How often do opinion polls get it right and zero in on what Americans are actually thinking and feeling when it comes to elections?
We've all seen political polls get it wrong, miss the mark and be an inaccurate barometer of public sentiment.
With an election looming, polling is in overdrive.
Could AI make polling more accurate? That's what a group at Ohio State is working to find out.
Without question, work has changed post-pandemic. Workers and managers have had to navigate many obstacles, including working from home, Zoom, and more recently, the addition of AI.
But just because work looks different, it doesn’t mean the work isn’t working.
Smart lights are new ways of creating unique atmospheres in homes. Some of the lights are now using generative AI.
Dry cat food dispensers have existed for a while now, but what about cats who prefer wet food?
Dishwashers are beginning to include cameras. How do they help get your dishes cleaner?
We're talking about all of these topics on this week's Tech Tuesday.
Guests:
- Ryan Kennedy, political science professor and the Timashev Chair of Data Analytics, Ohio State University
- Gleb Tsipursky, author and CEO of Disaster Avoidance Experts
- Russell Holly, director of commerce content, CNET
Transcript
This transcript is generated with AI. To ensure its accuracy, review the audio file.
Amy Juravich: Welcome to Tech Tuesday from All Sides with Amy Juravich, a show where we share stories about science, technology, and the future of our environment. Polling is a science. We've all seen opinion polls and election polls that get it wrong, miss the mark, and are an inaccurate barometer of public sentiment. With an election looming, polling right now is in overdrive. So could AI make polling better, more accurate? That's what a group at Ohio State is working to find out. Joining us now is Ryan Kennedy, a political science professor and the Timyshev Chair of Data Analytics at Ohio State University. Welcome to All Sides, Ryan.
Ryan Kennedy: Thank you so much. Great to be here.
Juravich: So for two years, you and your colleagues have been working on AI assisted polling. So I guess first, tell me what traditional polling is how it's been done and how AI assisted polling is different. Give me a little compare and contrast here.
Kennedy: Sure, absolutely. Basically, since the 1930s, most people have settled on a kind of closed form interview. Basically, asking a question and giving people multiple choices with which to respond. So how much do you agree or disagree with the following statement? Now, the problem with this is, I'll ask you to imagine something. We're trying to communicate something. You're trying get some information from me about what I'm doing and what I am thinking, right? And I'm trying to convey it back to you and get back. Imagine if you had to do your job using only multiple choice questions. How effective do you think that would be at getting at what your guests are thinking and conveying that information to the people who are listening to your show?
Juravich: I'm very inefficient.
Kennedy: It would probably not work very well at all. Well, the same thing is actually true of public opinion polling. What we're really trying to do in public opinion polling is to get an idea of what the other person is thinking and then be able to analyze that and convey that to the broader public. And the way that we've been doing it frankly doesn't work. But there's a problem, which is that up until now the only alternative we've had would be to hire. Hundreds of people to go out and do a whole bunch of interviews where we would have to pay people quite a bit to do those interviews, and then take months to get the results before we would get any kind of result, an election would be over. AI actually helps us to get around that problem. It allows us to do hundreds, even thousands of interviews and be able to take those interviews and analyze them and get the result within a time frame to where actual people might actually care about those results.
Juravich: Now, is AI doing the interview or is a human doing the interview and AI is analyzing?
Kennedy: So, AI is interviewing a human. Okay. Yes, we do use human respondents. I know there are some people who are trying to say that AI can create their own respondents out of nothing. No, we're not doing that.
Juravich: Okay, so AI is, AI is asking the questions, though. Okay.
Kennedy: He is asking the question
Juravich: And are they asking the questions of a real-life person in person, on a computer, on on a phone? Like what, tell me about this.
Kennedy: So we do it on a computer. So we it all text-based right now. With that said, we're working on making it to where it'll be audio-based, to where we could do it over the phone, to where could do some of these things in person, if that seemed like it would be a good way to go. But for right now, we are primarily doing it where people will be on their computer, interacting with the AI chatbot in order to do the interview.
Juravich: Okay, and is the human typing the answers or speaking them into the computer?
Kennedy: They can choose which one they prefer, which mode they like.
Juravich: Okay, so what kind of topics are you asking about? Can you give me an example of what the chat bot is asking the person? Sure.
Kennedy: So a lot of what we've been doing so far has been analyzing other poll questions to see how effective or ineffective these multiple choice questions are. And what we found is that they're, in many cases, absolutely terrible. So let me give you an example of this. In 2025, shortly after the assassination of Charlie Kirk, YouGov ran a poll that, among other things, ask people. To what extent they agreed or disagreed that violence to achieve political goals may sometimes be justified. Now, if I just ask you that question, it's a very, very heavy question. This result came out, suggested that people were more liberal, were more likely to agree than people were conservative. It got taken by the vice president, got all over the news, tons of news attention. It ends up, so we went through, we asked people that exact same question just a couple of weeks later than YouGov did. And then we asked what they were thinking of when they were answering this question. What kind of scenario were they talking about, everything like that. And what we were surprised by, because we weren't sure what was gonna happen, but what we found out was that the people who were agreeing with this statement, they weren't thinking about the Charlie Kirk assassination or anything like it. In fact, if you think about the question, doesn't ask that. The question asks if violence to achieve political goals could ever be justified. So people were saying, well, yeah, the American Revolution, the French resistance to the Nazis. You know, basically what it was, was the people who agreed, they were answering a completely different question.
