Blog

Marco Berki of the University of Reading examines the rise of LLMs in legal practice, and the dramatic changes that the legal profession can expect from this rapidly evolving technology. Legal Market Transformed was the Best in Category winner for the 2024 vLex International Writing Competition category: Large Language Models.

Legal Market Transformed: The Impact of LLMs on Legal Practice

by Marco Berki

"Words are all we have," I was once reminded by a City lawyer. From the eloquent arguments in courtrooms to the meticulous drafting of legal documents, the legal profession is defined by a mastery of language and a monopoly over the nuanced worlds of words. Lawyers, as stewards of the law, understand complex legal jargon and navigate statues and case law. This linguistic prowess has been distinguished by them as a profession but also erected barriers to access for those outside the legal community. However, the advent of Large Language Models (LLMs), a form of generative AI, is poised to disrupt this linguistic dominion, challenging the traditional roles and responsibilities of legal practitioners. As we stand on the cusp of a new era, where algorithms can parse, interpret, and even generate legal texts, we must ask: How will the role of lawyers transform in a world where words are no longer their exclusive domain?

LLMs are a subset of machine learning models adept at tasks ranging from text analysis to answering queries. (1) They have been in existence for some time, but it is their recent advancements that are shaking up legal practice. Earlier AI models could only manage simple queries, like those about the weather. However, today's versions, such as ChatGPT, are redefining the field with their remarkable capacity. Their abilities stem from neural networks that interlink words and data through prediction and association (2), a capability expected to double in performance every 3.5 months (3): making it a tool especially effective in handling vast amounts of text and data.

The efficacy of LLMs is so notable that even traditional law firms like A&O are experimenting with them (4). The work of a solicitor, even in prestigious 'magic circle' firms, essentially falls into two categories: formulaic and creative (5). While there is scepticism about LLMs handling creative tasks, their use in formulaic tasks is less contentious. If it is generally agreed, as I think it is, that the work of a lawyer is predominantly to fit factual circumstances into right or favourable patterns (formulaic), then those who are engaged in this type of work will be struck by the proverbial hammer of generative AI the hardest (6).

Consequently, lawyers focused on creative tasks are likely to withstand the AI assault the longest. In "Tomorrow’s Lawyers", R. Susskind decomposes legal work, like litigation, into components from research and analysis to strategy and tactics (7). He identifies, the two latter as the key strengths of top practitioners – not a rare opinion. It is not surprising that tasks related to such creative activities will be best positioned to be complemented by LLMs rather than replaced (8): a safe haven.

But for all the doom and gloom looming large over lawyers’ heads, it is crucial to acknowledge LLMs’ current shortcomings. As previously mentioned, today's LLMs are dependent on existing data, which inherently skews their focus to the past rather than the future. This limitation hampers their effectiveness in tasks requiring innovative and creative solutions and raises questions about liability.

The quality of LLM outputs is directly tied to the quality of their inputs, a well-known fact that has sparked widespread concern over biased data influencing LLMs. Worse still, LLMs may resort to fabrications or 'hallucinations'. The case of Steven Schwartz (9), a New York City lawyer who prepared a court filing using solely ChatGPT, amusingly illustrates how far LLMs are from being a universally trusted independent tool in the legal field. The key takeaway? In a world increasingly dominated by digital technologies, one valuable asset lawyers might offer is trust, acting as a safeguard against the uncertainties of generative AI.

However, it would be dangerously myopic to underestimate LLMs' potential. Regardless of their current limitations, the economic pressures exerted by the wider market are formidable. In stark contrast to the 33% of UK adults who can afford legal services (10), a whopping 97.7% have internet access (11). This disparity suggests an imminent challenge to traditional legal services from emerging law-tech startups, leveraging LLMs to provide legal advice or document drafting. Moreover, this is not just a concern for private clients. The world’s 1000 largest corporations, which sustain the 100 largest law firms and have been cutting legal costs since 2008 (12), are likely to embrace LLMs as a means to further reduce expenses. The future economic pressure will be immense; survivors will cut prices, deploy technology, and thrive, while laggards will try to maintain that the best ways are the old ways and drown in the process.

But those who expect traditional practices to vanish swiftly may find themselves disappointed. In the coming years, we will witness a complex interplay of trends that initially bolster lawyers' dominance in linguistic construction and interpretation (13). This reinforcement will be driven predominantly by law-tech companies, whose primary clientele are existing law firms. These firms are integrating LLMs, such as Co-Counsel, a GPT-4 based American AI legal assistant, to enhance their operational efficiency (14) and solidify their market stance (15). This entrenchment, however, will be complicated by emerging disputes over intellectual property rights, adding layers of complexity to the scenario.

