Any sufficiently advanced technology is indistinguishable from magic.
—Arthur C. Clarke
I have elected, somewhat reluctantly, to devote this month’s Technology—Probate column to artificial intelligence (AI). My reluctance is not due to a lack of excitement over how AI will change our lives and the practice of law—I firmly believe that AI will be the most transformative technology of our lifetimes. Rather, my hesitation to talk about AI here is because it oddly feels both too late and too early to opine upon the subject, particularly generative pre-trained transformers or GPTs (the most well-known version of which is OpenAI’s chatbot, ChatGPT). The possibilities of how AI and GPTs will continue to evolve and what they can help us build and change about the practice of law—for better or worse—are simultaneously enthralling, terrifying, and mind-boggling.
Returning to my hesitations around this topic, it feels late to write about the emergence of ChatGPT because unless you’ve purposely tried to avoid the topic, you’ve likely been inundated with dozens, if not hundreds, of articles and podcasts that explain how ChatGPT and all of its AI-focused competitors (such as Google’s Bard and Microsoft’s Bing AI) work, what they are currently capable of, and how they could potentially reshape the legal world. Furthermore, AI is evolving so quickly and its uptake has been so rapid (ChatGPT reached 100 million users within two months of its launch) that articles that are only a few weeks or months old already feel remarkably out of date. This rapid transformation is also the reason any article about ChatGPT feels premature. At this time, it seems nearly impossible to accurately predict how or when ChatGPT will go from something you heard someone talk about at a cocktail party to a ubiquitous tool we rely on daily that is the modern equivalent of the influential transformations of the internet in the late 1990s or the smartphone in the late 2000s. Nonetheless, to observers like me, this transformation seems inevitable in the coming years.
Concerning estate planning specifically, we learned early on in ChatGPT’s existence that it is capable of drafting legal documents. Initial users who asked ChatGPT to draft a will, trust, or power of attorney were often surprised to see the platform respond with language that looked, to the untrained eye, like passable planning documents. However, a closer examination of these outputs revealed that they were often missing important clauses, failed to consider appropriate laws and statutes, and included confusing if not completely incorrect language. OpenAI quickly prohibited its software from honoring requests for legal documents. Many who feared AI’s threat to the legal profession breathed a sigh of relief with the implementation of this ban. Currently, if you ask ChatGPT to draft a will for you, you’ll get the following response:
I’m sorry, but as an AI language model, I cannot provide legal services, such as drafting a will. Drafting a will is a complex and important legal matter that requires the expertise of a licensed attorney who is familiar with the laws and regulations of the relevant jurisdiction. It is important that you seek the advice of a qualified attorney to ensure that your will is legally valid and reflects your wishes. They can also provide you with guidance and advice on other important estate planning matters.
This ban was necessary because ChatGPT wasn’t well-trained at the time. When a user asked for a will, ChatGPT provided a document but often fell short of giving a strong output because it didn’t have access to the best forms nor did it understand why certain clauses were important and how they are interconnected within the document. Similarly, some critics point to instances when users ask questions and ChatGPT answers with suspicious or even provably false answers as evidence that ChatGPT will never be a reliable tool for legal research. Unfortunately, ChatGPT is sometimes inclined to make up data to provide realistic-sounding answers rather than simply say “I don’t know.” But both of these problems are fixable. ChatGPT and other AI competitors can be trained to produce highly sophisticated legal documents and provide more accurate responses to complex legal queries, provided they are given access to the right data and are shown how to use it effectively. ChatGPT, in particular, can be thought of as a first-year associate that has a good understanding of the purpose and basic format of a will but may lack much knowledge beyond this. Unlike a human associate, however, ChatGPT can remember every instruction given to it and can learn and improve its skills at a pace that far exceeds human capabilities. Training a young attorney to draft legal documents correctly can take months or years, but with the appropriate prompts and inputs, ChatGPT can learn the equivalent in a matter of days. Thus, it’s clear that AI will eventually be able to draft flawless legal documents with the right training—it’s just a matter of whether laws or policies will be implemented to restrict it from doing so either with or without attorney oversight.
Although it’s not my intention to fearmonger about the future of our profession, we need to acknowledge that we don’t know exactly what lies ahead. It’s likely that when change does happen, it will be sudden and transformative, and we should be ready for a major shift in how AI will impact our profession. Predicting the exact timeline of this shift is difficult, and it’s possible that a year from now AI may not have had any significant impact on our daily lives. However, this reminds me of a quote from Bill Gates: “Most people overestimate what they can achieve in a year and underestimate what they can achieve in ten years.” I believe this is true for the potential of AI as well. I may be overly optimistic or misguided about its short-term effects, but I’m also aware that we may not fully comprehend the long-term implications. Therefore, we should approach this new technology with both caution and curiosity and continuously educate ourselves to adapt to whatever the future holds by gaining a better understanding of how to work with AI.
Effective use of ChatGPT currently requires repetition. In many cases, the output it produces from your initial request may fall short of your desired result. For instance, if you ask ChatGPT to draft an email advising clients on a planning topic, the first draft may loosely resemble what you want but be far from perfect and thus insufficient to circulate to clients without editing. If junior associates produced a similar work product, you would likely ask them to rewrite it entirely and give them specific instructions for corrections in the next attempt. Then, if subsequent attempts also fail to meet your expectations, your frustration (and the associates’) would likely rise each time you ask for changes and the associates fail to meet your expectations. Working with ChatGPT is different, however. The key to improving its output is to ask it to try again and again, with small tweaks each time to guide it in the right direction. And because new attempts are nearly instantaneous (unlike when working with the associates), it’s possible to be subtle in each of one’s numerous requests. For example, in successive attempts, you can ask ChatGPT to adjust its tone, add new facts, or modify the word count. Gradually, the work product will get closer to what you’re aiming for.
Developing effective prompts and understanding how to interact with ChatGPT takes time, but it’s not rocket science. Some users may give up after the second or third attempt, thinking it takes longer to refine their prompts than it would take to write the email themselves. Although this may be true initially, the process will become faster and easier as users gain experience. Perseverance is the key and once you master giving ChatGPT prompts to help you build what you’re looking for, you will unlock much more of AI’s potential usefulness. Additional tips for creating better ChatGPT prompts include:
And for those still leery of devoting time to working with AI configuration, fear not—the latest GPT development is “AutoGPT,” which is a user interface designed to automatically fill in prompts to make the entire process easier. Many consider this to be the next evolution of ChatGPT and although it sounds very promising in helping new users adapt more quickly to AI, and assisting experienced ChatGPT users to further expand what AI is capable of, it won’t completely replace the need for users to adjust how they approach their work, which requires repetition and experimentation.
In closing, I believe there are an unlimited number of GPT and AI applications that will benefit both planning practitioners and clients alike by providing faster, clearer, more financially accessible, and more accurate estate planning and administration solutions in the future. But AI is not magic. For now, at least, it requires effort and focus to understand how it responds to human prompts to maximize its potential. And, if you haven’t already explored this promising future by experimenting with AI, now is the time to try it out.