Juravich: The questioner had Charlie Kirk in mind, but the people answering maybe had studied history and had something.
Kennedy: That's exactly it. And it was almost down to the letter. The people who were agreeing, they could sometimes be justified, were thinking historically. They were taking the question at face value. And in fact, if we were having a discussion back and forth, I think most all of us, except for maybe the most extreme pacifist could come up with some situation where we'd be like, well, yeah, I would hope I'd be in the French resistance against the Nazis, or I'd hope that I'd.
Juravich: Yeah, if things got real bad, you would hope, yeah, okay. Yes, yeah. That's the thing. Yes, yeah, that's the thing.
Kennedy: And so... So you find that out from...
Juravich: So you find that out from a dialog rather than a black and white question, is that right? Exactly.
Kennedy: So what we've been doing is we've been starting with the questions that the pollsters use. And then we've gone through and actually engaged more deeply with the respondents. And we have found over and over again on issues like gerrymandering, on issues like Christianity and politics, on just issue after issue, that there are fundamental distortions that are taking place where the differences between respondents aren't so much their disagreement about the issues, it's that they're interpreting the questions in an entirely different manner.
Juravich: Are you asking them as as black and white of a question as will you vote for Vivek Ramaswami or will you go for Amy Acton or are you asking these more nuanced questions?
Kennedy: So we've mostly been asking the more nuanced questions. With that said, I think that there is good reason to try to dig in more to the questions like, would you vote for Vivek Ramaswamy? And the reason is because even with the questions that are really clear, where the context, if the election were held tomorrow, would you vote X? Even in those questions, there's a lot of ambiguity.
Juravich: So this would allow the person to answer maybe and then explain why?
Kennedy: Exactly, I'll explain why, or even the people who answer yes may have very different reasons for why they answer yes.
Juravich: It would be great to hear the why, but then it also, like it would truly help you find those undecided voters that everyone talks about all the time.
Kennedy: Absolutely, absolutely. And to get a sense of the maybes who are leaning one way or the other. The old example, this isn't my example, but a previous scholar talked about it. Think about this very simple question we get asked all the time. Do you approve or disapprove of the job that the President of the United States is doing? Right? Seems so very clear. But if think about it, a person who thinks that what the president is doing is bad would answer the same way as a person who likes the president but thinks that they aren't doing enough, right? Both of them might say that they disapprove, but for extremely different reasons. That why is really important to understand if we're going to make any kind of story out of this or any kind sense out of it.
Juravich: This is Tech Tuesday from All Sides on 89.7 NPR News. We're talking about AI political polling and we're talking with Ryan Kennedy, political science professor and chair of data analytics at Ohio State University. So what are some of the flaws in election, like what are the potential flaws in the AI polling though, because everything is not perfect, right? And we hear about AI hallucinating things all the time and we are always told. To check the AI summary because, you know, it could be pulling from a movie that is completely not true. So can we trust this?
Kennedy: This is, we are trying very hard to make sure that it is as trustworthy as possible. And there are several ways that we try to do this. Number one, we've, you know, been working with this AI interviewer for a long time and found it doesn't go off track very often.
Juravich: So like, let me interrupt you. Did you build your own?
Kennedy: Or are you using? We've done both. We used a commercial and built some of our own as well. Continue. Absolutely. So the AI interviewers are pretty robust. Every once in a while, we'll have something that goes a little bit off track, but usually it's not something like an outright hallucination. It's usually the AI trying to be too friendly. And so when somebody asks the AI a question back instead of maintaining the role as interviewer, it turns into a helpful assistant. But I think that's one out of, oh my gosh, I think we're approaching 4,000, 5,000 interviews. We found it. Anyway, the second thing, when we're doing the encoding, the annotation of the answers, we make sure that we utilize multiple LLMs to do the encoding and that we check the encoding using human annotators. I have read personally. A ton of these interviews to make sure that the AI is not going off track, making up quotes, doing everything like that. We even go so far as any quote that the AI pulls out to verify a statement one way or the other, that it matches verbatim what's in the transcript. So it's not making up. So you're spot checking. Well, we're spot-checking, but we're also utilizing, like, multiple layers. Of AI to make sure that everything is okay.
Juravich: Okay. And how are you finding the people, these 4,000 or 5,000 people that you're polling, how do you find them? Because that's a part of the problem with polling was that people were, they were using landlines were way too long. So then, and was landlines went away and then we're trying to use cell phones, but no one answers a cell phone number that they don't know. And so it's hard to find a good sample. So how, where are you finding these people?
