheless also easy to imagine the use of similar algorithms in the Polish system. They could conduct initial, formal screening of claims and legal remedies, conduct registration proceedings or make entries in land and mortgage registers. The key issue at some point will be how to distinguish between cases that are completely typical, which the system will easily handle, and the most complicated cases, whose resolution will require application of general legal principles or balancing of competing social values. The human factor will be essential to decide at least on key issues in such cases.
Virtual legal algorithm teams
The CaseCruncher experiment shows that algorithms can be applied not only at the initial stages of legal disputes, which come down to the aggregation of factual states and analysis of extensive documentation, but can also finally decide on the basis of a claim. Such steady creation of divisions and specialisation, with specific programmes for resolving distinct issues, is part of a trend already visible on the market. There are hundreds of legal startups seeking to create lawbots automating particular stages of lawyers’ day-to-day work. The market offers algorithms specialising in automated forms for starting businesses (Rocket Lawyer from WilmerHale), trademark and patent protection (TrademarkNow, Google Patents), compliance (Turbo Tax, Visabot), or drafting specific types of agreements and ensuring their confidentiality (Docracy, Beagle). More and more repetitive legal tasks (normally entrusted to young practitioners) are becoming susceptible to automation.
Apart from professional products intended for large, specialised law firms, algorithms have also started to appear on the market aimed directly at consumers. Paradoxically, these consumer products may contribute the most to changing the global perception of the legal profession. They are typically completely free of charge, and operate on schematic outlines familiar to users of such common technologies as social media and other online services.
One example with the highest recognition is the portal DoNotPay. It initially arose as a straightforward chatbot helping users in appeals against parking fines in London and New York. In 2016, it lodged successful appeals against fines totalling over USD 13 million. Its creator, 20-year-old Stanford University student Joshua Browder, was recognised by the Financial Times as one of the ten most important legal innovators of 2017. Browder is now expanding his product to help users obtain quickie no-fault divorces or compensation from airlines. He has raised over a million dollars for his research from one of the largest venture capital firms in Silicon Valley.
Another portal intended for end users and making use of smart algorithms is Justice Toolbox and its attorney analyser tool. Users describe a legal problem and the algorithm matches it to a lawyer. The portal analyses thousands of cases, divides them into categories, and draws up law firm ratings based on its statistics. This allows the service to present users with specific proposals for legal assistance, along with objective data enabling them to make an informed choice. The graphic display and operation of the service resemble popular aggregators of film and restaurant reviews. Despite its relatively small database (the service is available only in the United States for the time being), the portal shows an interesting trend that may lead to an increase in consumer awareness on the legal services market.
These examples show that the slow assault by artificial intelligence on tasks that have been the sole domain of lawyers has already started. Although the media often announce revolutions prematurely, and autonomous AI is still a long way off, changes in this field are unavoidable. The only question is how fast the legal profession will be automated.