This transcript was created using speech recognition software. While it has been reviewed by human transcribers, it may contain errors. Please review the episode audio before quoting from this transcript and email transcripts@nytimes.com with any questions.

I had a big technology victory this week. What was that?

Which is that I finally used my chord box. Are you familiar with the chord box?

Like a box where you store a bunch of chords?

Yes. Every man of a certain age has a box in his house with 20 years’ worth of random chords in it.

I do. And it’s sort of like charts the development of chord history.

Exactly. And you just keep this box, like, deep in the closet. You never use it for anything. And then you die, and you pass your chord box down to your heirs. And this is how chord boxes generally work.

That’s right.

But this week, I’ve been doing a technology project at home in my home office, which is I’m setting up something called a KVM switch.

A KVM switch.

You don’t need to know. But it’s basically —

No, I need to know.

OK. It’s, a keyboard video mouse switch.

OK.

And it’s basically a way for you to attach multiple computers to the same set of monitors and peripherals. Anyway, as part of setting this up, I found myself in need of a few cables. And I thought, I’ve got a cable box.

I’m going to go in there. I’m going to look for the cords. So I found them. I found HDMIs. I found display ports. I found adapters. It was like Christmas morning for myself.

These are some of the great cords.

And so I just want to say to all the people with cord boxes out there just collecting dust, keep collecting them. Just keep putting the cords in the box because you never know. Someday, you will need them.

You know, this story really struck a chord with me, Kevin.

Oh!

[MUSIC PLAYING]

I’m Kevin Roose, a tech columnist for “The New York Times.”

I’m Casey Newton from Platformer.

And this is “Hard Fork.”

This week, listeners tell us the wildest ways you’re using AI at work. Then, legendary YouTuber Hank Green stops by to talk about how creators are reacting to the prospect of a TikTok ban. And finally, deepfakes are coming to Main Street. We’ll tell you how one caused turmoil in a Maryland high school.

[MUSIC PLAYING]

So, Casey, about a month ago, we asked our listeners to send us examples, stories, anecdotes from their use of generative AI at work.

That’s right. Kevin, we spent so much time on the show talking about AI, what the companies are doing, what products they are making, how it might change the world. But honestly, it is a lot more interesting most of the time to just talk to real people about what they are actually doing with this stuff.

Totally. So we sent out a sort of call-out, asking people to send in their stories, and we got just an overwhelming number of them. More than 100 responses came in from listeners.

And to give you some sense of perspective, we normally only get that many emails when we make a grammatical mistake.

It’s true. It’s true. Or when I say “um” or “like” too much.

[LAUGHS]:

[LAUGHS]: So we went through them all, looking for sort of themes or commonalities and just interesting stories that stuck out to us. And today, we are going to talk about them. Yeah. So one of the most interesting things about the responses we got is just how wide the range was of things that our listeners reported doing and experimenting with AI at work.

But people also just reported having a lot of feelings about the use of AI at work. Some of them were sort of scared or intimidated by AI. Some of them were sort of delighted by AI and how it helped them do more work or be more productive. And so, today, I think we should just give people a taste of that range.

And it makes sense when you think it, Kevin, because this is a general purpose technology. You go to ChatGPT. It is a blank box. You can put anything in it. And so, of course, people are going to have a really wide range of experiences, but that’s why I think it’s so important to just go in and do that check and try to take the pulse of how people are using this stuff.

Totally. So, for this segment, we’re going to bring you a few short stories that we got from some of our listeners about how they are using AI at work. Some are going to be voice memos that people recorded and sent; some just through the emails they wrote to us. And we’re going to just kind of react to and talk about those. And then at the very end, we’re going to zoom out and talk about some broader takeaways about some of the patterns that are emerging about how people are or aren’t using AI at work.

All right, well, shall we here the first one?

Let’s do it. So first up, we have a story of a listener who figured out a pretty creative way to use AI to get a client to make decisions faster. Let’s play the tape.

[CHUCKLES]: I love this. What did you think of this?

Well, it raises the question, like, aren’t most CEOs kind of a little synthetic to begin with?

It’s true.

The question of what is a CEO doing at any hour of the day, I think, has always been very mysterious. So the idea that you could come in and partially replace them with a chat bot I think makes some intuitive sense to me.

Look, this is a very creative idea. This is a very fun idea. But of course, it also has me wondering whatever feedback that these folks are getting from the chat bot, how closely does it actually mirror what the CEO would have said?

Yeah. So Alec told us over email that he doesn’t actually know whether the client ever told the CEO that they made a synthetic version of her. According to him when they first brought this out, the client just kind of laughed and thought it was a novelty.

