The Ethics of DALL-E, with Harry McCracken

Hello and welcome to The Media Copilot. It's a new podcast and newsletter all about how AI is changing the media and how the news gets made. I'm Pete Pashal, longtime tech journalist and media executive, and with this podcast, I aim to have interesting talks with fellow journalists, decision makers at media companies, and just fascinating people doing interesting things with AI that newsrooms should know about. I'm very pleased to welcome my very first guest to The Media Copilot podcast, uh, Terry McCracken. Harry is the global technology editor for Fast Company, and he, like myself, has been a lifelong tech journalist at the forefront of pretty much every major innovation of the past 20-plus years. Um, he's also the author of Plugged In, Fast Company's weekly report from the world of tech, and he's here today. Well, Harry, welcome to the podcast.

Hey Pete, great to see you.

Great to see you. Awesome. So I'm going to start off with A question I aim to start off most of these podcasts with, but you know, you as a journalist, you're a reporter, a writer, an editor, you're a creative guy. How are you using AI today in your work?

Well, I'd say that probably 80% of the stories we do in our tech section at the moment are about AI. So, uh, we've shifted a very large percentage of our editorial and mental bandwidth.

I was going to say, just 80?

Yeah, it might be higher. And then that's just the tech section, and we were doing AI and other sections we have as well. Uh, but we do— I mean, we really shifted our editorial and mental bandwidth to it dramatically and, uh, realize that because all of our readers and people who, uh, consume our content kind of need to know about it, we need to know about it a lot too, and to be able to convey why it matters to the people we care about. Internally.

Do you have like a lot of subbeats within AI? Like—

Well, I mean, a lot of what Fast Company covers is about the intersection between tech and something else people care about, whether it's politics or creativity or business. So rather than having these really well-defined beats, we sort of have people who write about these intersections. And I do have a colleague, Mark Sullivan, who really is formally charged with primarily writing about AI, and that's great. But I think it would be a mistake to say that AI is a silo. Uh, AI is touching almost everything we cover. And so all of us need to, uh, start to grapple with it and cover it. So I'd say at the moment, how AI has impacted our world as journalists is primarily it's had this huge impact on coverage and we are at a much earlier stage and actually leveraging it to do the work that we do.

So I feel like there's always sort of, or at least there used to be, this sort of semantic debate around AI about like what's really AI and what isn't. And to a large extent, I feel like that's become irrelevant because, you know, it's just clearly some kind of force that is sort of pervading everything, as you said. But does it— maybe it bears some, just a pause here at the moment just to define what exactly the world has come to know as AI? I mean, whatever we define it as in this podcast, again, totally irrelevant, but like, what, what does it even mean in the world now with AI? Because I think it is helpful for, particularly for a publication like Fast Company that's obviously just dived right into this. Like, how, how, when you, when it comes time to separate like what someone's really doing that's innovative versus something that is sort of an also-ran or just hype and slapping a label on something, Maybe it helps to define it. So how would you say the world has defined AI at this point?

Well, the thing that we're all fixated at the moment is really generative AI. So AI, which is using large language models to create content of some type, whether that's words or still images or videos. And generative AI is also what lets AI understand what we type in prompts and respond to it. So, um, that's a huge and transformative part of AI. It's certainly not the only part of AI that's worth thinking about. I do think that over time, though, the way that we really know that technologies matter is when we stop talking about them because they're just part of the air that we breathe. Um, you go back in this business a long time like I do, and I remember when, when multimedia was a thing and we were starting new magazines about multimedia, uh, and people were amazed by, you know, full motion video and high quality sound and CD-ROM. And very, very quickly it turned out that multimedia was not a separate topic from computing. And so nobody talks about multimedia. And like a decade after that, people got obsessed with mobile, and there were people who had mobile as a beat. And pretty quickly almost everything either was fully mobile or mobile was an aspect of it, and we stopped talking about mobile as a thing. And I really don't think it's it's going to take all that long until AI is just everywhere and, uh, we don't think of it as a separate category. It's really just fancy software.

