This state senator drafted legislation regulating artificial intelligence technology with the help of ChatGPT.


The bill, called “a bill drafted with the help of ChatGPT to regulate generative artificial intelligence models such as ChatGPT,” would take a series of safeguards, including requiring companies to disclose information about their algorithms to the Attorney General’s Office. Perform regular risk assessments and program the model to include unique watermarks to help detect plagiarism. proposal.

The bill’s purpose is to help “protect public safety, privacy and intellectual property” as newer and more powerful technologies emerge and become more accessible.

It is not exactly clear that Finegold used the program to create realistic-looking bills. The point he is trying to reach.

“What we’re trying to do is try to put in place failsafe-slash-guardrails that can help this technology grow without negative consequences.

“There are a lot of things that can be used in a negative way,” he added. “But it can be very powerful when used in the right context.”

A chatbot that can imitate human language and generate a variety of content in the blink of an eye was unveiled at the end of November. It was an instant hit and has sparked conversations across the internet about its many uses ever since, with people arguing the pros and cons of its amazing features.

According to The Wall Street Journal, the technology is so impressive that Microsoft Corp. recently announced a “multi-year multi-billion dollar investment” in OpenAI, the California-based lab that created ChatGPT.

In December, The Globe did a test run by asking the chatbot to create a series of comical scenarios, delivered in traditional newspaper format, based on a series of Boston-related prompts. Not disappointed.

Finegold, along with his chief of staff, Justin Curtis, experimented with the program to write the bill he presented Friday.

So, how well has technology served its political mission?

“I thought it was pretty good. I thought that was fine. “[ChatGPT] I got about 70%.”

Curtis said it took several tries to get the program to understand the allocation. At times, ChatGPT has struggled to combine information and language that reflect human-written legislation.

For example, Finegold and Curtis on ChatGPT Mass. When asked to draft a General Laws-style bill, the chatbot apologized and said, “Unable to draft a bill.” (Perhaps you sensed where this bill was headed.)

“It definitely needed a little bit of a nudge,” Curtis said. “One of the first attempts we made was a complete rejection of the attempt.”

Finegold’s office shared a series of screenshots of them trying ChatGPT to fill out an invoice. The image shows that you need to give very specific clues to get the program’s output across the finish line.

They were impressed with the speed of the program and its ability to add its own context to the bill, but Curtis said the proposal itself ultimately needed to be fine-tuned before the final product passed muster at Beacon Hill.

“I think the human mind is still better,” said Finegold.

As part of the bill, ChatGPT was asked to include a line indicating that the proposal was written with assistance. ChatGPT complied, but couldn’t help but accept what Curtis said was “a bit of digging.”

ChatGPT notes that “any errors or inaccuracies in the bill are due to human authors, not due to language models.

Finegold retained the line in the final version of the bill.


Steve Anear can be reached at steve.anear@globe.com. Follow us on Twitter @Steven Nier.


Leave a Comment