On today's episode of WNYC's new popup advice show, tech writer Manoush Zomorodi, host of the NPR's Ted Radio Hour and author of Body Electric: The Hidden Health Costs of the Digital Age and New Science to Reclaim Your Well-Being leads a conversation about artificial intelligence and our jobs, with Paul Ford, technology writer and founder of the AI startup Aboard. Listeners call in with their experiences and questions about navigating the impulse toward productivity that AI seems to offer, against the more unsettling aspects of this new technology.
Note: You are listening to the podcast version of a live radio show. The call-in number only works between 1 and 2 PM Eastern, while the show is live.
If you'd like to submit advice questions to be featured on the show, or topic suggestions for future shows, please use the form on our showpage at wnyc.org/advice
[music]
Manoush Zomorodi: Welcome to Please Advise, the pop-up advice show from WNYC. I'm Manoush Zomorodi, host of NPR's TED Radio Hour, author of the book Body Electric, and this week, your host on Please Advise. Thank you so much for being here. All week, we have been giving and getting advice on how to navigate all kinds of tech, from screens and social media to innovations in health like GLP-1s and wearables that can track our every breath.
Today we're getting into a topic that makes pretty much everyone nervous. AI and work. What is artificial intelligence doing to our jobs and our livelihoods? We want to hear what is happening with AI in your particular profession. What's working, what isn't? Where would you like some help? Call or text us at 212-4433-WNYC. That's 212-433-9692. Maybe you're concerned that AI is going to eliminate your job altogether. Maybe you're worried that your younger or older colleagues know how to use these tools in ways you don't. We want to help you work through the challenging questions around AI, the workplace, and your professional life.
We're taking your questions at 212-433-9692. We also want your advice. Are the ethics of AI dividing you and your coworkers? Is it causing conflict between management and employees? How are you resolving that tension? Tell us. 212-433-9692. Not to worry, we also have help here. Paul Ford is in the studio with me. You may have read Paul's op-eds in The New York Times. He's a software developer and a writer who, for the last several decades, has really helped me understand how tech is shaping my life. Brooke Gladstone, I think, refers to you as the poet of the internet. So good, because Paul understands humans and machines. Paul, I'm so glad you're here.
Paul Ford: Manoush, thank you for having me. I don't know if I understand humans or machines, but I'm trying.
Manoush Zomorodi: Okay. You, at least, are able to explain your thought process as you try to figure them out. Listeners, you should call in. This is great. Paul Ford's amazing. Tell us your questions at the intersection of AI and work or if you have some advice that you'd like to share with Paul, you can call or text us at 212-433-WNYC. That's 212-433-9692. Paul, we already got one note from someone who has been a designer for 30 years. I hate to start with a downer, but here we go.
They write, "The last year of AI advances has caused major depression for me. I am watching my life, my livelihood, and my love of work disappear. People say this is like when computers and the internet came to be, but it is entirely different. I have tried to use AI in my work for production-based tasks, but I am seeing my clients adopting AI to take over literally everything I do." I'm guessing, Paul, that you have heard this from many people.
Paul Ford: It is a very common story. I'm assuming, graphic designer, maybe somebody who works on websites or on things like that. This is a weird moment. I feel that somebody like that-- My instinct, I can't tell you about the future. My instinct is they are just caught in the crossfire of this moment because here's what's happened. These big companies come out with these cool tools that let everybody pretend to be a designer. They come up with things that look a lot like design.
They're maybe not thought through, they're maybe all kind of beige, but they look like websites are supposed to look. The people get really excited, and they're like, "I solved it. I don't have to pay thousands of dollars anymore." Meanwhile, the person who wrote in, they're caught in that crossfire. I think longer term, everybody's going to get pretty bored with generic output. That's just how it works. That's why not every website is on Squarespace or not everything is built the same way. For now, the artifacts look so real that people just assume that they are.
I'll just tell one little tiny story. A lot of times at work we consult, and we use tools like this. People will bring us these really big documents they created that look like formal documents. They're decks, they're product requirements documents, or things like that. It's the clients, they've built them. They're so proud. They think they've done the real work, but they're not useful. If they'd given us the prompt, we could have helped them.
Manoush Zomorodi: When you say they're not useful, do you mean like they're slop?
Paul Ford: They're not even slop. They're thought through, but it's just humans. When we see something that looks like the thing, we always undervalue the expertise of the person to our left, right?
Manoush Zomorodi: Gosh, yes.
Paul Ford: Like, look at you. You come on this radio show, and you do your little talking.
Manoush Zomorodi: Yes, a little talking.
Paul Ford: That's it. You just got up this morning, and you walked in, and that's all it takes, right?
Manoush Zomorodi: Well, sort of. [laughs]
Paul Ford: That's what. That's what most people think. Most people are like, "Oh, I could be on the radio." They don't see all the late nights where you're-
Manoush Zomorodi: Oh, so many late nights.
Paul Ford: -looking at just cutting and pasting into Google Docs. There's a craft, but it's very, very hard for us to see craft. This person, I have to say, many such stories, it's a bummer. There is a case I think to be made for holding on. That depression's really common. You're not alone. Start talking to other people.