Kennedy: It is very hard to find a good sample. And this is where I should make sure to be clear. We don't trust our top line numbers. So I would never go to one of our polls and say, well, 76% versus 24% on this issue. What we're getting into is the meaning behind why people answer a particular way. And the reason why is because, you know, we're able to get... Relatively high quality samples in terms of respondent engagement and them being reasonably like the more general American public. What we haven't set up yet, although this may be a future step, is something to where we can say with the kind of confidence that say Gallup does, that our sample will be within plus or minus 3% of the entire population.
Juravich: So where are you finding the people right now, those 4,000 people you're talking about?
Kennedy: So we usually find it through, there's a company called Cloud Research, another called Prolific, and what these do is they have panels that we are able to draw from in order to get our respondents.
Juravich: Because I get text messages almost every day asking me to take a poll. I promise those aren't ours. Yeah. I was just wondering.
Kennedy: I promise that none of those are ours. All of our people are being paid, but they have all signed up to receive surveys through this system.
Juravich: To end on, like, do you have any results that you're going to release, or is this all still too early?
Kennedy: We have a lot of results that we've already released, for those who are interested we have a sub stack that is free for anybody to subscribe to, it's secondraven.substack.com
Juravich: Why is it called Second Raven?
Kennedy: So second raven comes from Norse mythology. Odin had two ravens who told them what was going on in Midgard. One was thought, the other was memory. And so we say that poles sometimes capture memory. They rarely, if ever capture thoughts. So we're providing the second ravens. That's it.
Juravich: Okay, that's very deep and very professory of you. Okay, so give me an example of something that you've released, that you learned.
Kennedy: Sure. So there was a political poll that came out that they claim showed that majorities in both political parties supported gerrymandering for political advantage. I know, it's weird. It seems to go against a whole bunch of other things that we've seen and even political leadership says. So we went through, they actually used a two question system that's not uncommon in polling, but it's something that is a conversational tactic. And they were interviewing people online. And so what it was was the first question would ask, do you think that you should do gerrymandering in response to the other party doing gerrymandering? The second question was, then do you that your party should do gerry mandering for partisan advantage? The problem is in conversation, you almost can't say that without giving some kind of emphasis or even nodding your head or something like that to indicate to the person that you're saying. A different question with a different meaning, that doesn't come across in text. And so what we found was that very few of the people, it was about like a third of the people who agreed with that second question actually understood what the second question was asking. They thought the second was the same as the first question. Basically, should your party gerrymander in response to the other party gherrymandering? They should have reversed those questions first of all, or something. It's just one of those things where it's a common conversational thing that we don't think much about and I will tell you there's an old saying among pollsters that there's a common misperception that anybody who speaks fluent English and has an ounce of common sense can write a good polling question. We're finding not so much because even polling questions that we look at and think, Oh, this one seems like it's pretty good. I'm not sure that it's going to pose problems. Turns out they end up posing problems. So you're.
Juravich: So you're right now investigating polls that have happened and trying to figure out if they were accurate. That's your mission right now.
Kennedy: That's a large part of it. But going into the election season, we are going to try to establish our own panel that's going to track through the election seasons, asking them questions like, for example, you know, the classic been asked for over a half century question, do you think the country is on the right track or going in the wrong direction? And by doing this, what we're hoping to do is to be able to provide information that is more real-time relevant for elections, but also to see how consistent people are in their stories over time. And anyway, that's part of it too. That's more forward-looking than trying to evaluate other people's questions. Okay.
Juravich: Are you going to ask something as specific as the, um, are you going to vote for Brown or Houston or Ramaswami Acton?
Kennedy: We'd like to, admittedly we're caught in a little bit of a catch-22 of that we don't want, so national polls are cheaper than Ohio polls, and we don t want to pay for an Ohio unless we know that there'll be some interest in it.
Juravich: I'm interested Okay, but I'm not gonna pay you anything
Kennedy: I don't expect to be paid, this is, yeah. I understand.
Juravich: All right, well, maybe we should have you back again right before the election and to get more of a pulse on the midterms.
Kennedy: That would be fantastic, yeah.
Juravich: We've been talking about AI political polling with Ryan Kennedy, political science professor and the Timyshev Chair of Data Analytics at Ohio State University. Thanks for joining us, Ryan.
Kennedy: Thank you so much.
Juravich: And coming up, we're gonna talk about AI in the workplace. That is when Tech Tuesday from All Sides continues on 89.7 NPR News.
You're listening to Tech Tuesday from All Sides with Amy Juravich, a show where we share stories about science, technology and the future of our environment. Work has changed post-pandemic. Workers and managers have had to navigate many obstacles, including work from home, Zoom, and more recently, the addition of AI. But just because work looks different, it doesn't mean work isn't working. We are talking about AI in the workplace with Gleb Tsipursky, author of the new book "The Psychology of AI Adoption at Work: From Resistance to Results". He's also the CEO of Disaster Avoidance Experts. Welcome to All Sides Gleb.