Yeah, it reminds me of when at Mashable we were like, in the early days, we were the social media guide, and there was even like a social media day. And then a few years after that had sort of started, we were, is this even relevant anymore? I mean, everything is, everything's social media. Social media is just a thing. And now it's funny, we're on sort of a bit of a crest in social media, it seems like now. But anyway, that's not what we're here to talk about. So, well, generative AI, 100%, that's, that's where all the excitement is right now. And, um, so as a, as a publication, I'm sure Fast Company is sort of struggling with, or, you know, maybe excitedly thinking about, um, the potential of this technology to change not just what you're covering but how you're covering things. And I understand that you, uh, probably have some input into that. And to the extent that you can talk about it, like, what— how is Fast Company approaching this? And, and what are you guys thinking about in terms of using generative AI, um, in, in your whole process?

Sure. Well, I, I'd say that on the highest level, we're trying to approach it thoughtfully. I don't think at this point we feel like we're under an enormous amount of pressure to use as much AI as possible Partially because we're a relatively small, focused organization. We're not trying to cover the waterfront and do hundreds of stories a day. Um, there are all kinds of topics we don't cover because when, when we do try to cover everything and someone else can cover it better than we do, typically it doesn't do all that well for us. Uh, but we're really focused on business and particularly innovation in business and people who want to be forces for good within business. And there's, there's a lot we don't cover and that, that seems to to be fine. And generally speaking, our business does best when we're most focused on this very specific group of people we care about.

So, um, you seem to be saying that like you're the sort of most obvious use case of generative AI, which is to say scale up like what you're publishing isn't really that high on the list of things you're looking at doing. Is that fair to say?

Yeah, I mean, we're not really under the gun to cover as many stories as possible with as few people as possible, which is a good feeling. And like everybody else, we've also seen a number of news organizations that really did start publishing AI-generated stories as quickly as they could, and all the trouble they ran into. And it certainly has not been good for their, their brands in most cases that they, they moved with, with great speed.

I mean, yeah, let's get into that a bit. Again, just more as an observer here, like, I feel like that had a major searing effect on the rest of the industry to both sort of avoid that and— but to also get, get some kind of perspective on this technology, lest it sort of, um, sort of manifest on its own, sort of a grassroots way in their own newsroom, right? And, uh, what— but that, that also, you know, is a little bit in the past now, right? Because that, that all sort of went down in early this year, early 2023. Um, I feel like we're sort of entering a new phase with that and, and sort of how the media regards AI and uh, specifically hallucinations, uh, and maybe it's simply avoiding that use case. Um, I don't know, how— where do you feel, um, the media is at? And again, if, if— again, if, uh, steering it back to Fast Company— if you're thinking, well, let's, let's not— let's avoid generative articles, what, what else can we use this for?

Yeah, I'm— I mean, I imagine that probably at this point the majority of, of people in the media have seen all the, uh, kind of weird things that can happen and feel good that they haven't dived in. It does feel like at some of these media outlets, the people who are most excited are not the journalists, they are the executives who pay for the journalists. And at some places, um, those folks may be putting great pressure on, on the editors to do stuff as quickly as possible. But I, I'd say that it's not that the entire industry is trying to do this at this point. It's, it's, um really a smaller number. And I think we probably will see fewer companies doing that just because it doesn't seem to have a whole lot of upside. It's certainly true there are a ton of other ways that AI can have a positive impact on what we do and make us more effective and efficient in our work. I've been using Grammarly for several years now, and I've written about it. It's not a deep dark secret. At first I thought, um, maybe it was something to be embarrassed about that I wanted to like proofread and copy edit my own work, but I actually find it quite helpful. And Grammarly was kind of an early, um, canary in the coal mine of AI because it's, uh, been using it for several years now, and I run a lot of my work through it. So I certainly couldn't say that I'm opposed to AI having some impact on my work because there are certainly cases where AI suggested I make a change to a draft and I accepted it, although there are tons of places where AI makes helpful suggestions that I don't like and I ignore.