Manoush Zomorodi: Okay. Let's talk to other people. Leah from Charleston, West Virginia. You are an attorney, Leah?
Leah: That's correct, yes.
Manoush Zomorodi: What are you seeing in the legal field?
Leah: I have an especially interesting, I hope, perspective on it because my daughter is in law school right now. I graduated in 1998. I was trained in the old-school way of legal research, and she is in the AI world of legal research. We got to compare those extensively over the summer while she was a judicial clerk. I got to take a legal research for a spin, and my brain was thoroughly blown. I could see after being somewhat of a Luddite, it's absolutely mind-boggling what you can come up with with just a well-crafted search.
The disadvantage in it that I saw was, and especially in writing from that, because it'll generate a ridiculously beautiful product. When she did that, she hadn't integrated it into her knowledge of the law and how the law works. She didn't learn things that would then enable her to see unique, previously unthought-of angles into a legal issue that you get when you're pawing through indexes and pawing through case law and searching and searching on your own.
You build that. You build those neuronal connections between these different areas of law and these different approaches that you don't if you're just generating it with even a really good query. I think the ability to internalize and really make it part of your knowledge, and then how you can apply that to your clients' situations, because every legal situation throws up a novel question.
Manoush Zomorodi: I think that's such a great point, Leah. Thank you so much. Paul, I think we see that in other places too. How do you become good at your job? You do it badly at first when you're young, and you do it again and again and again. You develop judgment, and you know when something is good or mediocre. Is that basically what Leah's saying is that we run the risk of having a young group of people who don't get the reps in?
Paul Ford: We absolutely do. Look, the easiest way that I've found to express this is that the artifacts of craft are very different than the craft itself. You have to put the work in, and you have to what Leah described as those neuronal pathways. That's what we need in the world. We need people who can see the system and operate. Otherwise, you just have a robot making a document and a robot reading a document. That's just not how it's going to go.
We can't really have a society built on that. You can have a system that does that. No, it's very real. Getting people to slow down and internalize is really hard because everyone's getting a quick hit, and you get the thing, and it looks like your homework is done. That feeling is so good. Everybody loves it when their homework's done.
Manoush Zomorodi: Oh, yes.
Paul Ford: It's just wonderful, but it's not actually done because you didn't do it.
Manoush Zomorodi: It might be fine, but probably a D plus.
Paul Ford: Yes, or it could even be an A plus, but you didn't do it. What's the point? You learned how to write a couple sentences.
Manoush Zomorodi: We want to hear your questions, listeners, about using AI at work. Maybe you're a small business owner who's curious about where to start because maybe we're going to talk about this. It's not all bad. Or maybe you've noticed people using AI at your office and you want to have more open and honest conversations about it. You can call or text us at 212-433-WNYC. That is 212-433-9692. We've touched on the part where it does feel like magic.
You are able to use these LLMs, and they spit out seemingly smooth and professional-looking reports or diagrams, et cetera. If you haven't put in the intellectual work into crafting it, you may not really even understand the concepts. What about the good stuff, Paul? Because I know you are actually a fan of some of the tools that are out there.
Paul Ford: Oh no, absolutely. First of all, that legal research is meaningful if you've done that for years and years. It's important to note that that legal research, though, is probably based on a system like Westlaw, and it's based on the actual documents. They're probably using a system at that law school for lawyers. It's not just making stuff up. It's going and gathering stuff. It's gathering cases, and then it's summarizing them. It's really good at summarizing. It can take lots and lots of text and make it into less text, and that is super powerful. We do a lot of that when we're doing any kind of work.
The other thing I love it for is coding. Really, last November, a new version of Claude came out that could code, and it changed the world. Suddenly it really got good. It could write software and build websites. This is where our friend, the designer, got hit. If you're trying to make things or explore new ideas, it's really wonderful because software used to take forever. Now you can show it to people and be like, "Is this what you want?" The tricky part is there's lots of slopware. App Store submissions are up 30% for Apple, but downloads are only up 3%. There's just a lot of stuff that nobody cares about because it's cheap and easy to build. If you use it with discretion and you actually work on the craft and put a little friction in the process, you can get stuff really quickly.
Manoush Zomorodi: Like what? What's the good stuff that you're getting out of it?
Paul Ford: Oh, some of it's really boring stuff, but it used to be very expensive, like taking an old system and updating it, moving data from one place to another. This is sort of bread and butter of technology work, and it sucks. Nobody's ever liked it, and many projects fail. Our content management system from 15 years ago needs to be made modern, and then 18 months later, nothing has happened. That was the common case in software, literally, for decades. Now it's increasingly not because you say, "Hey, robot, I want you to take this one, and I want you to make it good like this one." It's like, "Cool. Here are the steps we got to follow."
Manoush Zomorodi: I want to go to Josh, who's calling in from Cobble Hill. This is a question about getting legal expertise, but from a different perspective. Is that right, Josh?
Josh: Yes. I was just listening. First of all, thank you for taking my question. Listening to the comment about punching a couple things into AI, not having done the thinking yourself, therefore less valuable. One of the ways that I find it very valuable is doing a lot of research ahead of a call so that can show up better informed on the call and, in the case of lawyers in particular, use those billable hours in a more productive ways. Just going to get your perspective on that.