Echoing Susskind, a significant shift is anticipated once an innovative company diverts from the beaten path to concentrate on the long-term evolution of LLMs and their application (17). This shift will likely be propelled forward by the advent of multimodal AI (18), which, with its unprecedented formulaic capabilities, can only be expected to deliver the final blow to those without the sufficient talent to engage in creative legal work.

The essence of legal profession has been encapsulated in its historic reliance on linguistic mastery. With the rise of LLMs, legal practice is shifting; routine tasks may be automated, reducing need for lawyers for formulaic work. However, the future brightens for creative legal work, where lawyers leveraging LLMs can innovate and strategise. This evolution offers a vision where lawyers excel in crafting innovative, ethical solutions in synergy with technology, marking a transformative era where creativity and tech-savviness define success.

References

1. Margaret Rouse, 'Large Language Model (LLM)' (Technopedia 2023) https://www.techopedia.com/definition/34948/large-language-model-llm accessed 27 November 2024.

2. Damien Charlotin, 'Large Language Models and the Future of Law' [2023] Social Science Research Network , 4.

3. P Moorhead, P Moorhead and Karl Freund, 'AI Hardware: Harder Than It Looks' (Moor Insight & Strategy 2019) https://moorinsightsstrategy.com/ai-hardware-harder-than-it-looks/ accessed 26 November 2023

4. Laura Lorek, 'How lawyers can take advantage of ChatGPT and other large language models disrupting the legal industry' (American Bar Association 2023) https://www.americanbar.org/groups/journal/articles/2023/how-lawyers-can-take-advantageof-chatgpt-and-other-large-language-models-disrupting-the-legal-industry/ accessed 27 November 2023

5. Damien Charlotin, 'Large Language Models and the Future of Law' [2023] Social Science Research Network , 18

6. Ed Felten, Manav Raj, Robert Seamans, 'How will Language Modelers Like ChatGPT Affects Occupations and Industries?' [2023] , 15

7. Richard Susskind, Tomorrow's Lawyers (3rd, Oxford University Press, Oxford 2023) 53

8. Damien Charlotin, 'Large Language Models and the Future of Law' [2023] Social Science Research Network , 26

9. 'Generative AI could radically alter the practice of law' (The Economist 2023) https://www.economist.com/business/2023/06/06/generative-ai-could-radically-alter-the-practice-of-law?utm_medium=cpc.adword.pd&utm_source=google&ppccampaignID=18156330227&ppcadID=&utm_campaign=a.22brand_pmax&utm_content=conversion.direct-response.anonymous&gad_source=2&gclid=EAIaIQobChMIydjjrZLOggMVo4toCR1piQaQEAEYASABEgJYT_D_BwE&gclsrc=aw.ds accessed 27 November 2023

10. Lee Marston, 'How Can People in the UK Get Better Access to Legal Services?' (Lawyer Monthly 2021) https://www.lawyer-monthly.com/2021/07/how-can-people-in-the-uk-get-better-access-to-legal-services/ accessed 26 November 2023

11. Ani Petrosyan, 'Internet usage in the United Kingdom (UK) - Statistics & Facts' (Statista 2023) https://www.statista.com/topics/3246/internet-usage-in-the-uk/#topicOverview accessed 27 November 2023

12. Richard Susskind, Tomorrow's Lawyers (3rd, Oxford University Press, Oxford 2023) 121

13. Richard Susskind, Tomorrow's Lawyers (3rd, Oxford University Press, Oxford 2023) 250

14. Rosemarie Miller, 'Meet The Legal Assistant For Lawyers' (YouTube 2023) <Meet The AI Legal Assistant For Lawyers, Rosemarie Miller, Forbes, [2023]> accessed 27 November 2023

15. Tom Saunders, 'Legal tech teams turn to AI to advance business goals' (Financial Times 2023) https://www.ft.com/content/9a117ac7-29ae-43fe-b840-a04005b98799 accessed 27 November 2023

16. Heloise Wood, ‘PA's Conway tells MPs AI Large Language Models breaking law 'on massive scale' (The Bookseller 2023) https://www.thebookseller.com/news/pas-conway-tells-mps-ai-large-language-models-massively-breaking-the-law accessed 27 November 2023

17. Richard Susskind, Tomorrow's Lawyers (3rd, Oxford University Press, Oxford 2023) 251

18. Damien Charlotin, 'Large Language Models and the Future of Law' [2023] Social Science Research Network , 9