What Can Lawyers Expect From AI In The Near Future?
Navigating the AI Hype Cycle
New technology historically follows a hype cycle1, defined2 by the research firm Gartner as: a peak of inflated expectations, followed by disillusionment and leveling out to pragmatic applications that fall short of the initial hype, but provide lasting value.
Gartner published a report3 in August 2023 that predicted the progress of AI. We are passing from peak hype, or the Peak of Inflated Expectations and into the Trough of Disillusionment. The focus of this newsletter is to manage the hype and provide insights and value in this rapidly evolving space.
Applied to the specific needs and uses of the legal industry, the hype cycle peaks as we understand the limits of applying general purpose AI tools such as ChatGPT and Claude to legal work. The disillusionment sets in as we learn the limitations of those tools. Those limitations include:
Lack of precision in crafting arguments – Or, AI that writes like a 1L.
Limited ability to produce valid citations.
Missing tool integration.
Inability to match tone, language, and structure of work created by lawyers.
Concerns over security and bias.
Time to learn tools and adapt work processes to integrate AI tools.
For example, we can look at the arguments crafted by Claude and see a marked improvement from even 6 months prior, but it and other tools are simply not ready for use as a “cut and paste” author of legal writing.
In addition, emerging tools offer the ability to personalize the tone and language style of the AI generated content to match the human author. Grammarly4 is a good example of this. But legal professionals need far more customization. For example, my ideal AI writing assistant would know that I work in the 8th Circuit. The AI tools need to be able to reference my prior work in related cases, but be able to navigate different areas of my practice. The AI tools need to know that I write a motion for summary judgement differently than a motion to dismiss.
As your guide to AI in the legal profession you might read that last paragraph as peak disillusionment, but I am optimistic. The general foundation of AI has shown broad applicability and potential.
What We Have Gained Thus Far
Here is what we have gained in the last year of our journey:
LLMs have steadily improved – the quality of output in ChatGPT 4.o is substantially better than ChatGPT 3.5.
Tool integration is steadily improving. Microsoft Office Copilot is showing the value of tightly integrated AI features in common tools.
Privacy and Bias concerns are getting resolved, or at least understood and managed.
Value and investment opportunities are clearly understood. We continue to see investment in AI tools in general, and legal-focused AI tools are continuing to come to market.
Looking Ahead
Looking ahead into 2025 I see needed growth in the following areas. Failure for AI providers to make progress in these three areas will keep us in the Trough of Disillusionment.
Productivity
The financial success of a lawyer and a law firm is heavily correlated to their ability to manage time for peak productivity. My focus is providing expertise to apply AI to improve productivity, but tools need to continue to evolve.
Legal practitioners can look for valuable productivity gains simply by applying AI tools to simple, redundant, or menial work. Staff can automate steps in document processing and authoring. AI tools can even reduce support staff overhead by performing work. These operating cost gains are tangible and valuable, but only the tip of the proverbial iceberg.
I am hopeful and excited to see the emergence of LLMs trained and focused on legal documents and concepts. This will most likely take the form of a plugin for Microsoft Word that offers a Copilot style authoring environment that is a mix of chatbot and inline suggestions. This type of system should deliver deep productivity gains directly to lawyers.
Tools
Westlaw, Harvey, Spellbook, Robin.ai, and others are pushing ahead and gaining market traction. As these firms compete and stake out market segments we should see continued improvement in capabilities and features.
The cost to train a custom LLM is significant. Tool providers will need capital and expertise to complete this work. Harvey.ai has raised over $100M in funding to do this in partnership with OpenAI5. This tool is in limited release, with sales focused on the largest law firms.
It will take time for highly trained models and deeply integrated tools to work their way to general availability at an affordable price point; it took OpenAI nearly 10 years to open their AI to the public.
We should also expect continued integration of AI into tools such as Westlaw, where AI powered document classification and search will continue to improve. The companies providing these tools have a lower level of technical challenge since they are integrating AI into core features, rather than building new features from the ground up.
Finally, general purpose tools such as Microsoft Teams, Google Search, and social media will improve and optimize our “non-billable” time.
Tool Accuracy
As mentioned above, I predict that legal-specific LLMs will emerge, perhaps trained from the start solely on legal corpus, or built as an extension to general purpose models like Mistral. As developers successfully train these tools, we should expect to see increased specialization in the output of AI tasks. This will take many shapes, including document generation that is specific to a type of legal form, but also improved transcription, document processing, legal reasoning, and predictions.
Regulation
There is an adage among pilots that “safety rules are written in blood6”. While rules and regulations of the legal profession might not be as dramatic, I expect to see increased refinement and clarity on how AI can be used in our profession. Some of this clarity will come in the form of restrictions, but will still provide guidance.
I also expect to see (and share with you) high profile examples of lawyers misusing AI. These examples will help shape regulatory action. Unfortunately, we will likely continue to see rules and laws fragmented across states, circuits, and districts – making alignment more challenging.
Like the Gartner consultants from August of last year, there is always risk in predicting the future. Use this article as a hopeful guide to navigating the Trough of Disillusionment as we look forward to the Slope of Enlightenment.