Manoush Zomorodi: What do you think, Paul?
Paul Ford: That's back to the, like, it's so good at summarizing. I use it before I go into a pitch meeting or I'm meeting with a client. It gives me a briefing not just on the client, but also on their whole industry and terms that they might use that I might not know, and the software that they use, and so on. That is so valuable. That's something that a big consulting firm used to pay five juniors to do. It's not that I'm taking that job away, I just could never have it before.
Now I get a PDF, and then I have it draw little pictures to explain what's going on because I'm a simple-minded person when I'm in a meeting, and I just want to have that background. It's a jewel that way. I think what Josh is describing is great. It empowers the individual to make better connections and actually learn more about the people in the room, as opposed to them throwing it across the table and going, "I did my job."
Manoush Zomorodi: Josh, I have to admit that I too have used an LLM so that I didn't have to spend three hours on the phone with my lawyer who costs $500 an hour, but maybe two hours instead. She's got plenty of clients, so I don't feel that guilty. There is this sense of uncertainty and shame around AI use, Paul, like not wanting to admit when you're using it, being uncertain if you are disrespecting a human when you do. How do you navigate that?
Paul Ford: With some guilt-
Manoush Zomorodi: As well.
Paul Ford: -it's very extractive. There's data centers nobody wants going up. The water's getting heated, and climate change is still here, and some of the worst dudes on earth are in charge of these things.
Manoush Zomorodi: Oh, boy.
Paul Ford: It's out of the box. I don't think you can understand the change unless you participate in the change. I've always felt that way. I need to get my hands on it. I think there's a little bit of a cop out. I noticed that all my slightly more progressive nerd friends like me, they're like, "I'm just learning it because I feel that I can then guide and help other people not to."
Manoush Zomorodi: That's what I think too. Are we misguided?
Paul Ford: This is hypocrisy.
Manoush Zomorodi: Oh, no.
Paul Ford: Yes, because the minute you get that summary and you're like, "Oh, it's a lot easier than it used to be," now you're in. We have to figure out-- One of the things that I think about is I think this is a very temporary state. I think for the most part, people like to subscribe to things on the internet and use them, but a lot of models are going to be able to run on your local computer in a couple years. You're going to be able to do a lot of this stuff without working with enormous technology companies.
You're going to be able to do stuff like working-- I use Chinese models for stuff that is completely transparent and open, never client-touching. They're cheap, and they work fine. It's just a little less headed to Silicon Valley. I think it's not going to be as monolithic. I think everybody's got to get through their IPO and make their trillions and then they can go wreck the world in different ways. Then we'll figure out what we're going to do after that.
Manoush Zomorodi: Oh my gosh. We're getting so many interesting comments and questions. We've got one, a text that says, "I am a programmer developer, but the non-coders in my office, including my own managers, are starting to generate code using AI, and I am being made to deploy it to our website, making me troubleshoot it and be responsible for it going forward. I find this extremely distressing to be on the hook for code that I didn't write or generate. How can I push back or otherwise protect myself?"
Paul Ford: Boy, this is real. I've seen CEOs write whole systems. It's the same problem where they haven't filled in all the blanks, but it looks like what they want, and then everyone's on the hook to deliver it. If it's your boss, it's really hard because you're going to have to tell them no.
Manoush Zomorodi: [laughs] Wait, you're going to have to tell them no?
Paul Ford: Look, here's the thing. It depends what you're deploying. If it's something that's read-only, doesn't have a lot of passwords in it, whatever, give it a go, see how it goes. If it's real infrastructural code, you owe it to be like, "Look, this is a--" What I always love to do is I call it the run-to-the-whiteboard move where you're like, "This is the greatest thing I've ever seen." You run to the whiteboard, and you scribble a couple things. Then you smack yourself in the head, and you go, "Oh, no. This genius idea. We have to slow it down. I'm so sorry. I thought we had it."
You let them feel that you've acknowledged their genius, but you're like, "If we're going to do a lot of this because I think we can ship so much more stuff. That's what they want. They want to ship more code. They want more stuff in the world. We can do so much more stuff, but I really want us to have just a solid security review phase. I want us to do this, and I want to help us automate this, and I want us to move stuff along." A little bit of cynicism will go a long way, but you are on the hook to put that friction in because they don't understand why that friction's there in the first place. Also, interrogate yourself and ask why you want to slow things down. Sometimes we do just want to slow things down, to slow them down.
Manoush Zomorodi: Because humans work at one speed, not the speed of machines.
Paul Ford: And not 50 machines.
Manoush Zomorodi: Yes, totally.
Paul Ford: It's a lot. Our brains, it's like every meal is just nougat. You're like, "I like nougat."
Manoush Zomorodi: But not like so much nougat.
Paul Ford: Not like a Thanksgiving dinner of nougat for breakfast.
Manoush Zomorodi: I feel nauseous just thinking about it. Let's go to Bob in Shepherdstown, West Virginia. Bob, you're a teacher.
Bob: Yes, I am.