But once they actually solicited feedback from the AI CEO with the client watching in real time, he says that the client’s tone shifted. And he also said that they’re now regularly running their work by the synthetic CEO as part of their process with this client. So, to me, I think this speaks to one of the most uncomfortable things about this technology, which is that it is often better at doing the kinds of work that managers and bosses do than the work of individual contributors.

Say more about that.

Well, a lot of what managers, especially at big companies, do is synthesizing. It’s spotting patterns. It’s taking data from across the company and kind of putting it together and projecting forward in some way. It’s like it’s this sort of work of prediction and agglomeration and synthesis that AI is actually quite good at doing.

Now, obviously, there are parts of a CEO’s job that an AI can’t do. A lot of that is sort of leadership and setting the tone and the agenda for an organization. But I think if a lot —

Covering up crimes.

Exactly. But I think if a lot of CEOs and leaders of big companies are honest with themselves, they’d find that actually, maybe this stuff is better at doing our jobs than some of the people who report to us.

Here’s my thought. My suspicion is that this chat bot probably is not doing an amazing job at mimicking the actual CEO of this company that they’re working for. But I wouldn’t be surprised if it was like a pretty good median CEO, and it’s able to give the median CEO answer.

And for folks who are working on this sort of creative work, it is useful to get the median CEO answer, the median CEO feedback. And so for that reason, I say this seems like an interesting tool to put in the toolbox.

Totally. It also just opens up a radical new avenue of workplace conflict, which is that if your boss doesn’t want to talk to you or hear your pitch, and you’re trying to get a meeting with them, they’ll just be like, well, I’m a little busy right now, but you could talk to my clone, and they’ll give you some feedback on your project.

I thought you were going to say that conflict is you go to the real CEO and they hate your idea, and you’re like, well, don’t look at me. The fake version of you love this.

[LAUGHS]: That could happen, too. All right, next example. Our next voice memo comes from listener Jane Endicott. Jane is a freelance writer. She makes short form video scripts for a media outlet that she did not name. And her story stood out for how she has incorporated generative AI into literally every single part of her job.

September 22, of course, right before ChatGPT came out. So, Kevin, what do we make of Jane’s AI workflow?

So I’m impressed by the extent to which Jane has managed to sort of automate certain parts of her job, but I am a little worried because it does sound like she’s kind of automating herself out of a job gradually. It seems like if she can sort of give every part of her process over to a bot, eventually, the people who employ her might just say, well, why are we paying Jane then? Why don’t we just have the bot do the whole thing?

I mean, this is the whole question, right? Because I think if the average manager hears, I’m using AI and it sort of helps me with the first 20 percent or 30 percent of every assignment, people say, oh, yeah, that’s going to make you a lot more productive. That’s good. We can focus your attention on more creative matters.

Once AI is doing 60 percent to 80 percent of the job, I do think that’s where the manager is like, wait? What exactly are you doing over here? So I think there’s just going to be maybe longer than we expect period of arbitrage essentially, where a bunch of really savvy workers who know how to get the most out of AI are going to basically be living on easy street until their bosses catch up. But it’s also possible that bosses are going to catch up faster than I’m thinking.

I will say as a creative worker myself, Kevin, I do bristle a bit at the idea of leaning this hard on these bots, right? If you’re making video, my hope would be that you want to put a real personal stamp on that. You want to put some human ingenuity in that. You just don’t want to rely on the regurgitated writings of every human to come before you. But I don’t know, how would you react to this one?

So, yeah, I agree with that. I think there’s sort of a commodity part of the media industry where you are basically just producing scripts or little short videos out of articles where you are essentially taking things from one format and putting them into another format. I think that is very low hanging fruit for AI.

All right, next up, an email from a listener named Rick Robinson. Should I read this one?

Yeah.

All right. Rick wrote in to tell us that he started using AI to help him navigate difficult situations with colleagues, and here’s what he said. Quote, “I work for one of the country’s largest nonprofits in tech —” OK, brag — “with a recently expanded staff of folks across a range of ages. To get to know one another, we all took a DISC profile assessment.” You ever done DISC profile, Kevin?

No.

DISC, capital D-I-S-C. Rick writes, quote, “It’s like a Myers-Briggs or other personality assessments — Anyway, after getting together to understand each other’s DISC reading, we promptly forgot about it,” which I think is sort of the median outcome for taking any personality tests. OK, back to the quote. “I thought that was a shame, so I built a GPT bot that was trained on a portion of everyone’s results, leaving us with an AI that could answer questions on how to deal with difficult situations that arose with one another. Example, I need to tell Sally and Ned some bad news, but it actually impacts Ned a little worse, I think. What’s my approach? So as to minimize his anxiety and make Sally understand how this could actually be a good thing. Boom, a series of suggested approaches based on their profiles and on how they might react to one another. It’s a skeleton key for lazy bosses.”

What did you make of this?

Wow. Well, number one, I want to learn more about the DISC reading because I’ll take any personality test.

