AI-Generated School Board News Story Gets a Scoop But Raises Questions

Remarks about racism by Pascack Valley High School valedictorian Kareena Shah (upper left) were the topic of discussion at the Pascack Valley Regional School District board of education meeting but the conversation was about regulating student speech content, not about addressing the allegations of racism she made.Remarks about racism by Pascack Valley High School valedictorian Kareena Shah (upper left) were the topic of discussion at the Pascack Valley Regional School District board of education meeting but the conversation was about regulating student speech content, not about addressing the allegations of racism she made.

BERGEN COUNTY, NJ (SBN) — A website featuring stories generated exclusively by artificial intelligence has raised eyebrows among New Jersey journalists and educators by highlighting a school board’s discussion of censoring student speeches following a controversial valedictory address.

The story was first noted by Politico New Jersey, which reported on the website, The site is apparently a creation of Mat Hernandez, a school board member in Allendale, NJ, and Matt Post, both of whom are attending Yale Law School.

The site is designed to analyze recordings and transcripts of local school board meetings and summarize them in a news-style article. Major news organizations have been using similar technology for a while for routine stories like sports box score reports and financial earnings reports. This is the first time we’ve seen it used to fill the gap at municipal government meetings where reporters once doggedly covered the activities of local officials.

What’s interesting about this case is that the artificial intelligence apparently made the connection between the speech and the school board meeting, even though the school board meeting discussion was vague about the reason for raising the issue.

The article written by the AI website is called “Valedictorian’s Candid Speech on Racial Discrimination Prompts Censorship Push by School Board,” and you can read it here (

What’s so interesting is that the AI tool apparently connected the content of her speech and the discussion at the board meeting and “realized” there was some glossing over the issue going on at the board meeting, and drew an incredibly accurate inference about the discussion.

In an email interview, Post explained that their AI model “generated summaries for both the graduation ceremony and the school board meeting. So we reran the model to take into account both, which is how it had that additional context.”

Here’s the background.

It seems that Kareena Shah, this year’s valedictorian at Pascack Valley High School, went off script in her remarks, telling the graduation audience that she faced years of racially tinged bullying.

“Whenever someone says my name, Kareena Shah, they say ‘that smart girl who’s going to MIT?” she began. “But little do they know of the many challenges that I have faced to get where I am today.”

Shah, who is of Indian descent, went on to say that “my entire life as a resident of Hillsdale, I have been treated as a pariah. I was forcefully put in a box and confined there as a brown girl who was different, inferior from the rest. Along with the overt forms of racism I faced, I’ve dealt with deliberate microaggressions constantly, from my ethnic food being ridiculed at the age of five to my Indian identity being laughed at in history classes in middle school and high school.”

Shah went on to describe other ethnic slurs leveled at her sister and family over the years.

You can watch her remarks in this player, which is cued up to the valedictory speech, about 43 minutes into the ceremony.

At the school board meeting a month after graduation, you might think the discussion of the speech focused on how to deal with the allegations of racism across a broad swath of the town’s school system.

But video of the board meeting shows the board was more upset at having to hear complaints from other residents about the tone of the speech and was looking for ways to prevent it from happening again.

“I would like the policy committee to look at what the policy is and process for approving speeches,” said board president Joseph Blundo. He asked the committee to consider policies that would “drive the content of speeches” by students and teachers.

“People weren’t thrilled with some of the content of one of the speeches there,” says Blundo at 9:58 of the board video. “I got accosted this week at a public pool for an hour,” Blundo said.

You can watch Blundo’s remarks in this player, cued to 9:58 into the board meeting.

Nowhere in the board discussion is there any outrage directed at the actual allegations of racism. Just the question of how to control what students say in their commencement speeches.

In response to a follow-up email, school superintendent Sarah Bilotti (right) told SBN, “We encourage student expression and student voice. We will not be instituting policies to censor students.”

Bilotti said “Like all school districts, annually Pascack Valley determines a theme and agenda for graduation activities, which includes time limits, number of speakers, and other logistics. The Board was discussing how these decisions would be communicated, not the censorship of speech.”

But that answer doesn’t seem to respond to what the school board president is heard asking the board’s policy committee to do. A follow-up email to Bilotti has not been answered.

Joe Amditis, Associate director of products + events, Center for Cooperative Media at Montclair State University, (left), has expressed concerns about the lack of transparency in the way the AI tool is generating stories on, and the risk that the AI tool may generate stories with factual flaws that could lead to a lawsuit.

“My biggest concern is the lack of human oversight or even basic copyediting,” Amditis said in an email comment. “The chances that the bot will eventually spit out something that creates a serious legal liability are effectively 100%. It’s absolutely going to happen, especially if LocalLens is able to achieve even a fraction of the national scale they say they’re aiming for on the website. All it takes is one slip, one defamatory or damaging statement that makes it onto the site and gets picked up or seen by an affected party and then the whole house of cards comes crumbling down — not to mention the potential harm and damages to the person or people involved.”

Post agrees that there are limitations to the use of AI for news reporting.

“The potential for errors is one of the many reasons AI will never replace journalists and why what we do is not journalism,” Post said. “We think of ourselves as a tool and a starting point. We note on every page of our website that there may be inaccuracies and always link to the meeting record we used to generate the summary. We also link to a correction form at the bottom of every summary. If anyone notices something inaccurate in a post, we’ll fix it immediately.”

While he says the AI tool is good at generating condensed summaries of school board meeting transcripts, it does struggle to spell names right, and also doesn’t do well at Google-style search questions.

“When we first got the idea for this project, we ran hundreds of tests on meetings we either watched all the way through or attended in person in our roles as school board members,” Post said. “We haven’t encountered any of the AI “hallucinations” yet where the computer makes up a completely fake fact or situation. We’re giving the LLM [large language model] a meeting record and asking it to summarize it so it doesn’t need to draw on or come up with any additional information.”

Post says he thinks the common AI errors that are getting attention may not apply to the model.

“I think the classic hallucinations (like the lawyer using ChatGPT-made-up cases in court) come from overestimating the limits of what LLMs can do reliably,” he said. “They’re not good at search engine-like queries. But they’re pretty good at taking a lot of information, like a school board meeting transcript, and compressing it accurately.”

And, says Post, they are learning from the pilot project and hope to improve the results.

Correction 8/23/2023 8:06 a.m: An earlier version of this story incorrectly rendered the URL of the AI-assisted website. It is, not