Manoush Zomorodi: Are you using AI? What are you up to?
Bob: I use AI not just for my job, but also just extensively a lot of different places in my life. I have a comment about the general sorts of things, but also in terms of AI and education.
Manoush Zomorodi: Oh, I think we lost Bob. That might have been my fault. Are you there, Bob? Sorry, go ahead.
Bob: Yes, I am. Moravec's paradox basically says in terms of AI, what humans find easy, is really hard for AI, and what humans find hard is really easy for AI. We have all these math papers are just coming out from OpenAI, but if you ask an LLM a common-sense question, they tend to fail at it. I think something that's been a common theme of some of the earlier callers is that AI can produce some beautiful solutions, beautiful papers, whatever, but it has zero real-world relevance. That's one thing that AI can't do right now is grasp what's meaningful to humans and what can be used.
Manoush Zomorodi: Are you talking about this stuff with your students, Bob? I'm so curious.
Bob: I'm talking about it in general. For my students--
Manoush Zomorodi: I mean, like, do you use your experience with AI? Do you talk to your students about what they're thinking and just debate this stuff with them?
Bob: Oh, absolutely. I tried to give them some instruction. It's one of the biggest problems, I think, with AI in education right now is people don't know how to make the best use of it. This has been something that's been true of any technology in education, from computers to when Wikipedia was introduced, all the way back to the ballpoint pen. If you research that, the outcry about ballpoint pens instead of fountain pens was crazy. I think we need to learn what is the best way of teaching students how to use AI. The big thing right now, people talk about the students asking, "Write my paper for me." In the literature, they call that cognitive offloading, with the idea that they're giving their thinking to the AI and the AI does the thinking, but it's not even that. There's no thought going on.
Manoush Zomorodi: Thank you, Bob, so much. The other term I've been hearing is cognitive stunting. This is by a researcher at Brookings. What do you think education-- I like that Bob is trying it out so he knows what he's talking about because he's getting his hands dirty. On the other hand, he doesn't seem terribly impressed.
Paul Ford: I feel one of the easiest ways to ruin your life is to talk about AI in education without-- This is the hottest of the hot buttons because it's the young brain. Educational software companies don't necessarily seem to have the interest of the young brain at heart half the time. They jam a lot of stuff in. You always are hearing stories about contracts getting canceled. I had this amazing moment where a bunch of students from the new school came to our office where we do AI stuff, and we're like, "What do you guys use it for?"
They're like, "Sometimes for work, but we're paying a lot of money." We try not to use it because we're invested in our own educations. It was so hopeful. I was just like, "Okay, good. You want that." I think it's how do you build a curriculum where they can understand how it works, why it lies, why it hallucinates, where it's good, where it isn't? Also, really, what do we want to guide students to do right now? We want to guide them to deal with the amazing amount of change that's coming in the world.
Manoush Zomorodi: And to get excited about learning.
Paul Ford: That's right, and frankly, for me, excited about technology. There's a wonderful educator named Julia Stanovich at NYU. We were talking to her, and I was like, "Are you optimistic?" She was just like, "Look, we have tools." It actually is that simple. If you think about it, we have tools to change and manipulate the world around us in our own interests and to empower ourselves. Go to that just for a minute and think what you can do with the tools. You owe that to yourself.
Manoush Zomorodi: I'm Manoush Zomorodi, and I'm with Paul Ford, and this is Please Advise on WNYC. We're going to take a very quick break and then more of your calls and texts on navigating AI and work. Stick around.
[music]
Manoush Zomorodi: It's Please Advise, the pop-up advice show from WNYC. I'm Manoush Zomorodi, host of NPR's TED Radio Hour, author of the book Body Electric. My guest today is Paul Ford. He is a programmer, a writer, a philosopher of technology, and we are taking your calls. We would love to hear from you. Maybe you've never even touched AI, or maybe you are using it all day long. What are your questions? What's your advice? Let us know. You can call or text us at 212-433-WNYC. That is 212-433-9692. I definitely want to answer a couple questions that are like, "What do people mean by slop?" We got to remember people are using terms that not everyone is familiar with. Another person is asks, "Paul, can you explain token and why an AI token is such a hot topic?"
Paul Ford: Great. We'll start with slop. AI generates things. Let's start there because actually the token part will make this a little more clear. It generates images, and it generates text, and it has a variety of methods for doing this. It's not one simple thing that it does. It's not a human process. It's not an edited process. It's not curated. No one is looking over its shoulder. As a result, it's actually gluing stuff together inside of its model and spitting stuff out.
It gets pretty close, but then it can't quite get that last little human bit. Sometimes it does get further. There's always this funny, and they call it glaze, especially on the images. It's like, "Why is it always a little blue-purple?" Or just, "It feels like the edges are off." It can be hard to put your finger on it. I think it's really hard for kids sometimes to tell the difference. If you look at it, if you see AI video, there's just something uncanny about the way things move and so on. You can produce a lot of it. You can produce it in huge volumes, and that's slop. Slop is slopped everywhere. It's the Instagram videos, all that.
Manoush Zomorodi: Or we have actually a texter who says, "I work in a butcher shop, and my boss insists on using AI to create posters of meat that look disgusting." That is definitely sloppy.
