Welcome to The Newsfeed.

I'm Paris Jackson.

A recent review of public records finds some Washington cities and departments are using artificial intelligence to handle varying degrees of government business.

In a new series, Cascade PBS and KNKX partner to dig through thousands of pages of ChatGPT conversation logs from city officials in Everett and Bellingham.

Cascade PBS and KNKX reporter Nate Sanford explains the questions surrounding government efficiency to crossing ethical lines.

What were officials in Everett and Bellingham using artificial intelligence for?

-Right.

So I filed records requests for about a dozen cities in Washington asking for, basically records of every time someone had used ChatGPT on a government computer.

And, you know, Bellingham and Everett were some of the quickest to get back to us.

But, it was, it was really it was a huge volume of records.

And it seems to indicate that it's, the technology is being used frequently, kind of at scale.

A lot of the records showed pretty mundane tasks, like summarizing notes or formatting spreadsheets, but there were also a lot of examples of people using it for some pretty complicated government tasks.

Right?

Like writing applications for, you know, federal funding or, analyzing policy proposals, writing speeches, generating talking points, writing responses to emails from constituents.

The mayors of Everett and Bellingham stress that their staff are focused on reviewing AI generated content for accuracy and bias.

Yeah, so Everett and Bellingham are still putting AI policies in place, but, the guidance has been that staff should review everything created by AI, right?

Because chat bots like ChatGPT are notorious for inventing facts entirely.

From the records we found, it seems like, there were many, many examples of ChatGPT just inventing things entirely while producing important government documents.

It seems like staff are being pretty attentive and noticing those mistakes before they actually get published in places.

But it does kind of raise that question of the fact that this is this is an unreliable, unreliable technology.

What is the pushback from academics and advocates on the use of AI in government?

You know, there's the big accuracy problem that everyone talks about, right?

And that it's just, it's unreliable and it makes stuff up.

There's also a lot of bias that can creep into its outputs, which could create problems if it's used for drafting important government documents.

There's also kind of a problem in that chat bots like ChatGPT tend to offer a lot of praise to the user.

They tend to be very sort of flattering and sycophantic in their responses, which is not necessarily conducive to the type of feedback someone might want on a government policy.

And there is also just lots of kind of broader ethical and moral questions about what happens when this important work is outsourced to a machine that ultimately doesn't have the same level of accountability that a human does.

Can you give us an example from your reporting in an instance where, one of these municipalities actually used ChatGPT to respond to a constituent question?

There was, there was one where, a senior citizen emailed the city because he was worried about not being able to afford utility bills.

And, a city staffer basically copy pasted his email into ChatGPT and asked it to generate a sympathetic response.

There was another instance where someone, they were worried about snowplows.

It was, it was a big, big snowstorm, right?

They didn't feel like the streets were safe.

So the emailed the city to complain about that.

And again, someone from the city basically just copy and pasted their email into ChatGPT, and asked it to generate a response.

Your reporting found ways that AI is eroding, from those examples, public trust.

Yeah.

So I ended up reaching out to that person who had emailed the city about the snowplow complaint.

You know, they had assumed that they were talking to an actual human.

But when they learned it was AI generated, they were, you know, they were pretty upset, right?

Because even if the content itself is factually accurate and not that dissimilar to what would have been said otherwise, a lot of people still find it kind of inauthentic and dismissive.

And so they're saying they, they just wish a human had actually taken the time to respond to them.

I'm Paris Jackson.

Thank you for watching The Newsfeed, your destination for nonprofit Northwest news.

Go to CascadePBS.org for more great local coverage.