Revolutionising Product Development: The Impact of AI

Rory Madden at UXDX EMEA. Video: https://youtu.be/AjDkX-9SCqc

Readable transcript: edited from the recording's captions for readability (fillers and false starts removed, punctuation and section headings added). Wording is the speaker's own. Timestamps are positions in the video. Names marked [?] could not be verified against the audio.

The forum format and who is using AI

[00:00:00] Rory: I hope everybody's enjoying this forum format. Again, just to reiterate, the whole intent of what we're doing with this is twofold. Yes, it's content, but you could go out onto the other stage, where people are giving you more direct content from talks and things like that. The purpose of this one, as I said, is twofold. It's great that we see these charts. You can see what other people are doing in the industry, you can see what the same problems or different problems are that people are having. But the second real reason for this is networking.

[00:00:33] We're about to break for another break after this, and one of the challenges that everybody always talks about is: I want to go to a conference to review my processes, check out what competitors are up to, upskill, but I also want to build my network. But then it's really hard. Networking's hard, networking's boring, and when I'm standing there it's much easier to whip out my phone and do something on my phone than go up and talk to somebody. So the idea here is to look at the face of somebody when they're asking a question or answering a question and go, "Okay, yeah, that's something I have in common," or "That's something I can go talk to that person about." That's what this whole stage is about: making that networking a lot more accessible and a lot easier for you after this, to go and talk to somebody instead of sitting on your phone, which is the much easier option.

[00:01:29] Anyway, with that aside, this one is going to be about AI, and AI in the workplace. I might delay another minute or two, or I'll get started. We're slightly ahead of time, which I think is the first time when we've been running a conference that we're ahead of time. I'm going to ask a couple of questions. It's the same format that you're used to at this stage. I want to see who is using the latest AI tools, and we're going to dig into what I mean, because that's such a broad term. Are you just playing around with it in your own time, or have you started adopting AI tools directly in what you're working on, or are you still a laggard and you're like, "I don't know what that's all about yet"? Obviously AI is such a crazy broad term, and we're going to dig into that in a minute.

[00:02:28] Okay, I'll give everybody just a few seconds. The QR code is on the side, so you can pop that open.

[00:02:40] Okay, so what I might do is start with the no's, because I actually want to understand: is there any reason? Is it that you just haven't gotten around to it, haven't found a need, or is there an ideological reason, or anything else that's stopping you from playing around? Does anybody want to raise their hand and start off, who has said no? Okay, we have somebody here.

[00:03:09] Audience: I think one of the main reasons is that I would be putting company data into generative AI tools, which I'm definitely not doing.

[00:03:21] Rory: And has that been a company-wide announcement, "Do not use this, they're sucking in all our data," or is that more each person deciding?

[00:03:30] Audience: I think I just saw companies that did it and messed up. Employees in other companies did it, and then data got released, or their source code got released. Anyway, I try to be a reliable employee. The company begged us not to do it, but we can still do it. Nobody's checking our laptops and saying, "Hey, you went to ChatGPT and you're using it with our company data." Nobody's going to do that, so it's on the employee not to do it, not to release personal data. But the AI we created internally is not as good as the ones in the other tools.

Using AI at work, and the fear of data leakage

[00:04:11] Rory: Brilliant. We're going to jump onto that internal and external bit in a second, but let's jump into the yes for work. Here we had an example where there's that fear of data leakage, of not knowing where that data is going to go. Who is using it for work, and do you think about that problem of data leakage? Can we get a few examples of people who are using it for work? Some hands. Okay, we have somebody here, thank you.

[00:04:40] Audience: Yeah, the company is actually really supporting AI for work. We are using a couple of solutions from Microsoft at the moment, with a small group, just to see how everybody deals with it. And it's amazing, because it helps you even to demonstrate your findings, so for researchers to create the deck to present to the stakeholders. It's very, very powerful, it's good. We also use AI for some of our solutions as well. At the moment we are working on IVR and a chatbot using AI, and we have almost 100 people working with AI at the moment in the company.

[00:05:22] Rory: Awesome. And is that company-wide? Is it the opposite of being told don't use it, is your company actively saying use it, speed up?

[00:05:30] Audience: Yeah, with moderation, because we are a regulated company, so data leaks are something very serious for us. We are really compliant, and data protection is a huge thing for us. However, we found ways to use AI, with Microsoft solutions for now.