Paul Ford: So many. It seems to have taken the food and restaurant industry absolutely by storm. You know what it is? It's the exact same vibe of the bodega posters where-- I read an article. If you go to the bodega guys and you're like, "Hey, let's make you a nice poster that just shows a couple sandwiches and brings people in," they're like, "I have 30 sandwiches. You put those sandwiches on there." I think that retail people are just like, "Show it all, make it exciting." Then they see it, and it's the same thing. They don't realize that a lot of people actually, like, "I don't want to see a butcher shop slop poster." You don't.
Manoush Zomorodi: Sounds disgusting just saying it aloud on the radio.
Paul Ford: Let me go to tokens. Okay?
Manoush Zomorodi: Yes.
Paul Ford: You type in the box. It doesn't actually know what the words mean. It cuts the words up into little tiny pieces. Those are your tokens. The tokens there's different ways to cut up. Then, the way those little pieces go into the large language model and go through many, many layers, that's how it talks to itself. It breaks it up into little symbols, and the symbols point to zillions of other little symbols. When you hear them talk about billions and trillions of parameters, that's what's going on.
Then it comes back out, the same way tokens get emitted, it puts them back into words. That's how its brain is working and talking. It's not thinking the way that we think. It's matching and modeling and putting. Tokens represent a volume of input in the same way that bytes or megabytes or gigabytes. You're measuring things in tokens, and tokens burn a certain amount of energy and use a certain amount to compute.
Manoush Zomorodi: The harder the question, the more probabilities it has to weigh and therefore uses more energy.
Paul Ford: The longer it runs, more tokens and video takes a lot of tokens and so on. What's happened is in a world where the machines that process the tokens, the GPUs, and that's NVIDIA and stuff like that, and the energy usage that those machines require, it's very, very expensive. This whole world of tokenomics emerges where, "Hey, I have a problem where I want to talk to this thing, and I have a billion questions to ask it," let's say. Now you've got billions and billions of tokens that you have to process. That's going to cost a lot of money. That's why the token is so important. Think of it as almost like not even just a computer thing, but an economic unit. That's why everybody talks about it all the time.
Manoush Zomorodi: We also got a text that would like us to clarify that we're talking about using the term AI is very, very broad. We've got the LLMs that's like Claude or ChatGPT, and then you have all kinds of other AI, for example, machine learning in software, in medicine, where, again, it's using probability though to run millions and millions and billions of queries to try to get to an answer simply faster than humans might.
Paul Ford: I think it's a catch-all marketing and conversational term. When people say AI out in the world right now, usually what they're talking about is LLMs like Claude or ChatGPT or Grok or the other ones. AI is a term with 70 years of history. Starts in the '50s. John McCarthy coins it at a conference in Dartmouth. That's the first AI conference, the artificial intelligence conference. It's meant something different, like almost every five years.
A good example of something that we talk about too, where we don't think of it as AI in quite the same way, would be like a Waymo. You get in the car, it drives around. It's using machine learning models, it's using a lot of processing, but it's not using LLMs. LLMs, you can't control them in a nice deterministic way, the same way that you want to control the car that your children are in, driving around San Francisco. You don't want to be like, "Hey, Chat, get me wherever," and hope it works out. It does all get lumped together.
I think the other thing that makes it a little complicated, a lot of these systems and a lot of these tools overlap. You might use machine learning in order to analyze and classify a lot of documents that you then feed into your large language model and so on. There's a pretty broad spectrum of things, and they do end up living together. Machine vision is what also allows a computer to turn a picture into words that it can feed into the model so that you can ask what's in the picture.
Manoush Zomorodi: People who think they're not using AI, chances are if they used their phone this morning, they did use AI.
Paul Ford: Even if you're not on the internet, it's just everything. Your phone has a little neural chip [unintelligible 00:30:17]. Yes, it's everywhere. I think the thing that people are upset about, the thing that we're talking about, where the controversy is, is large language models from big labs that pretend to be human and do people's work. We get upset about that.
Manoush Zomorodi: Let's go to Alex, who is in Bed-Stuy. Hey, Alex.
Alex: Hey. How's it going?
Manoush Zomorodi: It's going well. You're a video editor.
Alex: Yes. I work for a political production company, and we're ramping up for the midterms. I'm locked in at my desk seven days a week. I really come around on the AI stuff. I use it constantly. If I'm like, "I have to find a royalty-free photo of Trump to put in an ad," it does that for me. It puts it on my desktop. I can bring it into After Effects. Anyway. I agree with that there's a difference between generative AI and the stuff I'm doing. It's obviously a different conversation in the military applications. It's scary, but I've come to accept that it's a part of a world where the need for content is so aggressive that you just have to keep up with it.
As a working freelancer, I just have to do it, and it is amazing. My question is really like I don't really have an answer about this issue of energy and water and the environmental impact. Just feels like there's different answers out there, and I'm wondering if your guest can speak to. Is there any conclusive truth on that?
Manoush Zomorodi: You're not the only one asking that question, Alex. Thank you so much. My understanding, Paul, when it comes to use of energy, yes, these data centers use a ton of electricity. However, it's not necessarily water use because that water is being recycled. That problem has been largely solved, is my understanding.