Gleb Tsipursky: Thank you so much for welcoming me. Long time listener, first time guest.
Juravich: Oh, great. Thank you for listening. I'm glad you're here. So many employees fear AI is going to replace them, yet AI adoption for most companies is inevitable in some way, shape, or form. So talk to me about that fear. Why is AI causing so much hesitancy from workers?
Tsipursky: Well, it's not a surprise with all the headlines you get about AI layoffs that it's causing so much hesitancy among workers. There's a recent survey out with Pew Research showing that something like 52% of respondents are more anxious than excited about AI adoption. Only 9% are more excited than anxious and the rest equally excited and anxious. And you can guess which camp leaders of companies fall into and which camp employees fall into. So there's a fundamental empathy gap, emotional gap between leaders and employees. And that's what I explore in my book, "The Psychology of AI Adoption at Work", and how that gap can be overcome. Because it is to the benefit of employees to actually use AI. I can talk about that. But the problem is that leaders don't realize it. And the way they try to adopt AI results in what the recent research from MIT finds that 95% of AI pilots fail to scale. So they try and expand, they don't show return on investment, because leaders are implementing it in a wrong-headed way.
Juravich: Well, obviously, workplaces vary greatly for there's all kinds of different professions. So there's like, for example, like a PR firm versus a software engineering startup. Who is your book? Which workplace is intended for this? Because what happens with AI in a marketing firm is very different than a software company.
Tsipursky: What happens with AI is different, but people's psychology is not different. So that is the key. I'm not really focusing on startups. So it's not really intended for like small startups that have a different psychology than more established companies. And so that's really who the book is focused on. Middle market firms, small businesses, and so on, large companies that are established. Startups have a different mentality, but human psychology is not really different. So on humans, like I said, with a peer research survey, that's been showing consistent results. So, for example, the Pew Research did similar surveys in 2021, 2022, 2023. In 2021 and 2022, about 37-38% of people were more anxious than excited about the AI adoption. And there was a huge jump from in 2022 to 2023 to about 50% of the people were more anxious and excited, of course, with the rise of chat GPT. And that figure has stated approximately consistent after the rise at Chan GPT with 50%, 51%, 52% around that, showing that similar responses. So that human psychology is pretty consistent and it's going to be consistent across the large majority of firms, except small tech-oriented startups.
Juravich: Your book is called "The Psychology of AI Adoption at Work: From Resistance to Results". So obviously you have seen resistance. So what's your number one tip for a company that is looking to bring in more AI? How does a leader explain it in a positive way?
Tsipursky: So I talk about that in chapter six of my book where I talk about the three emotional blockers to AI adoption. And so the three motion blockers, one of them you already named, which is fear and anxiety. So I call that psychographic profile of people for whom that's the most prevalent emotion, the AI alarmists. So the leader first needs to understand AI alarmism and how fear and society for many people is the most probably motion. The second biggest emotional blocker is identity threat. Identity threats. So people feel threatened to their professional identity by a tool that can write great sales emails, can write great marketing emails, and do great coding work. They feel threatened by who they are. They don't feel threatened by previous technology like a CRM, but they feel threatened by a technology that can really replace who they are as a professional. We get so much meaning out of our sense as a professional that people feel frightened. And so that's the second big emotional blocker. And the third big emotional blocker is shame. So people who do adopt AI, many of them are reluctant to talk about it. They call this group the psychographic profile of reluctant adopters. And the reason they feel shame is because they're socially stigmatized by their colleagues. For using AI, for doing things like writing emails and writing articles and doing marketing using AI tools. So the first thing that a leader needs to do, we're not even talking about how to deal with things, I can talk about that, but the first things a leader need to do is understand the emotional blockers to AI adoption. They are different than blockers, the typical technology change.
Juravich: So how can the leader overcome those things? If their employees are anxious, they feel shame for using AI, if everyone doesn't agree with it, in some cases people even think using AI is unethical, but if the leader knows it's gonna be better for the business, how do they communicate that?