Yeah, you really like that kind of stuff.

Well, I’m trying to get a high score.

This is like astrology for people who went to business school.

It absolutely is. And if listeners want to send us their DISC readings, we’d love to take a look. But look, whatever the personality test was, I do think it is interesting. This idea of I’m going to use AI in a kind of benign way to help me understand my coworkers, to kind of store that somewhere. And if I’m in a difficult situation, lean on the AI to help me a little bit, give me some tools for how I might work through this. Now, I am going to say here, Kevin, I actually don’t think that AI is the most important part of the story.

What do you think the most important part is?

The most important part is that this person, when confronted with a difficult situation, took a beat and stopped and thought about how to respond before he acted. You think about most of the workplace conflicts that arise. People stew and they stew and then they see each other in the office, and it all just kind of boils over, right? And people act from emotion.

And the trick of it is to try to separate yourself from the conflict enough to say, OK, what kind of human am I dealing with? What do I know triggers them? Is there a way that I can approach this where I might get a little bit better result?

I think that’s probably like 85 percent of the reason why this succeeds. Now, at the same time, if you have 15 different coworkers and they all have very different personalities, and you sort of have trouble keeping track of this person’s drama and where’s this person at emotionally right now, keeping some sort of AI system, I don’t know. Maybe it could assist. But what did you think?

Yeah, I like this use case of sort o what you could call sort of conflict simulation. And I use AI for this sometimes, too — not in the same way that our listener Rick did. But basically, if I have to have a hard conversation with someone, or if I’m like — I was trying to help a friend sort of negotiate for a raise a few weeks ago, and I was trying to —

I didn’t get it, by the way, but go on.

You’re your own boss. You can just give yourself a raise any time you want. But I but I was basically — I was being asked for my advice. And I was a little unsure about how this specific situation would unfold. And so I did go to a chat bot and say, like, what advice should I give my friend? And it gave me some pretty good advice that I then relayed to my friend.

So this kind of thing, I think, AI can be very useful for. I also think that there’s an intriguing potential here, and I want to propose a new product, which is slack simulation mode.

[LAUGHS]:

Because how many conflicts at workplaces around the world are basically sparked when someone posts something sort of off topic or controversial in Slack and it totally derails an entire team? I think there should be a simulation mode, where before you derail a Slack conversation, you enter the simulation mode. You type whatever you’re going to type.

And all of your synthetic coworkers respond in real time. And so you can get a sense of how mad you are going to make people by posting the thing. And then you toggle off synthetic mode, and you go back into real Slack and you post your thing.

This is genius. I love this idea. I also just think it would be fun for people who work on really small teams like I do to just have six or seven synthetic coworkers who are dropping funny memes into the chat. I’d really enjoy that.

Yeah, Alan, VP of sales, what do you think about this idea? All right, I, do want to do our DISC readings after the show.

All right, so, next up, we got a bunch of responses from educators — teachers, even school administrators, and principals — who figured out ways to use AI in their jobs. And in particular, for one of the most painful and time consuming tasks involved with being a teacher, which is writing evaluations. So here’s an experience we heard about from one high school teacher.

This is so cool. I love this one. Do you like this one too?

So I do, with one caveat. So, obviously, there’s a lot of work involved in education that is not directly related to teaching students in a classroom. I would put evaluations, college recommendations, things of that nature into this category.

The work never stops for these teachers.

Yes, they have so much busywork and paperwork involved with their jobs. And if AI can help cut down on that, I am all for it. The caveat here is that I hope that James and other educators who are using AI to speed up the process of evaluating student work, of giving feedback to students on their report cards — I hope that they are still thinking as hard about those evaluations as they would have before.

I hope that they are not just kind of taking the stock output of the AI language model and giving it to their students and saying here is the assessment of how you did this semester, because those assessments can be very helpful for students in figuring out what to work on, what they need to improve, giving them positive reinforcement or feedback when they’re doing a good job. I think that loses a lot of its appeal and usefulness to the students if it’s all or mostly being produced by AI. What do you think?

I mean, that makes sense. My hope is that these narratives follow a pretty standardized structure, right? Like, ideally, you’re really trying to coach these kids in a few areas, and the narrative reflects essentially how well they’re doing in those cases.

If this were the case where it’s like, write a lyrical 20-page assessment of these individual humans on whatever rubric you want, then, yeah, you’re right, something would be lost. But let’s just say, if this teacher has — he was spending up to 30 hours a semester on these?

That is almost a full week of work that he was doing in addition to everything else that he had going on. That seems like a really tough requirement to put on these teachers. So if he’s able to get that down to two or five hours and lean on the AI to do work that it seems like can actually be automated decently well, that seems good to me.

