Thread by @UpolEhsan

Got obsessed with this story. It’s a wild one, y’all.

It’s about AI’s favorite child ChatGPT, lawyers, and a very angry judge😅

27 Twitter threads & 5 articles later, I finally have a straightforward picture.

The details, juicy. The implications, huge.

Let’s dig in 🍿

1/n

3 characters are important:
- Mata (plaintiff)
- Schwartz (attorney)
- LoDuca (attorney on record).

Mata sues Avianca airlines through Schwartz at NY state court.

Avianca transfers the case to Manhattan’s federal court.

That's when things get interesting🕵️‍♂️

2/n

Schwartz can’t practice there, so he gets LoDuca to be the attorney on record.

Remember, LoDuca does zero research. Everything is done by Schwartz.

Mata's lawyers file the brief.

Avianca asks the judge to throw the case out. Why? Because none of the cited cases exist! 🤯

3/n

Where did the fake cases come from?

You guessed it: ChatGPT hallucinated up the whole thing!

The cited cases were syntactically plausible, seemingly fooling even experts.

So, how did it all unfold? 🤔

4/n

The judge demands proof to back up the claims.

Schwartz goes back to ChatGPT and asks for the cases.

But the problem again is: all of it is fabricated 😬

Schwartz submits the ChatGPT output.

To make things worse, Schwartz and LoDuca sign affidavits. 📝

5/n

Now the judge is super pissed. 😡

Sh*t gets real.

Schwartz confesses he was "unaware of the possibility that [ChatGPT’s] content could be false."

He even shares screenshots where he asked ChatGPT to verify, and it confirmed everything was right. 😱🤐

6/n

⚖️Legal experts claim malpractice implications. Schwartz’s law firm could be in trouble. More details will emerge on June 8, 2023.

So, who’s at fault? What are the implications from a #ResponsibleAI angle?

7/n

The simplistic move is to place all the blame on the lawyers.

Yes, they royally goofed up.

Yes, this will be the case that likely ends up in law text books as a precedence setting case.

But is that all there’s to it?

Let's dive deeper. 🕵️‍♂️🔍

8/n

What about how ChatGPT was marketed and hyped as an all-knowing that could do almost anything?

Yes, the disclaimers about limitations are there, but are they hyped in the same way its capabilities are hyped? 🙃

9/n

What about the way ChatGPT is designed to output authoritatively and talk confidently to engender (inappropriate) user trust?

Remember, technology doesn’t “just” do things out of nowhere.

Creators of technology design it as such. 🤔🤖

#DesignEthics #UserTrust

10/n

What about the AI hypesters who continue to incessantly overhype AI’s capabilities?

What about the grifters who are feeding off the hypetrain? 🚂

11/n

Do none of these parties have any accountability?

Of course not, they all share responsibility.

Don’t get me wrong—Schwartz is definitely at fault, but increasing the loci of accountability does not lessen his culpability.

So now what?

12/n

We can't afford to look at this issue in isolation.

We need to see this from an ecosystem point of view—at an infrastructural level—so that we don’t just treat the symptom but also the root causes behind it. 🎯

13/n

While I feel at home with AI, law isn't my first language. Special shoutout to @bethwilensky, @KathrynTewson, @questauthority, @VolokhC from #lawtwitter who helped me piece this together.

Follow the docket: courtlistener.com/docket/6310779…

NYT: nytimes.com/2023/05/27/nyr…

14/14 (n=14)