Tsipursky: So each of those three groups needs different ways of addressing it. And I talk about that in chapter seven, eight, and nine of my book. So for the AI alarms, there are a couple of things to do. The first thing to do is to make a commitment that if the company gets more productivity from AI usage that the company will not fire anyone due to that. Instead, they will focus on growth. And there's recent research out from Stanford that AI forward companies in the same industry. Compared to companies that are less effective at AI adoption, that they grow their top-line revenue by 9% faster, but they grow the headcount by 6% faster. So they're using less people to do more work, but they're still growing their number of people because they're taking away market share from companies that have less effective with AI adoption. And so they're less productive. So that's one. The second thing that you want to do for AI alarmists is communicate to them that currently... What they're thinking is that by using AI, they're training their own replacement. You need to communicate to them that no, the company will keep around those people who are productive and efficient, and those are the people who are using AI most effectively. So you need to flip the script for them. You need clarify to them that by use AI tools effectively, they will be the ones who will stick around in the company and they'll even be more successful if they want to find a job elsewhere. So that's for the first group. For the people with the identity threat. The companies tend to focus on automating the most economically productive aspects of things, like a very top-down approach. But that's the thing that people care about most. That's what they feel tied to in their identity. So when I start working with companies, we do a survey on, hey, what are the tasks that you hate? What would you like to get off your plate? And what are blockers to the things that you love? For example, where do you have to look for information across a whole bunch of areas to prepare the PowerPoint or the report or the meaningful work that you love. And so focus on automating those things. You need to go slow to go fast. That is key. For the third group, for the people who feel shame. The key here is going to be two things. One thing is leadership model. Leaders need to talk about how they're using AI, to do their own emails, to their own performance evaluation reports, to P&L analysis. That's going really important, so that's one. The second thing is to create a venue where people can share about cool new AI uses and leaders can positively reinforce their sharing. Give them words of praise, make small rewards for coming up with tools that other people can use effectively. So basically, change the social norms by leadership modeling. And by having a venue where leaders reinforce positive use cases.
Juravich: This is Tech Tuesday from All Sides on 89.7 NPR News. We're talking about AI in the workplace with Gleb Sipursky, a behavioral scientist and author of the new book, "The Psychology of AI Adoption at Work: From Resistance to Results". He's also the CEO of Disaster Avoidance Experts. So can you talk a bit about that company I just mentioned, Disaster avoidance experts, because the name is very punchy. According to your website, you want people to stop overpaying for AI and build it themselves. So tell me more, disaster avoidance experts, what are you doing?
Tsipursky: We help people avoid disastrous decisions. That's the focus of our work. We've been doing this for a while. I've been a behavioral scientist, gosh, since 1999 now. So really focusing on helping people make the right decisions as the future of work adapts and changes. And disaster avoidance experts is not only about AI, it's about the future work. And the thing is leaders don't necessarily make the best decisions as the future-of-work approaches because they're finding themselves in new situations. AI is very different than previous technology adoption, as an example. So when you think about, let's say, adopting a CRM, that requires a really top-down approach, where everyone does things in the same way, so the data is entered cleanly into the system. With AI, it's a very different approach. It's not top-down at all. It's very flexible tool. Each person needs to figure out how it applies to their own workflows. So that's why with AI, it's not the right approach to just get a tool that somebody else built for you. You need to have your own people build tools that are going to be very customized to their own workflows. Each role needs to customize those. And you can't have IT build, let's say, so I was doing a training for an insurance company very recently, like right here in Ohio, two insurance companies. And IT doesn't know what the exact details of what claims does. They don't know the exact detail of what underwriting does or what auditing does. The underwriting, claims, auditing, they need to build their own AI tools. And AI is great because once you're motivated to use it, it's natural language programming. You talk to it, you ask it about, how can you help me build my own workflows? It's very exciting, it's very effective. But if people are scared of using it, if they're not motivated to do it, if they think they're training their own replacement, that's the underlying block. So that is the key to understand. It's a really different, it's a mindset shift, much more than just a technology issue.
Juravich: You were also named the office whisperer by the New York Times. So tell me about how your work combines psychology with these management techniques, because I would think that, um, a leader wouldn't think that they needed a psychology degree, but basically you're saying that that helps.
Tsipursky: Oh, it helps very much, yes. So the New Times had a profile of me in the Newtimes magazine named The Office Whisperer, specifically because of how people were thinking about hybrid work and remote work in an incorrect way. They weren't really thinking about hybrid work and the remote work from a psychological perspective. They were thinking about, okay, what do I want to accomplish? Where's productivity the highest? But they weren't thinking about how do people... Actually effectively collaborate, what motivates them, what incentivizes them, and with AI, it's even more so because there's such a big emotional gap. There was a pretty big emotional gap between the return to office debates about what employees wanted and what leaders wanted, and there's an even bigger gap here. Like I said, 9% versus 52%. And that's why pilots fail to scale when leaders roll them out, like 95% of pilots fail to scale, according to the MIT study. When leaders try to roll them out because leaders are excited about them and when they roll them out, employees are not. And so that emotional motivation is fundamental. So leaders really need to understand people's psychology. With AI, just because it's very different, like I said, than CRM or something like that, other technology platforms that you adopt. You can have standard operating procedures where everyone does things in the same way. With AI, people need to be much more motivated. They need to build the tools themselves. If you don't build the tools themselves, you will just be getting expensive off the shelf products that people aren't motivated to use and that they fail to scale. So that is the big problem with AI. And that is reason why companies that really want to succeed in the AI era need to focus on their employee psychology, not the technology, which is much easier to use than previous technologies.
