Artificial Intelligence and Machine Learning: The future of Legal Services

DeepBlue, thinking

These days when we talk about AI, we’re really talking about Machine Learning. Machine Learning is the term for computer programs which are able to adapt and evolve beyond the capabilities they are originally programmed to have, by using the data it sees to ‘learn’ more about the world.

A good example of the difference between traditional programming and AI is the difference between DeepBlue, the IBM chess playing program which famously beat Gary Kasparov in 1997, and AlphaGo, a Google program which plays the game Go, and beat the 9-dan professional Lee Sedol. DeepBlue was programmed to evaluate all the possible moves in a game of chess, and weigh up the option producing the best outcome. AlphaGo couldn’t do this for Go – as the total number of possible moves and combinations exceeds the number of atoms in the entire universe, so instead it played against itself and human opponents in millions of games of Go, learning the strengths and weaknesses of different strategies. (1)


Legal Implementation

Machine Learning is well suited to legal applications, where the tasks are varied, and require a more intelligent analysis than simply sorting the data. Tools which use Machine Learning are already available for a number of legal tasks – research, discovery, document drafting, due diligence, compliance, regulatory insights and a variety of other tasks. The Legal Robot software claims that it, “enhances or entirely replaces traditional legal processes like contract reviews with an automated intelligent assistant. Using the legal language model, the intelligent assistant flags issues and suggests improvements by considering best practices, risk factors, and jurisdictional differences.” (2).

Traditionally, many law firms have been reluctant to embrace new technologies, and a technology which claims to be able to make legal decisions is something which many would struggle to accept. As the Royal Society, put it, “AlphaGo demonstrated that machines could make moves that looked like mistakes to human players, but were based on seeking out patterns unseen to us. This highlights some interesting questions about how far we would be willing to trust predictions from machines, especially if it is hard to unpack why a machine has chosen a particular course of action, based on a learning process we haven’t been involved with.” (3)

There are however, examples of automated tools which have successfully given legal advice, DoNotPay is an online ‘bot’ that has helped 160,000 people in the UK and New York to successfully challenge parking tickets. These might’ve been straightforward wins, but this is not an isolated incident. An AI program named Brainspace analysed millions of emails surrounding the Enron scandal and discovered patterns and connections that lawyers had missed. (4).


The question in all professional services industries is not whether or not to use AI, it is how to best utilise it. IBM’s AI system, Watson, is used to advise medical staff in hospitals around the world, as well as a diverse range of other professions, including Education and Financial Services (5). The legal profession can make great use of AI, expensive and time consuming tasks such as detecting money laundering and fraud, due diligence, document version control, reviewing contracts, lead generation, amongst myriad others, could all benefit from AI.

The drive toward greater use of AI, led by pressure from tech-savvy commercial clients, competition from financial services firms entering the market, and the price demands on ‘high street’ firms, will also benefit Access to Justice (6, 7). As shown by the DoNotPay bot and other programs, simple legal advice can be effectively dispensed by low-cost tools, which may provide much needed support to the embattled Legal Aid system.


From a specifically Costs centred viewpoint, the often complex job of costs management could benefit greatly from an AI tool which intelligently predicted overspends in advance, based on similar cases and behaviours, even suggesting alternative courses of action, rather than the blunt objects of notification when agreed costs are being approached, and inexorably exceeded. Beyond this, the valuation of bills, responses to PODS, decisions to advance to assessment, and any number of other decisions could benefit from the assistance of AI programs.

‘Soft skills’ could also benefit from AI assistance, sophisticated and ambitious programs like Nudgr, which monitors the behaviour of potential customers on websites, ‘nudging’ them with targeted offers and incentives when they appear to be losing interest, hint at possible tools to monitor client engagement on cases, ‘nudging’ the case handler to contact the client and apply any TLC as needed (8). The publishing of judgements also allows the potential for AI to identify courts and arguments which are more likely to lead to a preferential outcome in a given case, spotting connections too esoteric for the human eye (no doubt a similar system could be applied to Jury Selection in the US and elsewhere).


Questions remain over how ethics can be applied within legal decision making if given over to a machine, and the social intelligence required in many legal situations is something not easily imitated. The risk of ‘pulling up the ladder’ also exists where tasks normally carried out by junior staff are replaced with technology – would this erode the traditional routes to progression too severely? Even the traditional hourly rate charging structure is called into question, how is the ‘work’ produced by a program to be billed?

Perhaps the greatest problem faced by law firms is access to data. Machine Learning requires data, just as experience helps to make a better human lawyer, more data makes better machines. Although the Machine Learning approach will overcome many of the hurdles of data being stored in a variety of formats (it can ‘read’ documents and understand them, rather than relying on well formatted data), it still needs a broad sample. There are around 10,000 law firms in the UK, and the legal and technical hurdles to be overcome in order for their data to be pooled are immense. It is possible, theoretically, for individual pieces of software to learn from anonymised data sets between firms, but this process would be replete with ethical and regulatory issues – perhaps the next legal AI program should work on those first.


Sources and further reading

Written by Daniel Heaton - Senior Associate and Head of IT


The purpose of this blog is to provide information and discussion. Nothing on this blog should be relied upon as a substitute for advice from a qualified Solicitor regarding any actual legal issue or dispute. Nothing on this website should be construed as legal advice or perceived as creating a solicitor – client relationship. Just Costs Limited can accept no liability in contract or tort to any person, firm or company that relies on or makes use of the above, or any part thereof.