Yeah, if it’s just sort of clerical paperwork happening here through the AI, I think that is totally fine, as long as there is actual substantive person-to-person feedback and assessment and guidance taking place in the classroom with these students.

Yeah. Now, I will say a sort of similar story that did make me more concerned was did you see this thing about Texas using computers to grade written answers on their standardized tests this year?

No, what happened?

Well, so they changed some of their standardized tests to involve more written answers. So like, less multiple choice more kind of open ended. And they are using some sort of AI system to do the initial scoring of these answers. They apparently will still send a quarter of responses to humans to be rescored as a kind of quality control, but the AI is still going to be taking a first pass. At and this is one where I say I actually am nervous because it just feels like in two or three years, we will find out that the system was biased against students who are minorities, basically.

Interesting. So your worry is that the AI just isn’t up to the task of evaluating these student responses.

Yeah. Yeah, that it’s going to have certain biases that are programmed into it, and that that gets reflected in the grades, and there aren���t that many humans to review the answers. And it winds up having sort of inequitable outcomes.

Yeah. I mean, I just kind of wonder what the status quo was there because it’s not as if standardized tests with essay responses are being sort of carefully and thoughtfully reviewed by a team of expert educators. They’re basically comparing them against model answers and figuring out does it address this point and this point and this point? Does it have a thesis and supporting evidence and a conclusion?

I think AI can take a first pass at that. But yeah, I would love to see humans kept in the loop on that kind of thing, especially if these are tests that might determine where students are getting into college or whether they’re tracked into sort of gifted programs or not. Like, things that have a real measurable impact on their lives.

All right, well, so far, we’ve heard some sort of fun and cool use cases. But we’d be remiss if we didn’t include a few horror stories, Kevin.

Yeah, let’s see the blooper reel.

All right, our first nightmare AI story comes from a listener named Colin Berry who has a regular work meeting where his writing is basically put to the test against ChatGPT, saying like basically, is this better than what ChatGPT would have come up with? Here’s what Colin said to us over email.

Quote, “I am a freelance marketing writer, and I probably have the worst interaction with ChatGPT that you’ll read. While I do use ChatGPT as a writing assistant — not in actual writing but drafting an ideation — my largest client insists on reviewing my work by giving the same job to ChatGPT and then comparing the two results. Every week without fail, they copy and paste my assignment into ChatGPT in the meeting with me. They aren’t any good at prompts, and I like to think my work is better. But with every new update, it inches closer.”

Here’s what I want to say about this. This is a human rights violation. [LAUGHS]

Whatever is happening in this workplace is specifically prohibited under the Geneva Convention, and this man needs to get a lawyer. I would scream if I were sitting in a meeting and people are looking at my copy and they’re saying, all right, Colin, now let’s just let’s see what the AI can do. And then it’s competing against me. Shut it all down, Colin. You need a new job.

Yeah, you need to fire this client. This is basically the digital white collar equivalent of the John Henry story where you’re racing against the machine. And it is truly dystopian that you would be constantly compared to an AI chat bot that gets better with every new iteration.

Oh, so it’s like what is Colin supposed to take from seeing the AI get better? It’s like I’m sure he is being as creative as he can be, given the terms of the assignment. I think just seeing your work next to an AI isn’t going to suddenly inspire you to become Shakespeare.

Yeah, I think the proper response to this sort of behavior is to write back to whoever is pasting in these ChatGPT answers and say, hey, it looks like you’re finding a lot of utility out of ChatGPT. Why don’t you do that? I’m going to go work with someone who actually wants a human being to do this.

For real.

Get out of here.

Get out — knock it off.

Yeah.

Well, Kevin, if we might do one more, this feeling that you’re going to hear in this call reflected a feeling that came up over and over again, which is that so many workers these days are turning to these AI tools because the demands of their jobs just keep increasing and increasing, and they are not getting any more help, right? So, a classic story of capitalism. And I want to play this one from a listener named Emma Fairchild Barge. Let’s listen.

What do you think about that?

I mean, it makes me really sad. I mean, I have to say I feel like I’ve been lucky in my life that I have not had managers who truly gave me more than I can handle for any extended period of time. But, of course —

Yeah, you work, like, two days a week. It’s like you’re basically on European schedule here.

I mean, I work a lot, but it’s like all on my own terms. I do what I want. I lead a charmed life. But this person’s story, I think reflects the experience of just a huge number of folks who do not have managers who are good, who are kind of working them to the bone.

And while I think it’s great that this person is able to use AI to reduce their workload and maybe gain a little bit of sanity, my fear is that as their bosses catch on, AI actually starts to become an excuse to give them more work, right? It’s like you actually should be more productive because now I know that you’re doing the first 40 percent of every assignment in something like ChatGPT. What do you think?