Paul Ford: That's right. Look, there are a lot of different ways to go get these stats. Then there are a lot of people who will tell you, like, "Look, golf courses are much worse." What I would say is, when you look at data centers, what you're seeing, there's a folk narrative that emerged, and we have to respect it because it's people saying, "This is my place. You've come and told me that you don't care about my job and you don't care about my family."
As far as I can tell, it's going to cost $90 million to decommission this data center when you abandon us, much like the auto companies did, and much like this. I'm going to tell you I don't like that. I don't like that, and that it turned out to be a very bipartisan issue. Look, here's what's tricky. If you take climate change seriously, it's like, "Oh, good, another thing. Throw it in with container shipping. Throw it in with the cars on the road." There are a lot of ways to contextualize energy usage. Then you'll get the big tech companies saying, "Hey, it's all good. We bought nuclear reactors."
We're just going down a path where we need enormous amounts of energy, and people are buying the product, and investors are excited and the future of data center energy is now like a tradable asset. There's enormous force trying to get these things built. There are very strong cases to be made that they're bad for the world, but I think you got to put them in with the other bad things for the world. The thing to understand is that is where people are saying, "Hey, I've had enough."
I think that if we like living in a democracy, which hopefully one day we do again, this is like, "You got to listen to that." I got to tell you, as someone who loves this stuff, if you told me I can't have any more, I'd figure it out. If you told me I had to just run the cheaper model, and I'd be limited a little bit, I'd be okay with it. We're so focused on getting there, getting there, wherever there is. Nobody can tell us. It would be okay to slow it down. It wouldn't actually be the worst thing in the world. We could get a little more efficient, we could get a little more smart. It's a fuzzy thing, and it got people all riled up, and I think we got to respect it.
Manoush Zomorodi: If you have a question for Paul Ford or an experience using AI at work that you want to share, maybe you've got advice for other listeners, or you really do not understand what is going on in your office with AI, we want to help you. Give us a call. 212-433-WNYC. That's 212-433-9692. One of the things I've been thinking a lot about is AI burnout, which is this idea that people are running multiple agents, six, seven, eight agents who are churning out work and then the work becomes checking their output, which is a lot. Then I also hear from a lot of coders or former coders, software people, who are saying, "Oh, I'm not working less, I'm working almost double because of all that I can do. There's just no stopping."
Paul Ford: I'm like, yes, is the truth. A lot of developers are depressed because they used to go in, apply their craft, see the outcomes, talk to their peers, and then go home. Instead, now it's like, "Hey, did you set up your agents to run overnight?" These things can create infinite volume. That volume can be good or bad. It can be slop. You're in there as a slop processor chewing through the slop every day, and it's a bad--
Manoush Zomorodi: Delicious.
Paul Ford: Exactly. Just like that butcher shop. I'll tell you, I'll go to personal experience. I got really into vibe coding because I'm a programmer, and I never--
Manoush Zomorodi: Define it for people who are like, "I've heard this term vibe coding. What even is it?"
Paul Ford: Thank you for asking. Just like, you can prompt an LLM to produce a document, you can say, write me some code. The code could run on a mobile phone, it could be a web app, it could be a webpage. Just all the things that run on your computer you can now do. If you're like me and you're like, "I like to explain technology, but I'm actually not that great a programmer," things like building machine learning models or doing 3D stuff have always been a little out of my reach. Boy, I can do it now. I can get it all scaffolded and built. I can understand it, and I get the architecture, and woo, it's fun. Then you're like, "I have become a God." You're like, "I'm going to do 20 of these at once because my brain is so amazing now. Oh my God, look at me."
Manoush Zomorodi: You're superhuman.
Paul Ford: I'm so superhuman. I probably built like 45 apps that no one will ever see that don't matter. Part of it is learning, but part of it was like, "Whoa, what are you doing?" Now I have--
Manoush Zomorodi: My sister gets so mad when I tell her the things that I've been trying with AI. She's like, "Great, thanks a lot. You're ruining the earth for the rest of us."
Paul Ford: It's literally really hot outside, Manoush. Thanks. That's you. Like when I go to the beach and it hurts, that's your fault.
Manoush Zomorodi: That's what she's saying, and she's not wrong.
Paul Ford: What I have learned, and this is partially my brain. First of all, there was a moment where everybody was talking about token maxing. Like, "Let's burn as many tokens, and it'll help us make lots of money." It didn't work. I've talked to so many heads of engineering and folks like that, and they're like, "It's a disaster because everybody keeps dumping more slop in it." They're like, "Okay, I did it. I did my work." Everybody has to go back. We had to have an all-hands meeting at work where we reminded everyone to do their job and to have meetings and make eye contact.
You're sitting there as an engineer, and you're like, "I can design," or you're sitting there as a designer, and you're like, "I can code," but you don't know. That's not what people are paying us for. It's really cool. I do it too. It's exciting, but you got to be very careful about where the boundaries are. I program one thing at a time now. When the LLM is working, I don't ask it for more stuff. I remind myself to do things like read books, think thoughts, sit quietly by myself, and it's cool.