Yeah, I gotta say, Grammarly is a really good example. Um, again, now you might get into the semantic debate of what's actually AI and what isn't, but one of the things that, uh, we explored at, uh, CoinDesk when I was there and heading up their AI committee was AI copy editing and I've— for using these newer chatbots, whether it's, uh, you know, OpenAI's or something else, um, I found that they did a decent job, but there was no paper trail. And because of that, the reliability was always sort of a question. And so, you know, Grammarly, on the other hand, uh, because of the, you know, infinite settings, the way it works in terms of like approving changes, etc., it's just sort of this custom-built AI tool for that purpose, you, you have that paper trail, which I think, um, is— gives a lot of assurance in a sort of formal editorial process. And I, I feel like for even, even tools like ChatGPT or similar stuff, the showing the work part, it will become sort of increasingly demanded by media companies that are sort of want to use these tools. Would you, would you say that's maybe a trend we might see?

Yeah, I mean, I think Grammarly certainly benefits from, from the fact it's this purpose-built tool. Um, yeah, and I find it to be very helpful. I mean, there are also lots of suggestions that aren't so helpful, and just recently I found it more often being kind of out and out wrong on stuff or making suggestions that, that don't make sense. So, um, it's not a substitute for human copy editors, and everything we publish is run run through humans, at least one extra set of eyes. But I do feel that on balance, it's helped my work. And the fact that it was built specifically to improve writing is helpful and lets it do things that ChatGPT is not going to do. And in fact, I think probably over time, the world will be less interested in these Swiss Army knives of AI and more interested in products that were created to do one thing and do it well.

One of the things I sort of feel like is a, uh, that comes up a lot in, uh, when you talk about AI is sort of the metadata and the social copy and all the things you kind of fill in usually after you're sort of done your article, right? So, um, I, I feel like, you know, there's this been this trend, that undoubted trend in journalism the past 15 years where every reporter and every editor has also had to become kind of a content marketer. Right? You kind of like, okay, I have to— I've done my article, now I have to understand the SEO audience, the social audience, the sub-audiences within those social networks, think about the thumbnail image and how is that different from the other images, and like all the ways my article will compete online in these distributed systems, and sort of trying to maximize my chance of someone clicking on it. And that's— some people are good at that, some people aren't, some people like it, a lot of people don't. I find. And I feel like with AI, generative AI, there is this promise almost that it has that it's like, well, maybe that will reverse itself now. Maybe there will be a, you know, generate for me button in every CMS in the near future that essentially taps into AI and does that for you. Are you seeing the beginnings of that with what you're doing? Are you using AI for social posts for SEO titles. Um, how far along do you think we are in that? Is— and is that thinking even correct?

I certainly think it's going to happen, and I hope it does happen, because, because you're right, we have a number of fields here we fill out that are for SEO purposes or promotion purposes, and it's not even clear to us as humans, um, the best way to fill those out. And it's kind of time-consuming and not a lot of fun. And we're not using AI yet to do that, except occasionally if, you know, occasionally just as an experiment, I will plug that into something like ChatGPT. It wrote one headline for me, which I was pretty happy with. And then the next time I tried, I was not happy. AI is very good at summarizing things. So I think it's probably gonna be way better at something like composing a potential tweet. And there's a big difference between expecting AI to write a 600-word story and get it right and expecting it to write a sentence or two, which you can see before it goes out and edit. So we're not actually doing any of that on a formal basis yet, but I expect we will. And if I imagine that there are people right now writing like the WordPress plugins that will make that easy, and maybe even WordPress will just build that stuff in because it has so much potential to do a better job than we can. And save us time, we can apply to the stuff where humans are still way better. Summarization just in general, I think has a lot of potential. Like earlier this year, I did a story for, we ran in print on the history of LinkedIn where we talked to about 45 people and I ended up with hundreds of thousands of words of transcript. There's no way I can read all of that and identify the best nuggets to use. And so I did muddle through it on my own because this was earlier in the year, but I think AI will be really helpful in terms of kind of synthesizing raw materials like that and helping. And again, because it's doing so behind the scenes, if it screws up, that's not a big problem because there are gonna be humans looking at it before anything goes out.