[00:05:51] Rory: Okay, so you're trusting the vendor that it's not going to get it. Okay, brilliant. Does anybody else have any cases of how they're using it for work? Any hands? Okay, we've got two people here. Sorry, we'll come to you next.

[00:06:15] Audience: Yeah, I'm using it for processing quite a lot of qualitative data from interview scripts and focus groups. I've been putting samples into ChatGPT and asking it to pull out themes and then relate quotes to themes, for the purpose of user research reports. I share the same concerns as the gentleman over there, so we have to try and sanitize the data before we put it in, to make sure there are no names being mentioned or any sensitive information. But by and large it's mostly generic questions about use of technology and use of products, so it's mostly fine, but we do need to be careful.

[00:06:59] I would say the results of the method are mixed. Sometimes the way it relates quotes and themes isn't wholly accurate, so you still have to go back and check everything and group your themes appropriately and do your due diligence with your research. But it's definitely a time saver as a first draft of the research process, for pulling out themes and getting quotes. It's very useful, and it saves a lot of time, I would say.

[00:07:24] Rory: So even with the hallucinations or the problems, you still find it faster, even having to go back and double check it, than if you had to do it in the first place?

[00:07:34] Audience: You can notice the errors pretty quickly if you have a keen eye for the way quotes are, and especially if you've done the interviews yourself. You can immediately see that that quote doesn't quite relate to that theme or whatever. And even if there are a few errors, no one is going to give out to you too much about it. They're just the same errors you could have if you were doing it manually. So yeah, I found it very useful.

[00:08:02] Rory: Just pass it to the lady here.

[00:08:05] Audience: Hi. Well, you kind of just stole mine as well, but I'm coming from the UX lecturer side of things. It's really interesting: we're trying to teach our students how to use AI, but discouraging them from using it for their assignments. We're basically saying the same thing: even if you do use it, you have to double check it anyway. I'm not sure we're quite there yet, but it is very interesting to see the applications for it so far.

[00:08:31] Rory: Okay. And is there anybody who's been flat out mandated that you can't use it? So it's not even just trying to be protective about it, but you just can't use the tools.

[00:08:48] Audience: Not with my current company, but in the past, yes, because there was a data leakage, so they closed it and we were banned from using it.

[00:08:55] Rory: And was there anything like the data sanitization that might happen before?

[00:08:58] Audience: It was a code leakage. They just pasted the code inside, and it was a protected one, so of course there was a problem with that. The company was pretty big as well, so that's why it got cut.

Which tools people use: language, image, video and embedded AI

[00:09:14] Rory: Okay, brilliant. Let's move on to the next one, because AI is quite broad and nebulous, and just as we're chatting here I realized I left out an incredibly obvious option: code generation would be a very obvious option here. But what are the tools that you're using? We've talked a lot about large language models, your ChatGPTs, your Claudes, your PaLMs. You've got your image generation, so that's your Midjourney, your DALL-E. You've got your video generation, so Runway, Synthesia. Embedded AI: I've separated out internal AI from embedded AI. Internal AI is something that you've built internally that you're using in your own products, versus embedded, which would be like Notion, Office 365, as was mentioned here, or Intercom. There's a great talk coming up tomorrow on how Intercom built Fin, so do go see that. Or other, so code generation would probably pop into other. I'll just give people a few more seconds to vote.

[00:10:24] Okay, I'm curious: have people heard of or played around with video generation at all? It's just because that's the one that's on zero. Show of hands, who has heard that you can actually create videos with prompts? Okay.

[00:10:40] Audience: Yeah, I work in the pharma industry, and a lot of the time they like having a professional voice on a live website. But the reason why we haven't really gone for it is a disingenuous feeling from looking at those videos. So while it was a nice idea, we haven't really proceeded with it.

[00:11:05] Rory: Yeah, there's still that uncanny valley that you get with the videos. It's close, it's really close. Has anybody seen the new metaverse Lex Fridman interviewing Mark Zuckerberg? A couple of hands. The metaverse is everybody's favorite joke, but it actually looks like a good product now, so do check out that video. It's photorealistic, and they say the uncanny valley has disappeared. It is like you're talking photorealistically to another person. If it really has gotten rid of the uncanny valley, that's an incredible achievement.

[00:11:47] Okay, let's talk about the embedded AI. We had Microsoft here. Who else is using it, and can you give me a couple of case studies of what you're using embedded AI for, in Notion or elsewhere? Pretty much every app now has their AI products. Who has a case study that they're using? Nobody? Were you scratching your head or putting up your hand? Okay, let's talk about the image generation then. Oh, sorry, did somebody...