Manoush Zomorodi: I try to remind myself to read all the things that the LLMs spit out and actually just pick out the good stuff.
Paul Ford: That's right.
Manoush Zomorodi: That alone, I don't need to keep churning and churning.
Paul Ford: Stop trying to turn a wheel faster. It actually doesn't get you anywhere. You're spinning.
Manoush Zomorodi: Speaking of which, we have to take another break. I'm Manoush Zomorodi. You're listening to Please Advise. We're going to take a very short break and then more of your calls and texts about work and AI. We'll be right back.
[music]
Manoush Zomorodi: You're listening to Please Advise, the pop-up advice show from WNYC. I'm Manoush Zomorodi, host of N NPR's TED Radio Hour, author of the book Body Electric. My guest today is Paul Ford. He is a technologist. He writes about technology. He makes technology. He's also a human and actually really thinks about how we humans should use technology. I would love to go to Miriam in Westchester. Miriam, would you have a point you'd like to make?
Miriam: Yes. I'm concerned about the theft of intellectual property. Right now, if you search something on a browser, the first thing that comes up is an AI response that is scraping information from all these authoritative sites, essentially stealing the information. I think there should be a law where you have to opt in if you prefer to have an AI response come up first. I think I have to search harder in order to get to the actual authoritative official sites. I'm just curious that any legal exploration of this issue.
Manoush Zomorodi: That's a great point. Miriam, you're absolutely right. Paul, to me, it's like if we keep training and training these models on existing intellectual property and we don't create any new intellectual property, aren't we going to turn these LLMs into essentially churning butter that just tastes like nothing, essentially.
Paul Ford: Slop butter.
Manoush Zomorodi: Slop butter, exactly.
Paul Ford: That is what we are making. This one's the hard one.
Manoush Zomorodi: Oh, Miriam.
Paul Ford: This is the bummer. This is the real bummer. I'll tell you why because first things happened. Like the large LLM, they just helped themselves to all the pirated books on the internet. They just went ahead. That's why Anthropic had to pay out to thousands and thousands of writers. They got to the point where they're like, "All right, fine, 1.8 billion. Let's just get this done with." That's not usually something you do if you're feeling really good about the decisions you made.
They really did help themselves. They helped themselves to the open web. Then it's very strange because they digest it and then Google, which was pretty much the trusted brand, just goes ahead and puts AI stuff at the top, which is often incorrect. It has little tiny links to the places where the data comes from, but they're little tiny. You got to work to click on them. It is utterly possible, and I can tell you it's utterly possible because I have done it. A lot of my work when I build things with AI is to just get back to things that are like search results. I built a local news search engine. I built tools to help me research, and I've built tools to help me find good, interesting links to talk about in different places.
You can actually go and find the original text, and you can link to the original source. You can drive traffic and engagement back to the people who created the material if you choose to. You can use AI to do that, and that would be good for the web because it would incentivize people to create new things and maybe sell some ads. At least they would feel basically respected. They came into the comments and things that weren't in the comments, things that were copyrighted, and they help themselves, and they just aren't that interested in giving back.
This has been there on and off throughout the web. A big company is always, "We're going to pivot to video and help the publishers," and then, "Whoops, sorry, actually we're--" The thing to remember is publishing is just so tiny compared to these numbers. Everything we do in this world is just a little tiny percentage on top of real estate sales. They want us because they want really good content to feed the thing, but ultimately we're not their money. You just end up in this place where it's just like what is it when you don't replant properly, and the soil dies?
Manoush Zomorodi: It's degraded, and you need to rotate your fields.
Paul Ford: That's it. We're not rotating the fields. I don't know what motivates. A legal structures might motivate [unintelligible 00:42:53] You're seeing things. Also, California is starting to invest in local journalism.
Manoush Zomorodi: Like CalMatters.
Paul Ford: Yes, that's right. That, to me, unless asking the giant tech companies to wake up with a really fully formed conscience about helping other people generate and publish great ideas and rewarding them for doing that, I don't think that can happen. I've given up. Maybe government needs to fund a content ecosystem and support it. Maybe there needs to be a really-- You have to opt in to this. I don't mind that idea.
Manoush Zomorodi: The thought of having to opt in and not automatically being enrolled in something that seems like the tech companies never want to do it. Then meanwhile the EU has done it, and all it is is annoying, and there's more pop-up boxes that you can click for GDPR privacy protection.
Paul Ford: Everything ends up being a legal battleground every time you open up a webpage. Look, a lot of times there are better ways to do things, but they won't be the most profitable. The one sort of beautiful thing in this is I literally was like, "Boy, there's a lot going on with really local, hyperlocal, tiny coop news organizations." I like reading local news, I like to see what people are up to. I want to find out.
I made, it's called GiveMeSomeGood.News. Just one weekend I made a website, and it indexes that news. You can look through. I check it every week or so, and it has RSS feeds and everything, links to the donate button and the original article, and so on. I've now got my local news engine, and I could build it in an afternoon. I just think we need thousands and thousands of things like that to build an ecosystem that we care about. Then we can all go back to linking and blogging like it's 2004. It's going to be amazing.