Yeah, I think the take that our listener has here is the optimistic take, which is that this technology is going to help stressed-out, burned-out workers sort of speed through their excessive workloads and basically come to a more balanced place at work. I think the pessimistic take is that what you’re saying is true — that as soon as this technology becomes truly useful for improving efficiency, managers, bosses, people who run companies are going to say, well, if you had 10 meetings before, now we’re going to give you 20 meetings because you’re going to use AI to get a bunch of stuff done faster. And they’re going to raise their expectations.

And actually, I was thinking about this because I was reading a study recently that was done by Accenture the consulting firm, where they basically surveyed workers and bosses about their feelings about generative AI in the workplace. And one of the surprising things that stuck out to me was that there was a huge disparity in how workers and bosses responded to prompts like, I am concerned that I may increase my stress and burnout. So 60 percent of workers in this survey said that was their concern that AI was going to make them more stressed and burned out. Only 37 percent of bosses felt that way.

So you have a situation now I think where a lot of managers and people who run companies, people who get excited about AI as a productivity enhancer are thinking this is going to make everyone more productive. They’ll be less stressed out. They’ll be less overloaded. And workers, meanwhile, a lot of them, are saying to themselves, wait a minute, that just means that my boss is going to expect more of me, and that’s actually going to increase my stress rather than decreasing it.

Well, I’ll be very curious to see how this one plays out. I mean, I hope that we continue talking to our listeners about those moments at their workplace. Like, is there going to be a moment when you first realize that your boss does actually expect more now that AI chat bot exist? We would love to hear those stories.

Yeah. And it speaks to one of the central themes that kind of emerged out of not just the literature so far of what we know about generative AI at work, but just our listeners’ stories, which is that there are sort of two trust problems with AI at work right now. The first is trust in the technology. There are a lot of reasons that people still don’t trust this technology. It gets things wrong. It makes things up. It causes embarrassing errors in client facing work. But I think there’s also a trust problem between people that is sort of being illuminated and exposed by this technology, which is that workers don’t trust that if they use this stuff, their bosses will consider them valuable and allow them to be more productive without just piling more work on.

Bosses don’t trust workers to not cut corners and automate themselves into a very easy job. And so there’s just kind of this mutual unease and distrust that I think is happening at a lot of workplaces right now, spurred by this technology.

That’s really interesting. What do you think workplaces should do about that?

So one thing I think could be helpful just as a first step is just to bring this discussion out in the open. I think at a lot of companies right now, these discussions are happening in private sort of side channels around the water cooler, maybe at some meetings.

Dissapearing signal chats.

Yes. But I think there’s a real value in just sort of putting this all on the table saying, hey, this technology exists. What is it good for? What is it not good for? How can we use it? And how should our jobs change as a result?

And I think involving not just senior managers but everyone at a company in that discussion is a really important piece. I think right now, a lot of companies are trying to kind of implement these top down rules for how we are going to use AI in this workplace. And I think it should be a more organic bottom up process where everyone from individual contributors all the way up to the CEO feels like they have a voice in that conversation.

Very interesting.

So those were incredible and illuminating stories. Thank you to all of the Forkers who sent in examples of AI at work. We love our listeners and their very thoughtful submissions.

Yeah, those were really great.

[MUSIC PLAYING]

When we come back. Let’s bring a little green energy to the podcast. We’ll talk to legendary YouTuber Hank Green about what it’s like to be a creator in 2024.

[MUSIC CONTINUES]

Casey, I’m very excited for our guest today.

Oh, me too.

This is someone we’ve been wanting to get on the show for a long time. Really, I would consider him a friend of the show even though he’s never been on before.

But he did email in a question one time.

It’s true. So today, we’re talking with Hank Green. Hank is, I would say, a legend of online content creation. I don’t know of really any other way to describe him.

He has been making videos on the internet since 2007 with his brother John. And they have created this whole online empire. They’ve got educational video channels, including SciShow and Crash Course.

He also is one of the creators of VidCon, the big annual convention for online content creators. And I would say he’s been sort of a driving force behind the rise of the creator economy, this world of people trying to make a living by putting stuff on the internet.

That’s right. And I would say in recent years, he has become a sort of elder statesman of the creators, a very smart, reasonable voice on issues affecting creators. And so whenever I want to know how changes in platforms and algorithms are affecting the people who make their living on them, Hank is always the first person I want to hear from.

Totally. So he’s someone we’ve wanted to talk to on the show for a while. And there’s so much happening in the world of online video and media creation right now. We’ve got the looming ban of TikTok, which may happen at some point in the next year.

We’ve got the rise of TikTok competitors — YouTube Shorts, Instagram Reels. We’ve got the larger platform changes and kind of the splintering of social media brought about by kind of the decline of Twitter, now X.

And we’ve also got a YouTube channel that we need to grow, Kevin.