[00:12:25] Audience: Just an example of an embedded AI that we've been working on. In Google Maps, when you have a shop, a small business for example, we found that people don't usually spend a lot of time describing their business. So we just added a magic button that uses the information that we have about you to create a really nice description of the business. And what we found in A/B testing is that the businesses love it, and the descriptions are far richer than they would have written. So it's a very cool feature.

[00:12:55] Rory: Awesome, that's a good example. And that kind of embedded AI, the Intercom chat: I don't want to give too much away, but the statistics on how much time they're saving for a customer service rep when they're using AI to summarize your history. You know that very annoying thing when you get a call: "Give me a few minutes to catch up on the chat." They just automate all that into a nice little summary, and it saves a huge amount of time.

Image generation in practice

[00:13:20] Okay, image generation, because we probably have a lot of designers in the audience. Are people using this just as a hobby, or has anybody actually gone to the point of using it as part of your job? Okay, we have two people here, so we'll start here and we'll come to you.

[00:13:38] Audience: I can start first. We tried to use it to make fancier storyboards, because the stakeholders are bored with the cartoons, and that backfired big time. It didn't work at all. But now, if I have to do some PowerPoints, sometimes I just ask, let's say, "Hey, I want to see a yellow Tesla in front of a red building." I can't find that image on Google Images. It would take me 5 or 10 minutes, but through Midjourney or something it looks pretty reliable, and even stakeholders don't pick up on it. So it helps me get my work done faster, and you don't have to worry, there's no risk, to be honest. That's what we use it for.

[00:14:09] Rory: Is anybody else doing that, just quick presentation slides?

[00:14:16] Audience: Hi, yeah, I can talk a little bit, not about my work but about my personal project. There was a time I wanted to create a logo for the Twitter account for my personal project, so I tried to use some image generation. I just mentioned the key elements and the name, maybe used this keyword. However, the generated logos were not something I was really looking for, so I ended up with something simple: we just used the name as the avatar of the Twitter account. But I think maybe at some stage I can try again, because the AI engines are evolving all the time, so maybe at a later stage they will give me something much better. I think the key reason I didn't use it was that I gave it a certain word and I wanted that word to be part of the logo, but not enough of the actual English characters were there. It just ended up with some weird characters, some of which looked like English characters, and it just didn't read right. But I want to have another try after a period of time.

[00:15:38] Rory: So there are some prompts, I guess, that you could learn that it's perfect for, and there are some things, English, where I think they're getting pretty good now, but definitely in the past it would just give you hieroglyphics or something. Interesting.

[00:15:53] Is anybody worried about that from the role perspective? Webflow just announced their AI, and ChatGPT said: do a scribble on a piece of paper and we'll do a design and code it up for you instantly. Is anybody actually worried about that at all, or do you all see it as a positive that's going to help things?

What people are excited about, and rising expectations

[00:16:17] Well, let's jump on to that question. Let's go with what you're most excited about. What do you think is going to be the benefit of this? Is it that you're going to have increased capacity and speed, like the example here where you know it'll take me 5 minutes to find that, but if I just go and do it, it'll be done in 30 seconds and I know it'll be better? Increased quality? Unlocking capabilities? What I mean by this is that a lot of people always complain there's too much work for design, there's too much work for research, there's so much work that work gets skipped. So by having it become easier or quicker to do, you're unlocking more people to be able to do it. Or it'll replace work altogether. Nobody thinks that they're going to be able to sit on a beach sipping mai tais while an AI does their job for them? No.

[00:17:18] Okay, just on that last one: who has noticed, in the companies that are using it, that the second you started using it and started doing more, suddenly the expectation of what you were expected to do increased as well? Yeah? Has anybody got an example? I know personally I now think things take less time to do, because I know how quickly they can be done in ChatGPT, so I just expect more work to be done in the same amount of time. Is anybody... sorry, we'll come to you next.

[00:17:59] Audience: Mine is really quick. I would recommend, if anyone's using it and they're saving time, don't tell anyone, because people will expect you to do more work. I haven't told my boss that I'm using it for what I'm doing, and I'm going to keep it that way.

[00:18:13] Rory: Mic's behind you.