Manoush Zomorodi: [chuckles] Oh no, not the nostalgia argument.
Paul Ford: No, I know it's not going to happen.
Manoush Zomorodi: It's interesting, Paul, because I'm looking through all the texts that we're getting, and people saying-- One woman is saying, "I work for the New York State government. I use AI to help with procurement questions for example.
Paul Ford: Oh, yes.
Manoush Zomorodi: I need to find which vendors I can use. This particular contract has 508 vendors. To check this manually would take hours. I use Chat. It goes so much faster." Then there's an artist. I can't remember where they're texting from, but they're saying every time they post their art, they're just worried that it's just going to be scraped and used by someone else. There's this constant ambivalence and tension that we feel.
I want to go macro in the last few minutes that we have, because I don't think anyone knows how this is going to go. I was reading the headlines this morning. It said, "Will AI replace us or just change the way we work?" Then the FT asks, "Silicon Valley thinks AI will kill jobs. Economists aren't convinced." When you zoom out, how do we stay grounded? We don't know what is going to happen. Meanwhile, there's a lot of fear and confusion.
Paul Ford: Look, I saw someone give this advice the other day, and I think it's really good, which is you really should sit down and use for paid, not free plan with one of these things and try to solve a problem. Just make sure you understand it so that you're not talking about--
Manoush Zomorodi: Why paid?
Paul Ford: It's just better. It just does more, and you'll see what everybody's talking about. That person from New York State who's unlocking stuff, they might be using Microsoft and Copilot tools like that. Those are getting pretty good. They're useful, and they're there at work. The things that she's describing, anything where you're like, "I have a pile of 600 PDFs that are relevant to my job, and I have to do things with those PDFs," boy, does that get better.
For the artist, we're back to where we were talking. They do help themselves. Zooming out, I think we have a radically new and confusing thing that simulates a human, even though it is not human. The people who are creating it have a very unusual belief system compared to most other Americans. They really are very concerned. They think that superintelligence is around the corner, and they've been pursuing that ethos for like 20 years. It's kind of, "This thing is here." It all is really, really muddy. I think you just need to lock in on what feels and looks real. You need to try it, and you need to know what it does. Then you need to make up your own mind there. It's horrible to have to go, "Think. That's why we made AI."
Manoush Zomorodi: We're going to go to Jesse real quick. Jesse, I think you have a very small question, actually.
Jesse: Yes. Hey, guys. Jesse from Tenafly. Thanks for taking the call. I've heard two conflicting views. One was initially, be polite to your chatbot, and then recently I heard someone say, "Don't be polite," and it's a waste of tokens because it doesn't care. What are your thoughts on that?
Manoush Zomorodi: Thanks, Jesse. Paul.
Paul Ford: Be polite because you are a good, kind person, and you should always practice being polite. Who cares? Like what? Like, "Oh, no, I wasted those tokens." What? Then somebody makes a picture of a cat on a rollerblade. It's so incidental. Be a human first and foremost and be above this.
Manoush Zomorodi: I disagree. I think do not say please and thank you. It is not a human. You need to remember it is a machine and not give it that kind of extreme power.
Paul Ford: I'm polite to my cats.
Manoush Zomorodi: But cats, they can purr. They want to be with you. This computer does not want to be with me, or maybe it does. I don't know, but not yet.
Paul Ford: Be polite, Manoush. Be polite.
Manoush Zomorodi: No, no, I say no.
Paul Ford: Okay.
Manoush Zomorodi: The bottom line remains, though. Look deep inside yourself and figure out what's right for you.
Paul Ford: That's right.
Manoush Zomorodi: We're going to have to leave it there. Paul Ford, you're the best. I admitted on air that I had a crush on you. Did you know that yesterday?
Paul Ford: No.
Manoush Zomorodi: Thank you so much for being here and listening to Please Advise. Coming up next week, a new host, Naima Reza, host of the podcast Smart Girl Dumb Questions, leads a conversation about finding love and dating for compatibility with author and relationship expert Jillian Turecki. It'll be the first in a series of conversations that help navigate the various phases of partnership and family. Good luck. You can call in Monday with your dating questions or submit questions online in advance at wnyc.org/advise. I'm manoush Zomorodi. I have had so much fun with you this week. It has gotten really spicy, but it always does when you talk about technology. Thank you so much for having me. Paul, one last word from you. Are you going to use ChatGPT to figure out your next move with your company?
Paul Ford: I'm a Claude person, and other tools, DeepSeek. I use it for a lot of things. The company, by the way, aboard.com, go check us out. We build things. We help people. We love you, and we're your friends. We like to host events to help people understand AI in New York City. We're trying to be good citizens.
Manoush Zomorodi: I love that. You can be in tech and be a good citizen.
Paul Ford: We're going to do our best.
Manoush Zomorodi: I love that. Let's leave it there. Thank you all.
[music]
Copyright © 2026 New York Public Radio. All rights reserved. Visit our website terms of use at www.wnyc.org for further information.
New York Public Radio transcripts are created on a rush deadline, often by contractors. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of New York Public Radio’s programming is the audio record.