Have you heard much about how sort of applications might be sort of taking what you're talking about here, the summarization, the almost like the Uber look at essentially a dataset and pulling out the things that are most relevant. Like if you project that to something like investigative journalism, right, and building sort of an LLM or using an LLM to look at, say, that big swaths of public data, like, you know, a lot of cities like New York sort of put a lot of stuff out there and Um, but the thing is, you access it now, you kind of need to know where it is, you kind of need to go sift through a lot of searches, um, etc., etc. Um, do you, do you see those tools, uh, being designed so where it's like it could really just be a chat, right? Like, you know, go ahead, you know, chatbot, tell me, I don't know, how many firefighters got hired or fired last year in this place. And then you don't, you know, you don't even have to go anywhere. To do it. It's all just the simplest interface. It's bringing, I guess, that chatbot ability to these sort of big datasets. Is that something— I mean, it's obviously something that could be useful, but are you hearing much about that? Are you seeing that? Is anyone building that to your knowledge? I mean, I think I saw someone in Toronto. Toronto had done something like that, but not much other than that.

I haven't heard a huge amount. I think it, I think it does have a lot of potential, um, just in the same way that, um, like 30 years ago, wow, when journalists were researching stuff, they didn't even have something like Google to help. And, um, those tools have had a huge impact, um, on how, uh, serious reporting is done. I do think you have to be a little bit careful because there's so much important information you're not going to find via Google And even when, uh, AI gets better at, at helping look at some of these databases, it's not gonna find everything. And, and some of the most valuable research is the stuff that's difficult to find. And, and we shouldn't forget that. But, but yes, I, I do expect that some of these databases that are full of valuable stuff will be easier to make sense of, uh, when AI gets better at looking at them and, and AI just has access to them. Of course, there's this huge question of whether all the entities in the world that control useful information even want AI to get at the data they have. But I think in a lot of cases, they will determine it's a great thing as long as it's done in a way that makes sure that there's integrity to the data and it's not just about some third party scouring what you have and using it for their own purposes.

So you— this is— I wanted to ask you this because as a tech guy, I think you, you might have a good perspective on it. So there's, there's a lot of different sort of models out there, approaches to AI, sort of when you start sort of cutting into like the implementations of generative AI, you know, you have commercial solutions like OpenAI, Anthropic, you have these open source models like Llama and other things out there. Some people are thinking about doing their own models and like it's usually like a Bloomberg GPT has been announced, and that sounds expensive, but maybe it's worth it for them. And there's even other approaches that I'm probably not even thinking of. And as people think about this, particularly media companies, on sort of like deploying AI, how do you think that even deciding like what to use, how do you even start to frame that decision? You know, what, what do you think, given your knowledge of sort of the pros and cons of these things, how these models work, how they tend to compare?

Yeah, I mean, I think, I think the world is still figuring it out. Um, I've talked to a whole bunch of tech companies who are figuring, figuring out those questions for themselves. And, um, like just in the last week or so, I talked to, uh, the CEOs of Zoom and Dropbox, and both of them ended up deciding it made sense to use several models. Really? Yeah. So they're using—

That would take some tech support, you know what I mean? Like that would increase sort of the bandwidth needed on your engineering side.

It might, but I think that, you know, so they're using like GPT and they're using Meta's Llama and they find value in it. I think one of the basic things that has not gotten an enormous amount of coverage is that AI is really expensive computationally. And you don't wanna throw everything at GPT-4 because there are a lot of AI tasks that don't require that much power. And if you are able to have access to several models and weave them together, you can do it in a much more cost-effective way and use each model in the way that leverages their strengths. I do think that's a challenge for media companies though, because if you're Dropbox or Zoom, hopefully you have a ton of people who really know what they're doing, and you are by definition a tech company, and, uh, in theory at least, are in a place to make these decisions. Uh, you know, over the entire course of the internet, we've seen the fact that it's much more difficult for media companies to get these things right because very few of them are inherently great at tech, and we've sort of spent the last 25 years getting good at the internet. And, um, even if we have— and I think we have made a lot of progress— AI is a little bit intimidating because I think we will be required to have some understanding of this even newer, more potentially transformative technology. And very few media companies are going to be able to deploy hundreds of people to tackle this stuff. You, you might be lucky if even if you have like a small tech team that is able to understand AI at the same time that it's dealing with all the other day-to-day stuff it has to think about.

