Law Firms are Preparing for an AI Revolution, But Can They Prepare Fast Enough?
Right now, law firms and attorneys worldwide are readying themselves in one way or another for the inevitable impact of artificial intelligence—including large language models like OpenAI's ChatGPT—on the legal profession's future. And while at this point, it's almost impossible to imagine a future in which AI doesn't play a significant role in the industry, it's also equally challenging to speculate with any confidence exactly what that role will be.
Understandably, most attorneys and legal professionals are simply taking it step by step in the spirit of cautious optimism, assessing the benefits and risks, and experimenting with Chat-GPT and similar technologies to understand their capabilities and potential.
But as intuitive as it might seem, the industry's slow and steady approach to adoption has flaws. More specifically, it's proven incredibly difficult to keep pace with the rapidly expanding capabilities of any emergent technologies today and, most importantly, to understand their potential impacts before they begin to take hold. And the reality is Chat-GPT continues to demonstrate AI's broader ability to evolve much faster than we can decide how to use it.
GPT-4: From Underachiever to Top of its Class
Less than a year ago, terms like "large language model" and "generative AI" were rarely spoken outside of niche tech circles or by those directly involved in the technology's development. It hasn't even been six months since the first public-facing iteration of Chat-GPT was released.
While its eerily human-like ability to engage in conversation and churn out mostly accurate responses in mere seconds was undoubtedly impressive, many users quickly pointed out the initial chatbot's myriad flaws. In addition to concerns about informational bias—many of which remain valid and have considerable implications for the legal industry—ChatGPT-3.5 often misunderstood relatively basic commands and even struggled to solve elementary math equations.
Predictably, the limitations of GPT-3.5 also extended into the complex realm of legal knowledge, as evidenced by its failure to pass the Uniform Bar Exam (UBE) when deployed by research collaborators CodeX and Casetext. However, before anyone fearful of AI's impact on the legal sector could breathe a sigh of relief, OpenAI released GPT-4, which to researchers' astonishment, had been rapidly transformed into a far more capable legal student.
According to the study, GPT-4 didn't just pass the bar; it sailed confidently through both the multiple-choice and written portions of the exam, scoring in the 90th percentile in a blistering outperformance of both GPT-3.5 and average real-life legal prospects. In an interview with Stanford Law School following the results, Casetext's co-founder and Chief Information Officer, Pablo Arredondo, described the rate of advancement as "breathtaking" while noting that GPT-4's expanded capabilities were more the result of the sheer volume of data contained in its underlying model than any legal-specific training.
Notably, however, Arredondo also echoed the familiar sentiment of cautious optimism, expressing excitement regarding opportunities for legal innovation while highlighting the importance of ensuring that large language models "[don't] retain, much less learn from, the activity and information of attorneys."
Join our community and be part of the future of law.
This is an example CTA within a Blog.
Timeframe Unknown: When Will AI Transform Legal Work?
Large language models such as GPT are undeniably quick learners. Does that mean generative chatbots will soon replace new hires at law firms while simultaneously multiplying the broader risks AI poses to the legal profession?
According to those highly familiar with the technology and how it works, the answer is almost certainly no. Arredondo, for example, noted the common confusion between existing AI models and artificial general intelligence (AGI), a vastly more capable and potentially threatening version of AI that most believe will arrive sometime in the distant future, if at all.
There's an argument to be made for law firms slightly dialing back their caution and accelerating AI adoption to secure a spot at the forefront of innovation. After all, generative AI already has several benefits to offer the industry, from improved efficiency in drafting and reviewing contracts to significantly reducing operating costs through automation and addressing the ongoing access to justice crisis.
Are there risks? Absolutely. Even ChatGPT will admit that when adequate security measures aren't taken, its learning model's open-source nature may threaten the security of proprietary data and confidential client information. Moreover, despite its impressive bar exam performance, the overall accuracy of generative AI at this point is hardly something users can take seriously, and relying on inaccurate information can have particularly catastrophic consequences in legal work.
Overall, the trajectory of AI's capabilities and its ultimate impact on the future of legal work remains highly uncertain. Whether the next significant breakthrough will arrive tomorrow or two years from now is anyone's guess. In either case, law firms need to understand and address the risks of slowing down their experimentation with AI as much as the risks associated with speeding up. Ultimately, the best path forward may be one on which legal professionals travel at a reasonable yet accelerated pace while always remembering to keep their eyes fixed on the road.
