The robots are coming – and this time it’s lawyers who could feel under threat. Fortunately, artificially intelligent (AI) droids are not, yet, generally celebrated for their powers of advocacy. So while a crisp turn of phrase, engaging presentation, and persuasive argument remain the preserve of humans, advocates will still have an important role. But outside of the courtroom there is no doubt that AI bots are starting to undertake increasing types of legal work.
Bots can now have conversations with people about their legal dilemmas and use algorithms to predict the outcome of their case. Work undertaken by human lawyers for centuries is being computerised.
Elexirr: playing the language game of law?
Leading the way is Elexirr, a bot created by Cambridge students (www.elexirr.com). Their initial ‘lawbot’ helped people determine whether they had been the victim of a sexual offence. They relaunched it this summer as Elexirr – a tool that claims to be able to predict the outcome of virtually any legal dispute with 71% accuracy.
Elexirr uses the Facebook messenger platform. Users have a conversation with the bot using normal language. Elexirr asks them a series of questions about their dispute before providing a Win/Lose prediction. It then offers the chance to connect users to a law firm.
It sounds like an astonishing achievement – even if it is inaccurate 29% of the time. To see how well it works, we put it to the test by chatting to the bot about three fictional legal scenarios: a simple allegation of assault; a petty libel claim doomed to fail; and a complaint about a cowboy plumber’s shoddy installation.
We answered Elexirr’s questions about each of the three scenarios using normal, conversational language. We didn’t use legalese.
Win or lose?
The result? In all three cases the bot predicted the claim would ‘lose’. In respect of the libel complaint, Elexirr was spot on. However, the complaint against the plumber was straightforward and there was no reason why it shouldn’t have been pursued. The assault allegation was a criminal allegation, yet the bot didn’t specifically point that out, neither did it suggest contacting the police. Given that a large number of people don’t know there are different justice systems for criminal and civil complaints, that was surprising.
Elexirr is a tool designed to be used by the public, yet a number of the questions it asked were clunky. In response to being told that the plumber had fitted a leaking bathtub in a rushed job, the bot asked ‘How likely do you think this situation was? Why do you think that is?’
Many users are likely to be financially or emotionally wrapped up in their particular complaint and so may find it difficult to provide an entirely objective and dispassionate view of their situation. Yet, that is what the bot requires them to do, asking questions such as ‘How reliable do you think your account and your evidence is, Sir?’ Or: ’If the other party or parties were to testify before a court, what would be most convincing about their account, Sir?’
It is hard to image many users admitting to the bot that their evidence is not very credible. However, the bot did ask questions that covered the scope of the dispute, the reasonableness of all the parties and the remedies being sought. If an answer was unclear or too short, it would ask for more detail.
So, although it has great potential, our tests suggest that at the moment, for the public user, Elexirr is not the finished product. It needs to improve its language and the fact that it uses the Facebook platform may deter some from sharing sensitive information. While its ‘Win/Lose’ prediction is bold and simple, it neglects the reality that many claims are more nuanced, and a settlement could be reached. Also, to be a really useful tool, the bot should be able to signpost users away from the civil courts – be it to the police, a regulator, or ombudsman.
We’ve seen Siri, Cortana, Amazon Echo and other AI bots produce outstanding voice recognition interaction in recent years – if Elexirr can harness some of that user-friendliness, it could become a valuable tool for the public.
Aside from public users, the algorithm that powers Elexirr can be used by law firms and insurers to calculate the merits of a case. While it’s unlikely to be completely relied upon at the moment, it is not hard to envisage a world where insurers run road traffic accident claims through a bot to determine the position they take.
Encourage or fear?
Should the legal profession be concerned that law bots will take away their advisory work? No, is the simple answer. At least not in the foreseeable future. They are tools aimed at increasing access to justice rather than impeding it – in the same way that the Bar Pro Bono Unit and Citizens Advice Bureau complement the work of the legal profession rather than starve it of work. The likelihood is that a lawbot user who is told they have a claim with decent prospects would consider instructing a human lawyer to act on it.
As Elexirr, and its bot friends, develop and grow, there will be new opportunities for the Bar, particularly lead generation for direct access barristers. The innovative work of teams like those behind Elexirr should be encouraged, not feared.
TOP OF THE BOTS
We put some other legal bots to the test.
Initially launched to help users challenge parking tickets, DoNotPay has been expanded to cover areas including immigration and housing. It was set up by an 18-year-old student who claims it has overturned 250,000 parking tickets.
•The good: For appealing parking tickets, and other fairly straightforward flowchart issues, its quick and helpful.
•The bad: Type in ‘my plumber owes me money’ and it bizarrely offers to help you claim compensation for a delayed train. It doesn’t tell you what areas of law it covers and if you ask for more help, it just opens up a blank email message.
The legal intelligence support assistant is the creation of a lawyer from Yorkshire. It aims to help parties produce a written agreement without the need to resort to expensive, slow and adversarial human lawyers.
•The good: For the 54% of small businesses that don’t hire lawyers, it’s an innovative tool that could help reach agreements, reducing their costs and the risk of litigation further down the road. It is a great idea – if both parties are reasonable.
•The bad: While an online document maker might be fine for simple tasks like creating a straightforward will, many businesses would benefit from a skilful and creative human lawyer fighting their corner in order to secure the maximum benefit.
The brainchild of the forward-thinking Clerksroom, Billybot is an online direct access tool. Users can chat with Billy ‘a cheekie chappy who loves to chat’ to get quotes from barristers.
•The good: By far the most user-friendly of all the bots. It looks good and works well. It won’t be long until all chambers have something like this on their websites.
•The bad: The idea of Billybot as a ‘cheeky chappie junior clerk’ is a little contrived, but it adds warmth and could easily be changed to fit a particular brand.
A tool that allows users to upload contracts and have them analysed to point out potential pitfalls and risks.
•The good: In principle, a good idea.
•The bad: It didn’t work when we tried it.