Yeah, that's interesting because it's like it seems to be a bit of a catch-22 in terms of like what you might think of as the democratization of, of certain things, right? Where it's like, yeah, sure, AI tools are available to anyone who can pay for a ChatGPT subscription, but to really support them and sort of deploy them at any kind of scale, even for a small shop, I mean, you're going to need, you're going to need the engineering support for that. And then that's going to take up like lots of, of the bandwidth over there and you won't be able to progress on other things. So it is a tough thing.

Um, I do think that, um, well, we'll learn some lessons from the internet. Like, like in the early days of the internet, uh, all these media companies were building their own CMSs and pouring resources into them. And a few large companies still do that, but, but most of the rest of us are on WordPress, and WordPress does most of the heavy lifting. And what they don't do, third parties do, and they build plugins. And I imagine that a lot of the ways that AI will be useful can be packaged up and provided to media companies so they don't have to do vast amounts of heavy lifting on their own to get some value out of them.

Yeah, yeah, I definitely tend to agree. Um, so let's just switch topics for a bit. So like generative images, they've definitely gotten a lot more attention. Uh, I was just writing about them today about how sort of the ease of use of these things, which has been around for a while, is now starting to get coupled with what you might call safe services, you know. So Adobe Firefly is the one that always comes to mind because, well, after sort of the legal issues that happened with Getty and They essentially came— Adobe came out with a— their own Firefly, which— well, they had Firefly, but in terms of the generative capabilities, they made sure the imagery they were using was stuff that was licensed or in the public domain. And that seems to be the way to do it, and that seems to remove a lot of barriers in terms of legal concerns, ethical concerns. Um, so it comes to this to the point where, you know, like, why not use it? I feel like once that's there, even, even if it's sort of in a grassroots kind of way, it's sort of like at an individual level. And even the free plan gives you like 25 free images a month. Um, I don't know, like, I mean, there's a lot of stories I've written over the years, and I'm sure you have too, where the image that is— it's more of an abstract concept and there isn't like an image that you can really associate with it, and you're just hunting for stock images and you're constantly looking for the right one. And You know, that takes a half hour and it would just be like, if I could just tell Adobe Firefly or whatever, you know, generate this thing for me and slap it on and be assured that that's okay. Um, why not do it? Um, I don't know. Do you see that maybe that's going to become more of a norm anyway in, in media, particularly in smaller places? And then what does that mean?

It made me think of, uh, about 15 years ago when I had my own site, Technologizer, and I was responsible for doing all my own imagery. I would sometimes get very into it, and I remember doing a story on Palm, um, the smartphone company, and how I was taking on Apple. And I made an image of a Palm Pre phone riding a horse and tilting at a windmill, which required me to get a Palm and find a horse I could use and a windmill and adjust everything. And, um, I need to get it doing layers in Photoshop, right? Right, right. Um, This is something which today, if I was doing the equivalent, I could do in 2 seconds and probably would do a better job than I did. I do think for kind of more professional, serious media outlets that actually take imagery seriously and hire actual artists, there are some major issues here. Um, at Fast Company, and we have some policies in place, we actually, uh, do find value in generative art, but we have some rules like we're not going to publish stuff that was created purely through a prompt. Um, it's fine to use things that were generated as an element of something that a human being touched, in the same way that we use stock imagery. Um, we're disclosing when we use that. Yeah. Uh, we're not—

if I just stop you there, like, what, why, what is the issue with something that is generated by a prompt? What, like, can you really break it down for me? What's the problem?

