“Come with me, if you want to live!”
Kyle Reese to Sarah Connor-The Terminator
Introduction
Artificial intelligence (AI) (provocatively called by some futurists “synthetic life”) has not only arrived, but it has also been with us for a number of years, embedded in Iphones, and under the bonnet of various iterations of Google. But in the last year, it has exploded into view through the release of ChatGPT in November 2022, and it stands poised to become an embedded technology, potentially as transfigurative of our society as other embedded technologies such as electricity.
In this article I explore what artificial intelligence is, how it is currently being used by lawyers, how it will be used by lawyers and how the adoption of artificial intelligence in legal practice will inexorably mean the end of time based charging and the adopting of value based charging, both between lawyers and their own clients, and in determining the measure of recoverable costs between parties to litigation and arbitration.
Artificial intelligence
In his fascinating book, The Coming Wave, Mustafa Suleyman with Michael Bhaskar (2023) Mustafa Suleyman defines artificial intelligence as the science of teaching machines to learn humanlike capabilities. Artificial general intelligence (AGI) would be the point at which an AI can perform all human cognitive skills better than the cleverest human being.
AGI is someway off from becoming reality, but since 2022 a form of artificial intelligence called large language models (LLMs) has come to the forefront of public consciousness. Put simply, an LLM is a type of artificial intelligence program designed to understand, generate, and manipulate natural language. As lawyers stock in trade are “words” and “knowledge” the legal industry and AI can be seen as natural partners.
These programs or models are called “large” for several reasons:
They are programs which are trained on vast amounts of text data. This data can include books, articles, websites, and other sources of written language, encompassing a wide range of topics, styles, and contexts, literally billions of references.
The “large” also refers to the size of the neural network itself. These models consist of many layers of interconnected nodes (or “neurons”) again, running into billions of nodes. Each parameter helps the model learn some aspect of language, from basic grammar to more complex concepts.
Training and running these models requires significant computational power, often involving advanced GPUs or TPUs and considerable energy resources. If you use an AI, you are deploying vast resources in order to find what may be a specific answer to a particular problem, from all the information on the internet, as processed through the model.
These models are also capable of a wide range of language tasks, such as answering questions, writing essays, summarizing texts, translating languages, and even generating creative content like poetry or stories. Their large scale allows them to have a broad understanding of language and context.
They are typically trained using a method known as unsupervised learning. The model observes patterns in the data it is trained on and learns to predict the next word in a sentence, gradually getting better at understanding and generating coherent, contextually appropriate text.
However, they also have limitations, such as sometimes generating incorrect or biased information: infamously a number of American lawyers have been sanctioned for using AI to generate legal documents, which have been riddled with errors as the AI feels able to make up non-existent caselaw to support the arguments in a case.
How is artificial intelligence currently being used?
AI is increasingly being integrated into various aspects of litigation, offering ways to streamline processes, reduce costs, and improve accuracy. Some of key areas where AI is making an impact in litigation include the following processes.
The key one is that of document review and E-disclosure: AI, particularly machine learning and natural language processing, is used extensively in e-disclosure. It helps in sorting, reviewing, and analysing large volumes of documents to identify relevant information for cases. Tools like predictive coding can learn from lawyers’ document coding decisions to automatically classify other documents, dramatically speeding up the review process. Some of the AI based programs are sufficiently sophisticated that they not only flag particular words or types of documents, but they are also able to make decisions as to whether a particular document is likely to be subject to legal professional privilege.
Although these programs are principally deployed in commercial litigation or arbitration, where the volume of document may be very great indeed, there seems no reason they could not be deployed in other types of case, such as personal injury or clinical negligence, where the volume of disclosure material may be very great and run into many thousands of pages.
AI systems can also assist in legal research by quickly sifting through legal databases to find relevant statutes, case law, and legal writings. This research process should be a step beyond the experience of many users of electronic legal databases, as the interface should be more sophisticated easier to direct and save time.
A smaller user for AI currently exists in the form of predictive analysis: Some AI tools are designed to predict litigation outcomes based on historical data. By analysing past case law, verdicts, and judicial behaviours, these systems can provide insights into how a court might rule on a particular issue, the likelihood of success.
However, such systems are always subject to the (nonlegal) principle of “garbage in, garbage out”. As they will be limited to publicly available data, insight into what has really driven historical cases will necessarily be limited. If a firm decided to use its own data for predictive analysis, it would face two problems.
