Artificial Intelligence technology has developed by leaps and bounds over the past two years. The most popular application of A.I. in our everyday lives has been OpenAI’s ChatGPT, a chatbot that uses deep learning to perform natural language processing tasks.
By some estimates, ChatGPT has over 300 million weekly active users, mostly using its free web or mobile app. (I used Google’s Generative A.I. feature to find that fact!) In other words, a population roughly equal to that of the United States uses ChatGPT at least weekly to complete homework, write social media posts, and find answers to life’s complex questions.
That number is growing fast. I am seeing more and more intersection of A.I. with my daily practice. Recently, a high net worth client with a software engineering background told me he consulted a chatbot for answers to specific, complex legal questions involving his relative’s Will, a Trust, New York Trusts and Estates law, and some $20 million in family assets.
Given his pedigree and our conversation, I inferred that he scanned the 100-page Trust Agreement into a state-of-art, paid chatbot app that uses text-recognition software (i.e., not free ChatGPT), which analyzed the document against all that it could learn about New York Trust and Estates law. Then, it quickly and confidently spit out answers.
It’s a good time to say that none of this offends or worries me. I myself am a big user of, and investor in, A.I. technology. I believe that, if used correctly, chatbots like ChatGPT can bolster, not replace, a thriving law practice that smartly leverages the technology.
What does worry me is my own observation that very frequently, A.I. chatbots give comically wrong answers to even the most basic questions of law.
As an example, consider a test case I ran this morning. I asked ChatGPT the most basic and common question I thought anyone could possibly ask about New York Trusts and Estates law:
“What is the order of succession when someone dies without a Will?”
This was ChatGPT’s response:
ChatGPT’s summary of New York law is completely wrong in the most common instance- when someone dies without a will, survived by a spouse and children. This isn’t even arguable. (The joke is that we attorneys love to answer “it depends” to most legal questions. This is not an ambiguous law. It does not depend.)
Any human who bothers to find the correct statute governing New York intestate succession (Estates, Powers, and Trusts Law Section 4-1.1) would know that when someone dies without a Will (intestate), survived by a spouse and children, the spouse does not receive one-third of the estate! The correct answer is that the spouse receives the first $50,000 of a decedent’s property, plus one-half of the remainder. The children split the rest equally.
Somehow, ChatGPT flubs this lay-up. If I were to guess, the bot appears to have confused the laws of intestate succession with a spouse’s right to assert a 33% “forced elective share” against a Will that does not provide for them. I cannot emphasize this enough: these are two completely different legal issues, arising in two completely different proceedings, with two completely separate sets of rules.
What’s truly frightening is that, in my experience, potential estate clients are almost never asking ChatGPT to answer questions this basic. After all, most humans can find EPTL 4-1.1 pretty easily.
Instead, I commonly see clients asking chatbots to synthesize, interpret, understand, or summarize complex estate planning documents, which they themselves have interpreted or summarized. (Yikes!) Or, asking a bot to provide a summary of law for a nuanced legal issue that could be governed by a Will, statutes, local court rules, and 250 years of case law hidden behind a paywall- like Lexis or Westlaw- which is inaccessible to the bot. (Double Yikes!)
If modern iterations of chatbots can’t even comprehend the most fundamental building blocks of an entire body of law- “black-letter rules,” like the statute governing intestate succession- what are the odds they could provide accurate legal advice on the complex legal issues that build upon those rules?
It goes without saying that terrible legal advice can have terrible consequences. Back to my client with the $20 million estate. Imagine the deceased was his spouse, who left no Will and had children from another marriage. Now imagine he ran the same query I ran above.
ChatGPT, Esq. would’ve wrongly advised the client that he was entitled to $6,666,666.66, or one-third of the $20 million estate. In reality, the client would be entitled to $10,050,000.00. That’s a four million dollar mistake! Let’s hope the folks at OpenAI carry malpractice insurance.
Disclaimer: The information on this website is not legal advice. It is for information purposes only. No user of this site should act or refrain on the basis of this information without seeking legal counsel. This website does not create an attorney-client relationship.
Photo credit: Shutterstock image no. 659260093