[00:18:16] Audience: We are using it for the research piece with the team, so now other stakeholders are asking for more research and user testing, because we can present amazing results with dashboards and graphs and everything, and it cut something like 25% of our time on the research piece. So now even on medium and small epics, everybody's like, "Oh no, let's do user tests." And it's like, okay, you need to step back. It's not for everything that we do. But it's really useful and more precise as well. It gives you more data quality, in my view, for now.

[00:18:54] Rory: Yeah, I think the hiding will work for a small period of time. I think it's going to catch up with you, but hide while you can. My opinion is that that expectation is just going to grow. It's the unlocking capabilities: more people are going to ask for research because it's quicker and easier to do.

Quality concerns and simulated research

[00:19:15] Increased quality was lower. What are people's main concerns? Are you seeing a challenge with quality? There was the Midjourney example, where it's not good at English letters and characters. Is anybody else worried? There was a question over here.

[00:19:35] Audience: I guess it depends on the industry that people are working in. If you're working with complex products, I don't think it will increase the quality of work, because what you have to input there is already so complex that you're losing more time thinking about what you're inputting, and the output is not there. So I don't think it's increasing the quality. It depends on the industry that you're in.

[00:19:58] Rory: And is that a challenge of the input window, the amount of data that you can shove in? Do you think in a year or two's time, when you can put your whole company's data stream into it, that will solve it, or do you think there's always going to be a challenge when it's too big and too complex? I'm seeing a nod.

[00:20:22] Audience: I'm really not sure.

[00:20:25] Rory: Because there's one out there, and I'm guessing I know what the reaction's going to be: simulated research. Instead of going and talking to customers, because that takes time and we know we have to go quick, why don't you just simulate customers and run your research with simulated users? Who thinks that would be a good idea? It would definitely speed things up. Okay, we have some people saying it'd be good, so let's go here. The guy in the corner there.

[00:21:02] Audience: I work for an insurance company, and we have a very important persona on the claims side called claims adjusters. They handle the claims, and it's a complex job, and it's hard to get time with them, but over a period of three to four years we've done hundreds of interviews with them. I don't think this is unique, but we're taking all that research data, and you can build almost an internal LLM of all the questions and answers that you've got from them over the years in interviews, and you can create a real-life persona. If you're on a project and you need to do some quick research, you can pump questions to that, especially for development teams, who might otherwise make assumptions about how someone works. You can actually ask this digital persona complicated questions, and it'll answer based on research that we performed. So you get better answers, and it's better than the assumptions that some of the teams make. There is value in it. I don't think it replaces research entirely, but this dynamic persona approach is definitely something where we're seeing some value, for speeding things up and bringing research data together.

[00:22:07] Rory: Okay, is anybody else seeing value? Before I ask somebody, because I'm assuming one or two people are itching to give a retort. Anybody else see value? Okay, who wants to say why that's an absolutely terrible idea? Does anybody feel strongly? Okay, we have somebody over here. I think it's great if it's working, but I'm just trying to get a bit of tension.

[00:22:29] Audience: If it adds value, great. But fundamentally, humans are individuals, and we have variability within ourselves, and the whole basis of LLMs is that they go to the mean. There's a huge concept within psychology and behavioral science of regression to the mean, and when you only look at that mean, you are losing the variability that is important to understanding the human experience.

[00:22:55] Rory: Is it a question, though? If you are saying it's regression to the mean, well, it's the Pareto principle. I'm solving 80% of the problem. Yes, we lose the bits of the 20% on the outside, but I've got the chunk of it. Is that not better than no research?

[00:23:15] Audience: I actually think bad research is worse than no research, personally. And we have seen for hundreds of years that when we design for the edge cases, universal design, inclusive design, and we do research with those populations, we fix things for everyone else and make things better for everyone else. The curb cut effect. Closed captions are used by people to learn languages. They're used so that people can watch TV while their spouses are sleeping, or while their children are watching one thing and they're watching another and not being bored by Barney for the 50th time. There are lots of these cases where the thing that was designed for a very niche population makes life better for everyone. And if you don't have that variability in the data you're collecting, then you are never going to discover those things, and you are leaving out a huge part of the marketplace as well, if you are only designing for or doing research with the general population.

[00:24:19] Rory: I agree with your points. But you're pitching a business case, or you're pitching for some funding, and somebody knows there's this tool that will simulate our personas, because some company has gone out there and ingested a huge amount of data, and they say, "We've got the perfect personas for fraud, we've got the perfect personas for all these."