It’s true. So now seems like the perfect time to bring him in to talk about the many changes facing online creators in 2024 and what he thinks the future might hold.

All right, let’s bring in Hank.

When we come back, a wild story about a deepfake scandal that rocked a Maryland high school.

[MUSIC CONTINUES]

All right, Casey, we have one more story we have to talk about today because it has been on my mind all week. And that is the story of what I would call Main Street deepfakes.

Deepfakes, these AI-generated synthetic fakes — are a problem. We’ve known this. We’ve talked about it on the show before. Usually, in the context of big national issues — elections, protests, things of that nature.

Somebody created some synthetic audio of President Biden.

Yes, exactly. So that is a problem, and I think lots of people are rightfully concerned about it. But we’re also seeing a bunch of stories kind of trickle up from local media outlets about things that are happening in much smaller sort of settings, where AI deepfakes are creating major problems.

Yeah, this technology is very easy to use. It is almost freely available to use on the internet, and it is finding its way into more and more hands, and this is a story about that.

So we’ve seen over the past few weeks and months a number of stories like this. There was a story about deepfake nude images being created by and of high school kids that are going around in high schools. There was a similar story out of Spain that caused a big outcry.

And then there was this story that happened last week that my colleague Natasha Singer at “The Times” covered — really wild story about something that happened at a high school in Baltimore County, Maryland.

All right, so walk us through it.

So the story takes place at a high school called Pikesville High School in Baltimore County, Maryland. And in mid-january of this year, there were some mysterious audio clips that started going around this high school. These clips are purportedly of the school’s principal, Eric Eiswert, who is making a bunch of inflammatory, racist, and anti-Semitic comments.

He talks about ungrateful Black kids in the school who can’t, quote, “test their way out of a paper bag.” At one point, the voice says that if he gets one more complaint from a Jewish person in the community, he’s going to, quote, “join the other side.” So, very inflammatory statements.

And the audio clips kind of sound like he was secretly recorded making these statements. And a lot of people in the community are outraged when these clips come out. They think this stuff is real. They’re sort of harassing and threatening the principal and his family.

He was actually temporarily removed as the school’s principal and just triggered a huge wave of backlash, as you would expect if a school principal had been caught on tape saying these kinds of offensive things. But then people started investigating. The principal says this wasn’t me. I never said these things.

The police start getting involved. They send the audio recordings to experts and got access to some emails. And they find that actually this has a much stranger origin story.

Truly.

They find that the athletic director at Pikesville High School, a man named Dazhon Darien, was having a dispute with principal Eisert over some performance issues. The principal had opened an investigation into Darien over the alleged mishandling of school funds. Apparently, he was making improper payments to an assistant girls soccer coach who was also his roommate.

And who apparently was hired but never did the job.

Yes.

Which is a dream job.

Right. So basically, there was a conflict between the athletic director and the principal. And they conclude that the athletic director, Dazhon Darien, made this deepfake audio recording as retaliation for an investigation.

I mean, I have to say this feels like a real gift to the police department. They go in and start snooping around, investigating. Like, this principal, is he having a huge dispute with anyone in the school? It’s like, oh, yeah, well, there’s the athletic director who the principal had recently caught funneling school funds to his roommate. So I’m sure that got the police’s attention in a hurry.

Yes. So the police, after this investigation, they conclude that it was Darien who had produced this recording and sent it to two other teachers. One of those teachers then sent it to a student, and it sort of made its way around not only the school, but also local media outlets. The NAACP was also sent copies of this recording.

So after investigating this whole thing, the Baltimore County Police put out a warrant for Darien’s arrest. He was arrested the next day at BWI Airport after he was stopped for having a gun. And security at the airport saw that he had an open warrant. He was charged with disrupting school activities and some related charges like stalking the principal. And the school superintendent said that officials are recommending that he be fired.

This was the crime they got him on after all? This is disrupting school activities. That seems like small potatoes, given the gravity of this. I will also say, Kevin, you left out one of my favorite details of this story, which is that once the audio started getting around, one of these teachers sent it to a student who, quote, “she knew would rapidly spread the message around various social media outlets and throughout the school.”

So whoever this anonymous gossip was at Pikesville High School, you may have a future in journalism. Although, of course, we will want to help you vet your evidence.

Right. So, very serious story, very sort of disruptive use of AI deepfakes to sort of wreck the life of a principal who it appears did nothing wrong here. And I wanted to talk about this on the show today because I think we are at the precipice of a sea change in the way that AI is used in people’s ordinary lives, and a sea change in how we perceive things that appear shocking or surprising, and the kind of suspicion that will now have to be attached to any kind of evidence, no matter how convincing it is.