Sure. I mean, there are a whole bunch of them, one of which is AI is getting really good at creating things that look like photographs, and we don't want to run something where we had AI create something that people might think actually happened and was actually generated by a machine.

Well, couldn't you just put that in the caption?

Well, even then, I think there are issues. Uh, yeah, I don't— you know, personally, I'm not crazy about— in a lot of ways about the fact that, uh, we are entering this era where there will be things that look exactly like photographs and, uh, were generated by somebody in, uh, very quickly. So I do worry about that, and I'm glad— I'm glad we're erring on the side of caution.

I'm glad we're having this discussion, and the, the— because it's helping me think through this, because I do think there is a— there's sort of the microethical issue of like, was the work that this particular AI was trained on— were the artists compensated?

Yes.

But then there's also like a bigger picture ethical concern, which I think you're kind of getting at, which is to say like, well, if as a matter of policy we deem this okay to do, does that then cheapen the whole industry and just make things harder for everyone in general? That's a thing I haven't— I don't know if anyone's really fully grappled with.

Oh, and then also we have a rule that we're not going to try to create something in somebody else's style, even though you can tell AI to generate something in somebody else's style. I mean, that's— I think we see that as kind of a form of intellectual theft. And if you like somebody's style, you should hire that person to do it. Having said all this, I think that my philosophy would be different for very small media organizations such as newsletters, including yours, and places where you're probably not going to hire somebody to do art from scratch for every issue of your newsletter, probably economically.

Maybe if once I get a few more paid subscribers, once I get a few more conversations like this one.

Yeah, if it's the equivalent of what I was doing with my blog, I think that it makes sense. Um, even then, I mean, I think The commodification of imagery has some challenges. And even when I was doing my blog, I wanted the artwork to be distinctive. And certainly Fast Company as a brand has some definite theories about the stories we're telling through imagery. And we don't want to do it through generic stuff that was obviously generated by a computer that's generating similar imagery for everybody else. So we, we like it to be distinctive. And hiring a person to generate it and working with that person to create it is a great way to make sure it's distinctive. And I think over time AI will be able to do that. And I do feel good about the fact that we're not saying no way are we gonna do this. We are doing it, but we're doing it in a way that's erring a little bit on the side of caution at this point, which I think we've seen many pieces of proof that it's better to be careful at this point than to be too gung-ho about it.

Yeah, yeah, definitely, definitely would agree there. Um, so just big picture, all right? So I'm gonna ask you to put on your sort of wizard hat, staring in your crystal ball a little bit. But like, there's no question in the present moment we are at peak hype for generative AI, and every day just seems like there's more news, and it's just, it's pressed into something. Um, what's the next phase, do you think, once, once we're past that peak? Are we nearing the— are we at the peak? Are we past the peak? Maybe we're just past it. But, and do you think it is going to be sort of a sudden bottom falling out of this because some, some promise of AI that everyone seems to assume, whatever that is, isn't delivered on at some point and everyone realizes it? Or is it going to be more gradual? Is it going to be more Okay, like you kind of said at the beginning, it's eventually just going to be the air we breathe and we're just going to sort of stop talking about it because it's just going to be here. I don't know, it feels like we're not quite there, but like what, what, what happens in sort of the next 6 months to a year, do you think?

I think that right now we're kind of hitting against the reality of it, of it has to be economically feasible. Um, there were stories the other day about the fact that Microsoft's GitHub Copilot, which might be the single best example of something that actually is out there changing the world because an awful lot of people who write code are using that copilot to help. And it really has had an impact on how software is written. But these stories said that Microsoft is losing money on every customer because so much computation has to take place. For it to happen. And, um, Microsoft is not going to be willing to lose money on Copilots forever. And with it, with the Copilots it's introducing for Microsoft 365, which most of us still call Microsoft Office, they're charging $30 per user per month, which is really a ton of money. I mean, there are whole versions of Office that don't cost that much. And certainly there are many, many companies that are, are not going to decide to increase the amount of money they give to Microsoft by $30 per user per month, then I think just inherently, if that's the price, it might not have as much impact on the world, even though Office is potentially one of the richest tapestries for AI having impact. So I think that tech companies are gonna have to figure out how to do this in a cost-effective way. I imagine that they are assuming, probably correctly, that over time they can figure out some of those issues and, um, and run the AI machine in a way that is efficient enough that they can make some money at it. But that has— that isn't entirely in place.