Juravich: Well, I did a show on return to work, and we talked about how some employers just used it as a way to have people leave or quit their jobs because they didn't want to return to work. So I guess I could see the same thing with AI. If they need fewer employees for the bottom line, they'll implement an AI thing, and if people don't like it, they can leave and go get another job. Is that a bad strategy? Does the leader need to worry about the company's bottom line or cleaning house? Like talk to me about that.
Tsipursky: It's a terrible strategy. With return to office, people who are forced to return to the office, they will still do their work in the same way. Now, they might be not happy to return to the offices there. We see that definitely their employment satisfaction is down, they're definitely more likely to find a job elsewhere, but they're doing their work in the way when they're at home and then when they are in the office although they might well be less productive in the offices because they're distracted, which is a different issue. Now, when you're talking about AI, you need to get people to do things in a different way. You need to get people that change their ways. And the problem is with Hai is that when they're not motivated to do it, they're not going to change their way. So then they're going to do things differently. You can't really monitor this. With an ESSA with a CRM, you can monitor how people are using. You can see, okay, are people putting in the data in the right way so that a client's information that used to be in an Excel spreadsheet is now there in the same way in the write boxes in the CRM. That's monitorable. But with a AI tool, that's not monitorable, IT again, doesn't really know what's going on with underwriting claims, auditing, finance, they can't monitor this stuff. I've seen this in so many companies, it just can't moderate. You need to get people motivated. And otherwise you get what's called AI works long. This was a recent study from Stanford and better of our business review showed that. Something like 15% of all work products that colleagues are receiving from other colleagues are workslot by AI workslots, by which they mean products that are created by AI that are not high quality enough to actually be productive for their outcomes. So they spend about two hours per AI work slope incident fixing it.
So that's pretty significant amount of time spent on fixing things that result from malicious compliance. So malicious compliance is where people are using AI, but they're using it badly. So if you think about someone who's an AI alarmist, or someone who feels that identity front, a pragmatic resistor, when they are forced to use AI, when their boss tells them, You must use AI for this purpose. They don't use it well, they don't use it effectively. They're not motivated to do so. And when the boss comes to them and yells at them, like, OK, your colleague said you sent them a piece of AI work-slot. They say, well, you told me to use AI. I use AI like you told them. I won't use AI again. And what's the supervisor going to say? You use AI badly? Well, maybe they weren't sufficiently trained and so on. So this is not an issue of people being fired in the large majority of companies. We see that AI forward companies are actually hiring people at a higher rate. It's that when leaders are trying to use AI and they're using it badly, and they try to scale it up, that's what we see 95% of AI pilots failing, not people being fired.
Juravich: Okay. And so just to end on, are you seeing that the idea that AI is going to steal my job, is that a myth at this point? Or is it just, because a lot of companies leaned into AI really fast and let go too many employees and then had to hire them back because it wasn't working out. So is AI the future of work and not stealing my job?
Tsipursky: Hmm So what we're already seeing is in terms of actual research, so we see headlines about some companies that laid people off and so on, and we don't really know if it's about AI or if they over-hired during the pandemic and now they're using AI as an excuse to fire some people. So I think the headlines are not really informative. What's informative is the underlying facts from the research, from peer-reviewed studies. So there's a recent study from Stanford called Canary's Micromine, which showed led in June 2025. In the most AI exposed industries, compared to the least AI exposed industries, young people, 22 to 25, had 15% worse job prospects. They updated that data for July 2026, so about a year later, they found that the gap rose to 19%. So there's definitely something there clearly for young professionals, for people who are in the most exposed industries. So we definitely are seeing the beginning of the impact of AI on jobs. And there was a letter about a month ago from about 200 plus economists, including 16 Nobel laureates, saying that in the next several years we'll see severe AI job disruptions. So I definitely think we will see severe job disruptions. But according to the research I sent from Stanford, the job disruptations will not necessarily be in the most AI-forward companies. Because those are the companies that are growing faster, their headcount, and they're growing their revenue even faster. I think where the job layoffs will happen is the companies are losing market share because they're not sufficiently AI-forward. So those are companies where we'll be seeing job losses. So, if your company is adopting AI, most likely it will be growing faster, and most likely you will retain your job, at least if you're adopting AI effectively yourself. So, this is a subtle approach where you really want to think about which companies will be losing jobs and who will be laid off.
Juravich: We've been talking about AI in the workplace with Gleb Zburski, a behavioral scientist and author of the new book "AI Adoption at Work: From Resistance to Results". He's also the CEO of the consulting group Disaster Avoidance Experts. Thank you so much for your time today, Gleb.
Tsipursky: Thank you so much for inviting me.
Juravich: And coming up, we're gonna talk about smart lights, cat food, and cameras in dishwashers. That's when Tech Tuesday continues from All Sides on 89.7 NPR News.