The first is again, small sample size and absence of all the variables: the second is the privacy and data protection aspects of using client data for purposes additional and extraneous to their case, and the need for data to periodically be deleted, not least because of the right to be forgotten.
AI can also have role in content creation, although anyone who has used ChatGPT to write poetry, will also be aware of its limitations. But many legal documents, certainly in non-contentious work, are often heavily based on precedents, standard forms, or at least use standard clauses. AI can assist in drafting legal documents by suggesting content, formatting, and even providing compliance checks. It also helps in reviewing contracts and other legal documents to identify potentially risky or non-compliant clauses.
Moving beyond the commercial sphere, the insurance industry both on its own account and in conjunction with its panel firms has used databases, to detect fraud. AI can analyse transactional data to identify patterns that may indicate fraudulent activity, claims history, links between places, actors and claims.
Perhaps one of the most entertaining aspects of AI, is the use of Chatbots for the purposes of research, drafting or to act as “co-pilot” for a case. AI-powered chatbots can provide preliminary legal advice and answer basic legal questions but at the current time at too high a level, to provide the deep answers that a human lawyer can provide. They can be useful for increasing access to legal information for the general public or potentially to live on a firm or chambers website, answering questions or taking information.
What tasks can artificial intelligence be used for?
But this is just the beginning. If one considers that in terms of the hardware, the iPad was only invented in 2010, the scope for advancement on a software level is not only very great, but change can occur far faster than on a hardware level. In the next decade there may well be more advanced predictive analytics tools, powered by AI, several leagues beyond the current limited models, as more data becomes digitised generally in society. These tools could analyse not just legal precedents but also consider social, economic, and psychological factors that might influence legal outcomes.
There will undoubtedly be enhanced legal research tools: AI will likely become more adept at understanding and interpreting complex legal questions, offering more nuanced and context-aware legal research assistance. Moreover, as large models take their information globally, from across the internet it could integrate the laws from various legal systems and jurisdictions, providing global legal perspectives.
One can also see that the role of AI in document drafting, and analysis is likely to expand: It might automatically generate entire legal documents based on minimal input, and offer more advanced analysis of contracts, predicting potential legal issues before they arise.
Anyone who has an iPad, or laptop at court, and that is everyone these days, is able to draw upon online legal databases and undertake legal research on the spot: typically to obtain a transcript or judgment of a case that has just become relevant, and which can then be emailed to the judge or one’s opponent. But this could be expanded: it could provide lawyers with immediate information, case law references, or evidential analysis. This could include live fact-checking and rapid response research capabilities.
In the near future, AI will likely offer more advanced e-discovery tools, capable of understanding and categorizing more complex and varied types of data, including audio, video, and social media content.
Proprietorial and open market models
Of course, commercial litigators live in a different world to those lawyers who are undertaking personal injury litigation or other case types, which involve personal claims, rather than commercial claims. Lawyers working in consumer facing firms may not have the capability to develop proprietorial AI based tools in house and may seek to obtain programs on the open market. This may limit their options, and indeed preclude the early adoption of AI.
But it may well be that the dividing line between the two models is illusory. By way of example, Harvey AI is an advanced artificial intelligence solution specifically tailored for the legal profession. Developed on Open. Ai’s GPT AI, similar to ChatGPT, Harvey AI is designed to assist law firms in various legal tasks. It incorporates a combination of general internet data from the GPT model and legal-specific data, including case law and reference materials. When a law firm employs Harvey AI, the system undergoes further training using the firm’s own work products and templates, enabling it to provide assistance that is tailored to the specific practices of the firm.
As the founders note:
Harvey is starting with the legal market, which is $300Bn+ in the U.S. alone. Legal work is the ultimate text-in, text-out business—a bull’s-eye for language models—and Harvey has already begun to make an impact. Harvey helps legal teams find leverage in time-consuming tasks like legal research, due diligence and more, allowing them to focus on client relationships and strategic work that truly moves the needle. In February, Allen & Overy became the first announced enterprise customer. In March, PwC announced it was coming on board. More than 15,000 law firms are on the waiting list today.
https://www.sequoiacap.com/article/partnering-with-harvey-putting-llms-to-work/
The capabilities of Harvey AI include contract analysis, due diligence, litigation, and regulatory compliance. By leveraging AI algorithms, it can analyse data, generate insights, provide recommendations, and make predictions, thereby assisting lawyers in delivering faster solutions for their clients.