[00:24:41] Audience: I'll go back to the first presentation this morning: you have to validate that data. It's fine, use it as a starting point, but you'd better do a couple of interviews to make sure that that is the actual data out there and is truly representative. Because I have found with lots of things on ChatGPT and things like this, I'll ask it a question and it gets it completely wrong, and these are things that are verifiable. So you have to check where that persona data that's going in there came from, and check it to make sure. You can reduce the cost by checking with fewer people, but I'd argue that most of the time we recommend at least starting your research with six people and seeing what happens from there. Six interviews shouldn't cost you that much money.

[00:25:32] Rory: Yeah, it's a great point. I agree that small numbers, even six, can give you directional insight. Not validated, but directional.

A hybrid future?

[00:25:43] Rory: Who thinks the future is going to be more hybrid? I don't think manual is the future. I just don't think we can put the genie back in the bottle. Management are going to know we can do things quicker. You can hide for a while, but not forever. Who thinks the future is more of a hybrid, and where's the balance shifting? Is it more 50/50, 80/20? Does anybody have an opinion on what will happen in the future? Did you give him the mic? No? Okay, here.

[00:26:17] Audience: Yeah, I really like the idea. I think it's going to be hybrid. I think there's going to be a class of problems that are incremental UX improvements, and you can just throw those at the machine and it'll give you a reasonable signal on what you're doing. And then I think that will actually allow us to put a lot more of our human intuition into more pivotal shifts, maybe in product or whatever we want to achieve.

[00:26:42] Rory: Anybody else have thoughts? I haven't heard anybody from this side in a while, so we'll start here, because I'm feeling I'm standing over there too much.

[00:26:51] Audience: I actually agree with your point very much. I think it's going to be a hybrid. We've already seen it. This isn't really in UX, this is more on the video side, but for voiceovers it's extremely popular, not just on social media: being able to change the speed of someone's talking, or being able to change their accent, depending on where you're producing content, or just to make it sound more like who you're giving it to, a customer. I think it helps a lot, even for myself. It's not perfect by any means, but I think it's improving. For instance, for an iPhone or an "i"-related product, you have to actually type out "eye" as in eyeball for the "i", which is very funny. But I think it's definitely improving, and I think it's going to be half and half, and it's a tool that we can use to benefit our UX process.

Net win or net loss, and accountability

[00:27:45] Rory: Okay, before I jump on to what you don't like: who thinks it's going to make things better? Because everyone here has said that there are efficiencies and there's unlocking new capabilities. Who thinks all these AIs that are coming out are a net win? Ooh, not much. Who thinks it's a net loss? And then we'll go to the undecided. Net loss, anybody else? Okay, who's undecided? That's 95% of you then, because nobody went either way on that. Okay, so everybody's sitting on the fence at the moment on this one. Sorry, you had a point that you were going to make.

[00:28:25] Audience: I would say if we talk about the tool, it's definitely brilliant, but if you look from a business perspective, we're totally missing any accountability right now. If I'm doing research or anything and it goes wrong, I know it's my responsibility, but I can't tell my boss, "Well, the AI told me so," and you can't fire AI. So when everything works, the happy path, brilliant. But in the real-life scenario, if it doesn't, you can't blame it on a system, and we haven't thought about that whatsoever.

[00:28:48] Rory: People blame stuff on "computer says no" all the time. It's a famous one, "computer says no."

[00:28:56] Audience: If I mess up a lot, I'm made redundant or fired. But what can you do if you put millions into a product because it told you so, and it fails? Who will be held accountable?

[00:29:04] Rory: Hands up here: who has worked on a project that has utterly failed and not returned the value that was expected? How many people got fired? Okay, we got half of one. It happens today. In one company I worked in, they spent a billion pounds sterling on a brand new "we're going to do everything" transformation and wrote it off because it failed, and not one person left the company over it. It was over 13 years, so maybe people left.

What people are most concerned about

[00:29:45] Okay, let's just jump quickly into what you are most concerned about, because everybody was 95% sitting on the fence there. Is it that it's going to replace work? "Okay, I'll do a scribble on a piece of paper, now I don't need a designer, because it's going to create my designs and code up the website for me." Is it "computer says no," loss of autonomy, where it's just "the algorithm said this," and those really annoying situations where you're just frustrated because for some reason something is biased against you? Increased surveillance: we've talked about data leakage, but there's also the facial recognition stuff, the biometric recognition, all that kind of stuff. Surveillance is very easy to do. It's great if somebody has good intentions, but what if a bad actor gets hold of these things and does bad things with them? Or it's going to kill all of humanity, which is on the extreme end of things.