You're listening to Tech Tuesday from All Sides with Amy Juravich, a show where we share stories about science, technology and the future of our environment. Smart lights are new ways of creating unique atmospheres in homes, and some are using generative AI. Dry cat food dispensers have existed for a while now, but what about cats who prefer wet food? And dishwashers are beginning to include cameras. So, does that help get your dishes cleaner? We're gonna talk about these topics with Russell Holly, Director of Commerce Content at CNET. Welcome back, Russell.
Russell Holly: Thank you.
Juravich: All right, so these new, there's these smart lights and there are new sensors that will make the next generation of smart lights better. And the company is creating new smart light technology. So what makes this product so special and how do you say its name? Cause again, I don't know how to say its names.
Holly: The name of the company we're talking about today is Acara and Acara's lighting, like a lot of the other competitors, has spent the last couple of years doing this kind of race to the bottom to see how inexpensive smart lighting can be in the house, but not really improving what those lights are capable of. And it seems like there's been kind of a shift in how that works, especially over this current generation from the two main competitors in this space right now for the kind of budget focused lighting is Acora and Govi. Govee is most known for lighting that goes behind a television and kind of syncs up with what's going on on your screen, whereas Akara takes more of a whole-home approach. So they've got not just lighting that can go throughout your house that allow you to set whatever color temperature or color you want, but they've also started focusing on permanent outdoor lights. So instead of having lights that you just put up for the holidays, you can just your outdoor lights to holiday colors as an example.
Juravich: So how is AI being used and why are these smarter lights?
Holly: So there's two primary components to how Akara is using AI for this. The first is what they're calling scene generation, which essentially means you can provide a prompt or a feeling or something like that. And Akara will try and create a light scheme throughout your house to go into that. So if you tell the Akara app that you want something that is kind of a late night vibe to your house, then it may change the lighting in your house to maybe blues and purples. Or something that is going to be less kind of blue light responsive to your eyes if that's the kind of thing you're worried about. And it'll take that and make it very personal, so the request that you will make will be very different from the request I would make as far as the lighting and the colors that go through the house. The other part of this is the way that it's going to operate with older versions of its own technology. Akara is releasing a box that will allow you to connect all of the other smart lights that it has made in the past So that it the stuff you'already bought doesn't get left behind technologically
Juravich: Hmm. Okay. So the idea of being able to tell it, you know, what mood you want, or if you're having a party, you know, maybe you could tell it I'm having a beach theme party, make the lighting work, that kind of thing. I is that something that a lot of people want? I mean, I had never really thought about it. We talked about the other lighting company putting the lights behind the TV. And you kind of sold me on that. I didn't buy them. But you sold me on it, you know. So who is this for?
Holly: There are a couple different approaches. There are some who take their bedroom in particular and set it up to smart lighting so that the color temperature of the light matches sleep schedules. So there's a lot of color theory that goes into your lights should be warmer in color the closer you are to going to bed so that you're more likely to fall asleep a little faster, the opposite being true for the morning. So being able to set up schedules like that or something that is important to some folks. The other is just. Having cool color lights if you have friends over and are doing it just because you can. My personal favorite when smart lights release, like we're just becoming a thing, is making it so that the lights in all of my kids' rooms turned red if they weren't downstairs by a certain point in order to go to school. But I think that that's just a bad dad joke at this point.
Juravich: That's funny. Okay, and I also see the practicality of like installing nice looking lights and then changing them, you know, like orange for Halloween, holiday lights. You could do that kind of thing. I mean, that kind of makes sense rather than getting a ladder out and going on your roof once a year.
Holly: And you see, you know, they've started to show up more in my neighborhood and, you know, there are folks who will take outdoor lights and set them to their sports team color when, you know, their their team is currently playing or, you know, for 4th of July, you know, when you when it becomes very easy to do something like that, then you're just more likely to it.
Juravich: Until it annoys your neighbors, then we'll have a whole other issue, right? All right. Well, moving on to our next topic, dry cat food dispensers have existed for a while now, but what about a cat who prefers wet food? Tell me how this new dispenser is going to help pet owners, whether or not they want wet food or dry food. Um, is it called Rica? Rika?
Holly: Yeah. So this is called Rico and Rico is a wet food maker. So you're not taking wet food and putting it in directly and just having it dispense it. You have a water cylinder almost like a coffee maker and then your preferred dry food in a different canister. And when you set up a schedule for when your pet would normally eat and it actually brews the water and the dog and the cat food together. And then dispenses it at that same time as a complete meal.
Juravich: Yeah. Okay. So I actually wrote down, it looks like a coffee maker. So you and I agree. So describe it for us. It's actually making your cat's food on a schedule then, right?
Holly: That's right. You have the ability to set up, uh, you know, your, your own schedule for when it is that your pet eats. There's a, there's a small bowl built into the bottom and a great big water chamber in the back. Uh, it doesn't take up a ton of space. If you're, you're going to have it on a floor, um, you, but, but it is something that is also going to require regular maintenance. So if you're not a fan of crawling around on your floor, uh especially, uh, if you have a cat that is a little more on the curious side and we'll try and access the food when you are not watching. I imagine that that will be an interesting adventure at some point in the future.