Yeah. I think something we’ve talked about before on this show is that, with the synthetic media that typically gets written about, it is of these major political figures. It’s Trump. It’s Biden. And there’s a huge journalistic infrastructure that can go in, can investigate, can say, hey, this was real. This was fake and sort of help everyone adjust their settings in determining what is reality.

That’s going to be so much harder to do at the local level where journalism has been gutted. There are few sources of trusted authority in these communities who can weigh in and make these determinations. And so this Dazhon Darien appears to have been exploiting that and saying, I think I can get away with this because who is actually going to come in and be able to tell that the principal didn’t really say this?

And we should say like the recording itself is very realistic. It was actually quite convincing. So I think we should just listen to it not because I’m excited about spreading these fake recordings around. But I just think it’s useful for people to understand and hear for themselves how good this technology has gotten.

Yeah. So let’s hear a short clip of that audio.

So, Casey, when you hear that — if you did not know anything about this story — would your suspicion be that this is fake?

No, not at all. I mean, I think what gets me, Kevin, is that there is so much emotion in the voice that we just heard. And I think up until now, that has been the telltale sign for me is there’s a certain flatness in the affect of the voice that lets me say, OK, I’m hearing a computer.

That fake voice was angry. Also the rhythm of the words was such that he spoke faster in certain moments and slower in others in a way that also sort of mimics human speech. So for those reasons, I did think that it sounded quite realistic. What did you think?

Yeah, I thought so, too. I mean, this technology has gotten quite good quite fast. Just a couple of years ago, Even. The best synthetic AI voice software on the market still sounded like a robot. I mean, it was flat. It was sort of affectless. It would put the emphasis on the wrong syllables of words sometimes.

I believe it’s pronounced sy-lable.

[CHUCKLES]: And it was quite labor intensive to make a deepfake audio. You needed to feed it hours of someone’s voice samples in order to be able to clone their voice and make it say stuff that they didn’t actually say. Now, some of these technologies allow you to do realistic synthetic voice creation with as little as a minute or two of someone’s actual voice being recorded.

There are now technologies that can insert ums and ahs and these sort of stumbles that people make as part of our natural speech cadence. And in this case, it appears that there was actually background noise that was sort of inserted into this audio file to make it sound more like something that someone had caught on a sort of hidden recorder in their pocket or something.

Yeah, there was a kind of tinniness to the audio that also made it — just kind of give you that impression of this was surreptitiously recorded.

So when the investigators sent this audio to AI experts, they were able to flag some things that they thought made it sound more synthetic. They talked about there were signs of after-the-fact human editing. But I would say that this for me was a good reminder of just how convincing these fakes are and how little money and technical expertise you need to make then.

I guess it raises for me the question, Kevin, of like, as you start to hear these little controversies in your community, oh, here’s this snippet of audio, what are the signs that we’re looking for? How do we know when to react and when not to react?

I think this is really tricky because a few years ago, I would have told you there are a few basic indicators that something is being AI-generated, right? If the voice sounds kind of flat, if it has these sort of, I don’t know, computery effects on it.

You might hear something and think, oh, that’s not real. But I think this technology has gotten better her to the point that most people who are not experts would not be able to tell the difference. So I don’t know. If you have tips for listeners who may be worried that something like this could happen in their own community, what would they be?

And here’s the message that I would give. And this is a tip that is taken from essentially how to avoid falling for clickbait and rage bait online. I think sort of during the heyday of people sharing news on social media, a lot of times, you would see something that would produce this really strong emotional reaction. And you’d share it maybe without even reading the article, certainly without doing any work to tell whether it was true or not.

And in those cases, the advice that really worked for me was, number one, ask yourself two questions when you see a story like this. One is, is this story telling me something I want to hear? Is this about something I’m inclined to believe or want to believe? If so, that’s the first thing that should trigger your skepticism. Before you share, you want to do a little bit more digging.

And then the number two thing to look for is, is this producing a very strong emotional reaction in me? Do I find myself really angry? Am I really upset?

That should be the second sign of, like, OK, maybe somebody is trying to get me to feel this way. If you can notice those two feelings in yourself before you react to a piece of media, you’re going to be in a better position to understand whether somebody might be trying to mislead you.

Yeah, I would totally agree with that. I also think context is very important. If some recordings or video or audio emerge of someone in your community saying outrageous and offensive things, is this someone who has a history of saying outrageous and offensive things? Does it make sense for this person to be sort of recorded saying these things? Or is this kind of a departure from how people have heard them speak publicly and privately before?

But I would say looking to the actual media is sort of a sucker’s game because this technology, as we’ve talked about, is getting so much better and so much cheaper, so much harder to tell apart from real media. I think that, ultimately, we just have to kind of dial up our overall skepticism about the sort of media that we encounter on the internet or in our communities.