Charging clients value-based fees.
I have little doubt that the use of AI will increase, scale up, and also come down in price very quickly, much more quickly than the decades of time for hardware to develop, between the ZX Spectrum and the iPhone as two examples of computers used for consumer electronics. The consequence will be that there will be a reduction in the amount of time that firms spend working on a case, and an increase in the amount of technology, that they deploy on a case.
On a traditional economic analysis this could be described as a shift from labour to capital, and analogies can be drawn going back to the first Industrial Revolution and the adopting of the spinning jenny. But if a disclosure exercise can be done, effectively at the click of a button, then firms to capture the added value they are providing, will have to charge not by time spent, but on a value-added basis.
A de facto shift to value-based fees (fixed fees) will be undertaken. It may be that a law firm sets up subsidiary entities, who undertake, for example the disclosure exercise and charge a fixed fee for the use of their technology, as a disbursement, whilst the hourly rate model remains for the higher and better value work that is done on the case by experienced human beings.
However, as AI becomes more widespread and penetrates more tasks, reshaping workflows, I would suggest there will be an ever-increasing drive to value-based fees (or fixed fees as they are otherwise known) and a marked decrease in the use of the hourly rate.
Assessing costs between the parties
I suspect it will take the practice of the award and assessment of costs between the parties, rather longer to catch up. Or perhaps not, provided it is understood that in some types of litigation conducted by solicitors who use templates, formatted documents and letter banks, a de facto value based charge is often being levied. The item in the bill of costs is charged for as if it is wholly original but this is often far from the case. In the personal injury market, claims management software routinely assigns or estimates time to mundane tasks, and the use of “units” serves to disengage time claimed, from minutes and seconds actually employed.
AI will accelerate and bring to its logical conclusion the concept of “near zero marginal cost” a phenomenon of the digital age, where it costs virtually nothing to provide additional products at minimal, or even zero costs. The music and publishing industries, providing digital downloads of identical products to millions of users are the clearest examples of how services can be provided at near zero marginal costs. There is scope despite its bespoke heritage for much work within a case to be delivered in a similar fashion.
Increasing automation will make a nonsense of the notion that time claimed on a bill of costs, is a reliable indicator of reasonableness. In terms of high value litigation, AI tools permitting key word searches and automation of the disclosure process, will greatly reduce the amount of fee earner time hitherto routinely spent in enormous quantities. How does a solicitor quantify and charge for the use of a software programme in the context of a multi-million pound commercial dispute? Logically, it should be by the added value given to a case.
If fees are charged on a “value added” basis without reference to time are challenged, either by a regretful client or a paying party in a recoverable costs assessment, how should the court approach their quantification? The court’s approach to the assessment of contentious costs is governed by rule 44.4 CPR containing the seven (now eight) pillars of wisdom.
Of these factors only one, factor (f) specifically enjoins the court to have regard to the amount of time spent on the case. In addition, there is respectable body of case law on non-contentious costs, derived from the Solicitors (Non-Contentious Business) Remuneration Order 2009 and its predecessors where a value charge is the norm in areas such as probate, to allow the court to draw upon by analogy, when determining whether a contractually agreed fixed fee is reasonable.
Conclusions
As well as the work I undertake in costs and litigation funding, I undertake more general litigation, historically a large amount of personal injury litigation. I am actively looking at how that part of my work might be streamlined.
I am intrigued by the notion that a client’s medical records and other documents might be summarised and evaluated through AI using e-disclosure, that AI can be used to produce transcripts of conferences held with clients or to save my pen, when taking notes of evidence or a judgment in court.
If my time can be saved by increasing the power of the programs and platforms I use for legal research, or standard forms, skeletons or pleadings can be produced more quickly through AI that will benefit me and my clients.
I would also be interested to see AI deployed at trial, through producing summaries, visual aids or reconstructions through software that can walk a judge through the scene of a contested accident. Although the courts can barely keep the elevators functioning or their ceilings from collapsing at the moment, these things are coming.
Andrew Hogan is a barrister at Kings Chambers, Manchester. His blog on costs and litigation funding can be found at www.costsbarrister.co.uk. This article was partly written using ChatGPT. Can you tell which bits were written by a humourless automaton?
A version of this article was first published in Litigation Funding magazine whose website can be found here: https://www.lawgazette.co.uk/litigation-funding