Also, just in general, like the, the compute curve and the adoption curve, at some point that's just going to get unsustainable.

Well, um, I think they're— I think, yeah, I think they're hoping that by the time, um, they're not willing to lose money, that they will have figured out how to do it in a way that, that's, um, they can make money at. Um, they don't, they don't have all that in place, but I imagine there are a lot of techniques they'll be able to use, such as using more than one model and, um, using the most efficient model for any, any job that they are doing. I also just think in general that it's kind of remarkable how excited the world is about this, given that, that relatively little of it is actually out there, and almost every major tech company has been showing off stuff that they haven't actually shipped yet, right? Like, we've been, we've been thinking about these Microsoft tools and Google's equivalents for most of this year, but for the most part, those actually have not been deployed yet. And so it's a little bit hard to say how useful they will be or exactly how they will impact how work gets done until large numbers of actual human beings are using them, which by the end of this year will have happened, but, but it's not the case yet.

Yeah, it sounds like, yeah, there's a— there's sort of the economic issue, uh, potentially, uh, leading to a decline, as well as just the not delivering on a promise issue. Because like you say, there's all this stuff that's been announced, all this stuff that we wanted to be awesome. What if it's not so awesome?

There's a lot of vaporware out there. To think back again to the '90s when, um There were all these other competitive landscapes in tech where companies would show off stuff that was not ready, uh, in a lot of cases to get in the way of other companies that were showing off other things that weren't ready. And we're seeing an awful lot of that with AI. Like Dropbox yesterday at their conference showed off a bunch of things there that are in alpha, so they're not even in beta yet, but, um, they wanted to get them out there because they're competing against all these other companies which are are also showing off comparable things.

Yeah, like when Google came out with Bard earlier this year, and I think it still has the word experiment on it, which was sort of an unprecedented public-facing word for them.

It has a disclaimer saying something like, this tool may give you information which is inaccurate or offensive, which is not the kind of thing in the past I would've expected a company like Google to ship as quickly as they did. They did it very shortly after Microsoft announced the new version of Bing. And, you know, briefly people were saying that, that, gosh, maybe that there, there is this epic shifting moment in search where Bing could really take some share away from Google based on having embraced AI more quickly.

So let's keep our crystal balls out for a minute, but let's also for just a moment like adjust our skepticism a little and maybe loosen that journalist journalism fedora for a second and sort of try to see where AI delivers on some promise. But now we're predicting out further, say 5 years, some of this stuff really works. It ends up changing our information ecosystem in some way. Like what— basically that's what I want to know. Like what, what do you think the media landscape even looks like in 5 years if AI is changing both how we produce and how we consume information. You know, you think about these multimodal, uh, things like Artifact even, where it's like Snoop Dogg's reading you the news. And like, well, what, what does that look like when that's times 10 in 5 years? And it's, it's like whoever you want, however, whatever length you want, in whatever format you want, and brought to you whenever you want. Uh, and how, like, how do you even produce stories anymore, uh, when, when that is possible? What do you think?