Juravich: Is it foolproof? I mean, will it make the food whenever you want? It's not going to overfeed or underfeed your cat if you're out of town.
Holly: Yeah, you have the ability to, you know, determine exactly how much food goes into each mix. It's not something that it does automatically. You say I have a cat that is this size, I, you know, I want the amount of food to be this many ounces. And so you can build that up. And it's set on, you know, a weekly schedule. So you can say, you know, every day of the week at five o'clock, you know, turn on and dispense food, and it'll go through that brewing process, you know, regardless of whether you're there.
Juravich: Was this invented by someone who had a cat who had a dry food dispenser but the cat liked wet food and they were like there has to be a better way and then there wasn't and they invented one.
Holly: That that is a significant part of it. You know, dry food especially has been a thing that smart dispensers have had for a really long time. We even have pet security cameras now that launch, you know, dried food as a way to calm a pet if you're not at home. So there's clearly quite a bit that, you know that goes into that. But we live in a world where folks are a lot more conscious about the kind of food they give their pets. And you know because of that, either wet food or. You know, some of the dry age stuff or the like fan-cooled pet food has become a lot more popular and a lot of that stuff is not shelf stable. It either lives in a refrigerator or it's something that you make first and then feed to the pet. And so in this case, it is removing that step for you.
Juravich: Yeah, do you know how much it costs? It seems fancy, so I don't know. Is it expensive? Yes.
Holly: Yeah, so this is gonna start this is going to be $200.
Juravich: That's not too bad, I guess, yeah.
Holly: Yeah, I mean, it depends on, you know, the, the thing that they're not talking a whole lot about right now that I would love some clarity on is how well it's going to work with a bunch of different kinds of pet food, because I know that people are quite particular about the things that they feed their pets, understandably. And so if it's just one or two brands that will play nicely in this thing, then that may limit how many people would actually want one of these.
Juravich: This is Tech Tuesday from All Sides on 89.7 NPR News. We're talking about street lights, pet feeders, and now we're gonna talk about dishwashers with Russell Holly, director of commerce content at CNET. All right, so there's a new dishwasher that now includes a camera. Why do I need to see my dishes when they're washing, Russell?
Holly: I think there's a couple different reasons that you would want to see your dishes. I think if you are on the other side of the house and your kid tells you that they totally loaded the dishwasher, you can go and check without leaving the thing that you're doing. I think that understanding whether or not something is actually getting cleaned while it's washing, if you're the kind of person who maybe doesn't load their dishwasher in the most conventional sense, then being able to verify that things are actually getting clean so that you can go and stop something if necessary. I don't know that having a camera inside of your dishwasher is a need, but I am interested.
Juravich: Interesting. Yeah. Well, there's a camera inside a refrigerator to tell you whether you're out of milk or not. So what? Right? So this is the next thing. We're putting it inside the dishwasher.
Holly: Yeah, I guess, you know, the next step there would be, you know, if the if the camera was able to, you know, quote unquote, see the soil level of your, you know, plates and pots and stuff, and decide that the wash cycle should be longer because things aren't getting clean, you know, or shorter because they already are that that may end up being a case for water conservation.
Juravich: Yeah, I guess that would save money on electricity and water if it could see if the dishes were clean enough. But I have to imagine that this dishwasher is expensive if it has a camera in it, is it pricey?
Holly: It is. These are made by a company called Melee. It is called the Melee G8, and this is something that has been not as available in the U.S. This is something that is more internationally available, and so U. S. Prices don't exist yet. Oh, okay. In Europe, they're going to start at what is the U S. Equivalent of around $2,000.
Juravich: Okay, well, I guess I also read in the review of it that it can see if the dishwasher's full and start itself automatically, or like pick the right cycle. I mean, it's smart enough to know what you want.
Holly: It claims to be smart enough to know what we want. As you pointed out earlier with cameras that we have in refrigerators, they're not always the most foolproof things. I'm really curious to see how effective that ends up being in practice.
Juravich: Well, we have been talking about cameras and dishwashers. And before that, we talked about smart lights with AI. And we talked about pet feeders that can now do wet cat food, not just dry. And we've been talking about it all with Russell Holly, director of commerce content at CNET. Thanks so much for joining us today, Russell.
Holly: Thank you for having me.
Juravich: And I want to say thanks to the AllSides staff, producers Marcus Charleston and Erin Esmont Rabinowitz, also our student producer Brianna Fortunet and our student producer from Denison University, Ellen Hansen. Video production by the Ohio Channel and board operation by Chris Johnston and Trey Hayes. You're listening to All Sides on 89.7 NPR News. I'm Amy Juravich. Thanks for joining us.