Which does have this really unfortunate effect, Kevin, which is that when people do say really horrible, explosive things and then want to deny them, in many cases, they are going to now get the benefit of the doubt. This is a phenomenon which has a great name — the liar’s dividend, right? And the liar’s dividend is now that in a world where deepfakes exist, it is easier for liars to get away with lying. So that I think is really unfortunate.

The other thing that I’m just taking away from this is, man, do we need local journalism in this country. In fact, a lot of what we know about this story came from a great nonprofit called The Baltimore Banner, which is a relatively new organization in Baltimore that’s trying to pick up the banner from “The Baltimore Sun,” a once-great newspaper that is in some pretty rough times right now. And because The Banner existed, we were able to get some good boots on the ground, people who understand this community, people who could talk to people who knew and could give us some information that we can trust. Not every city has a Baltimore Banner. And so when we think about what is the solution to this deepfakes problem, it actually is, in part, a how do we solve a local journalism question?

I agree with that. I also think there are some things that the tech companies could do to prevent or at least to cut back on the abuse of these tools for retaliation or retribution or revenge. Some of these platforms now require you to prove that if you’re trying to clone someone’s voice, that you actually have that person’s permission. And I can imagine a number of ways that you could attempt to do that. But I also think there are limits to that because we now also have kind of more open source voice synthesizing tools that are just not going to take those same precautions.

Yeah, I think that makes sense. But I think there’s even more that the tech companies could do on top of that. I think they could experiment with some of these audio watermarks in the same way that we see watermarks and images. Can they put some tiny artifact inside the audio that can be revealed that it was made by one of these synthesizers? You can also just do something on the level of content moderation. If you’re one of these voice synthesizer studios, and somebody types in a sort of inflammatory paragraph to make a synthetic voice say it, it’s reasonable to me for the synthesizer to say, gosh, this doesn’t really seem like a great use of our technology. And maybe we’ll let you do it. But maybe we’ll actually make you tell us the reason for why you’re synthesizing this so at least we can have some sort of file.

Yeah, I totally agree with that. And I also think the other area that they should be really careful about is financial scams because this is the other big use that we are seeing already for these synthetic voice tools is that people are using them to basically place fake ransom calls to people’s friends and relatives, calling Casey’s mom, pretending to be Casey.

And I’m saying, I’ve been taken hostage, and I need you to wire $100,000 worth of Bitcoin to this address. And that kind of thing is not only plausible, but we’ve seen demonstrated examples of that happening. And so yeah, if you’re going into 11 Labs or one of these other sort of synthetic voice tools, and you are creating synthetic voices that are saying things about money, about sending money, about needing money, I think that should immediately raise a flag in the system, and they should probably not allow you to do that.

By the way I did text my mom a couple months ago about this exact issue. And I said, hey, by the way, if you ever get a call and it sounds like me, and I say I’ve been taken hostage, make sure you call me on the phone because it’s almost certainly a scam. And I was so proud of my mom. She was like, oh, yeah, I already saw a segment about this on the news. I’m way ahead of you on this. So one thing that you generally can count on parents is to be aware of scams on the internet because they love to read about that.

You’re never going to sneak one by Sally Newton.

(LAUGHING) No, you’re not.

Yeah. No, I think this is a really important thing. I think people need to be having this conversation with their family members. There needs to be a secret passphrase or some question that you can agree that you’re going to ask before wiring money or sending money or doing anything involving a credit card number or a Social Security number. You need to have a way of establishing with your loved one that it is actually you on the other end of that call. I think this is a conversation that people need to be having with their friends and family right now.

This Mother’s Day, when you call mom to tell her you love her, maybe just say, hey, what’s our secret passphrase we can always use so that I’ve not been taken hostage? And that’s going to warm mom’s heart.

Casey, what is our secret passphrase if someone ever calls you about me being taken hostage? What should we do to demonstrate that it’s actually the real thing?

I mean, my initial answer was we should just say sandwich time.

I like it.

Only you and I know that we like to eat sandwiches after the show. Now we have given that information away, and we will have to come up with a new passphrase.

OK. We’ll come up with a new one, and we’re not going to put it on the podcast.

We refuse. [MUSIC PLAYING]

“Hard Fork” is produced by Rachel Cohn and Whitney Jones, were edited by Jen Pollan, were fact-checked by Caitlin Love. Today’s show was engineered by Alyssa Moxley. Original music by Elisheba Ittoop, Rowan Niemisto, and Dan Powell. Our audience editor is Nell Gallogly. Video production by Ryan Manning and Dylan Bergersen.

You can check us out on YouTube at youtube.com/hardfork. Special thanks to Paula Szuchman, Pui Wing Tam, Kate LoPresti, and Jeffrey Miranda. You can email us at hardfork@nytimes.com, preferably with a deepfake of Kevin saying something inflammatory.

Please don’t.

[MUSIC PLAYING]