Yeah, yeah. I mean, um, first of all, I hate predictions, and the best way to never be wrong is to be very careful about what you predict. Um, but that said, I mean, I think that even if you are a company like Fast Company, um, that is not in a huge hurry to do as much AI as possible We live in a world in which there are going to be tons of media outlets that do do a lot of AI content, as well as kind of content farms. Remember in the old days of content farms, they, they were trying to do vast amounts of stuff that was generally created by people who they paid very poorly. Well, today they can do 100 times more content and not have to pay any human beings. So I think that will be part of the context in which media outlets operate. I think that consumers of media are gonna have to get even more savvy than they are now about what's out there and what they can trust and what they can't. I think it will probably have some impact on just on how content is discovered in ways that are a little bit hard to predict. But in the early days of the World Wide Web, media brands created these beautiful homepages, and we kind of expected that people would come to our homepage and read it kind of like you do the table of contents in a magazine. And people do that some of the time, but, um, after we built those beautiful homepages, it turned out that search engines were a much more dominant way for people to find content. And then a few years after that, social networks came along and people found an awful lot of what they read because other people were sharing it. And, um, I don't think I am smart enough to understand exactly how AI is going to impact that, but, but I imagine we'll see further waves of change and how people discover stuff that is worth consuming. And, um, I think there are some people who are kind of optimistic that there might be ways in which media brands leverage that to their advantage to get a little bit more control over this experience. But I think it's probably equally possible that we'll have even less control over how what we do is discovered. Um, I do think that there's going to be value for things created by humans, uh, forever. And it's not like we need to abandon what we do in, in terms of, of hiring people like me to write things. And, um, I mean, we have our, our last cover story was about Microsoft and AI. And, um, I talked to Satya Nadella, I talked to a whole bunch of other people within and outside of Microsoft. I got a lot smarter by talking to them. I was able to put this stuff into my story. We are not yet at a point where, where AI is going to negotiate an interview with Satya Nadella, although I'm guessing probably it probably could come up with some pretty good questions even now. But there just is a lot of value to specialness created by people who know what they're doing. And I think there's a chance that some of that material might look even more special if it's in the context of even more vast amounts of material. That is not all that special because at least right now, AI is not good at making stuff special. It's good at putting together something that's somewhat generic.

Mm-hmm. Yeah, definitely seeing that.

Although again, I mean, there are things that haven't happened yet that could be really powerful, such as what if AI starts to generate stories on the fly purely for me based on what it knows about what I care about? we could see a point at which, if that happens, stories that were created to be read by a critical mass of people in the, in the way that journalism is now could start to look a little bit old-fashioned.

Hmm. That's an interesting thought. I like that, how you transition that into you as kind of a news consumer and what, what sort of see the potential there. This has been great. I'm going to close with the question that I, again, I hope to ask, uh, everyone I end up talking to, but what is the one thing everyone in the media business should know about generative AI?

Well, there are probably 100 things, but I'd say the most specific thing is even before you start to understand it as a journalist, start to understand it as a consumer. In the same way that if 30 years ago, if you were one of the first people to get online, just because you found it interesting and useful as a human being, you were in a much better position to leverage it as a media person. I think the same thing is true with generative AI. So, so kind of don't think of it as this theoretical thing that might radically change your livelihood. Think of it first as something that's gonna change how people get and consume content. And if you do that, you'll be much better positioned to understand its impact on your livelihood.

Outstanding. Harry McCracken, thank you so much for the conversation. This has been great.

Thank you, Pete. It was a lot of fun.

Where can people find you on Twitter or on the internet? Assuming you don't, wherever the, where do you want them to find you?

I'm still figuring that out. That's a whole nother, Tafek. Uh, on X, I am still Harry McCracken. Uh, on Mastodon, which I'm also having a lot of fun with, I am Harry McCracken at mastodon.social. On Threads, I'm Technologizer. And, uh, who knows where else I am these days or where you'll be able to find me. Uh, and, uh, whatever I say now might not apply in a few months. But I'd say Mastodon and Twitter/X are still the best places to find me, as well of course as Fast Company where you can sign up for my newsletter Plugged In.

Got it. Thank you for correcting me on the Twitter/X thing. I keep slipping.

I'm still living in 2022. Actually, I call it Twitter and I refer to tweets. I kind of refer to buy into that particular branding change.

Sweet. Well, you have been listening, listener, to The Media Copilot hosted by me, Pete Pashal. You can find me on X @petepashal, but you can also follow The Media Copilot there at The Media Copilot. Uh, please subscribe to the newsletter at mediacopilot.substack.com. Thank you for listening and be seeing you in the future.