Posted on March 23, 2023 by Don Bluedorn
Unless you have been hibernating this winter, you know about ChatGPT, the artificial intelligence chatbot that rolled out late in 2022. Its developer, OpenAI, describes ChatGPT as follows:
We’ve trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.
Or, for those of us who remember watching Star Trek, ChatGPT functions a lot like the computer on the Starship Enterprise – you ask it to do something, and it does it. People have experimented with ChatGPT to write computer code, draft poems, write term papers, and create visual art, among many others.
Recently I experimented with ChatGPT, in an effort to not be that “senior” lawyer who in the early 1990’s said, “I don’t need to learn this new email thing . . . .” More specifically, I asked ChatGPT to do two things:
- Draft a short purchase and sale agreement for a 65-acre coal-fired power plant; and
- Prepare a five-page memorandum on the definition of “Waters of the US.”
The results were surprising but instructive.
First, ChatGPT’s purchase and sale agreement was so basic, and so vanilla, that it would be useless to a lawyer hoping to prepare the document in a real transaction. This surprised me because I had heard and read so many glowing reviews about ChatGPT that I anticipated a fulsome work product. It is quite possible, if not likely, that much of this is attributable to “user error.” If I spent more time describing the project and setting forth my anticipated parameters, I expect I would have received a better product. Nonetheless, I deemed my initial effort with ChatGPT a failure.
ChatGPT’s 5-page memorandum on the “Waters of the US” was surprisingly good. Now make no mistake about it – the draft would not pass muster as a writing assignment in even the most basic of environmental law courses, much less as an actual piece of legal work product.
Nevertheless, it provided a basic level of understanding of the issues and it provided a surprisingly sound foundation for such a memorandum. I could easily imagine checking the referenced statements and citations in the draft (unfortunately ChatGPT already has developed a fearsome reputation for confidently stating a “fact” when it is wrong), and then performing supplemental research and analysis to fill out the picture. If I were a new lawyer tasked with preparing such a memorandum and I did not already have a good starting place, the draft would have saved me a surprising amount of time at the outset. So, at least in that sense, I deemed my second effort with ChatGPT a conditional success.
OpenAI is quite open and transparent about many of the limitations in ChatGPT. Among other express limitations, it states on its website that “ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers. Fixing this issue is challenging . . . .” And of course the entire field of artificial intelligence is growing at a tremendous rate, so we can expect to see significant improvements in ChatGPT and similar products in the foreseeable future. Nevertheless, even in its current iteration it would be a mistake to dismiss ChatGPT as a viable tool for the environmental lawyer, particularly in the correct circumstances and with appropriate controls.
And with that final note, “Beam me up, Scotty . . . .”