[00:30:48] But wow, 25%, or 20%, of people are concerned it will kill all of humanity. I don't know. I'd love to hear some thoughts. My personal opinion is that it's a little bit alarmist, but every new technology has alarmists, like 5G is giving you cancer and all this kind of stuff. But I don't know, maybe I'm being super naive, so I'd love to hear. Let's go with the kill humanity, just because it's funny. Who has an opinion on this? Okay, over here.

[00:31:29] Audience: I'm not talking like Skynet or anything, but I think, especially using it for research and stuff, we're just going back to doing assumptions again. That's not research, that's just an assumption, which hasn't worked before. Also using it for illustrations of things: artists and illustrators have been trying to get paid for their work for years and years and years, and been pushed back, getting exposure instead of actual money, which is incredibly frustrating, to see a computer doing it in no time at all. But the main thing I think is that it's taking time away from you doing something, then you do more work, and we're just making more money, and that is capitalism, and that is ruining the world. And it's already happening at the moment.

[00:32:06] Rory: Okay, I've got to just separate two things there. The blacksmiths who made all of the shoes for the horses that got replaced by the cars: was it a bad thing that they went out of business?

[00:32:18] Audience: No, but you can also adapt skills to do that.

[00:32:21] Rory: Yeah, you can adapt skills. I know I'm intentionally trying to be antagonistic. But if you can adapt skills and you can get rid of the stuff that the AI can do in seconds, then the work that you're doing should be a different type of work, higher value work, more interesting work. Has it been? I don't know. Has anybody, with their newfound time, had more time? When we talked in the last one, it was about product vision and product strategy. These are the things that product managers want to do, but probably backlog management is actually what we do. Where's your percentage of time? Because I'd say it was more backlog management than it was on the vision and strategy stuff. We have lots of people with questions. Let's go here, front row.

[00:33:11] Audience: Mine's really short. I'm horrible at voice acting. When I had to do the voiceovers, I sometimes have a lisp, and it's just due to the fact that I had braces that expanded my mouth. But because of that, I was able to use the AI to produce a piece of content that I was proud of, and therefore I could focus on how the video wanted to feel, what it wanted to express, what the tutorial was based on. So it didn't limit me, if you get what I mean.

[00:33:44] Rory: Yeah, excellent. And you had a point here. We'll get back on to kill humanity again in a second.

[00:33:50] Audience: For me, I'm a product design manager. I don't know, but most people who are doing management are probably still doing some IC work as well. It's a bit of a hybrid role in the industry. What I've done with what we call Liberty GPT is it has helped me almost get back to a level as if I was a full IC, but I'm still doing all my line management duties. So it's been hugely helpful for me, massive, because it's changed my role essentially.

[00:34:20] Rory: And just to the point of: okay, so you're going to be, not so much overburdened, but you're going to be expected to do a lot more now. So you can be an IC and you can be a line manager. Are people just going to keep getting more and more burdened?

[00:34:33] Audience: Yes, I guess that's the balance that they don't like. It's the thing of keeping it a bit of a secret, it's a balance of that at the minute. The synthesis of research has been sped up massively because of it, and they're not aware of that yet, you know what I mean? But at the same time, the impact I can have is greater for it, and I'm not stressed out from trying to do extra work.

[00:34:52] Rory: Do you find that it's actually freed you up to do some higher-value, less hands-on, just-getting-stuff-done work, and more thinking about it?

[00:35:01] Audience: Yes, yes. Head space, definitely.

[00:35:04] Rory: Anybody have... okay, somebody over here.

[00:35:11] Audience: Going back to the creative points: everything that is in AI is derivative. It is stealing from somebody else's work and not compensating them for that work. So eventually we're going to get to the point where all we're doing is derivatives of derivatives of derivatives, if we keep going with just AI and don't compensate the original creators for creating original work.

[00:35:32] Rory: Is any work original these days?

[00:35:35] Audience: Yes, there is original work.

[00:35:37] Rory: Is it not derivative, because an artist learns...

[00:35:40] Audience: Everything is derivative, but there are gradations of derivativeness.

[00:35:47] Rory: So those spiral AI images that have become super popular, where they create these weird spirals, I don't know if you've seen those. It's an optical illusion where you can put faces on images, and there are all these optical illusions you can do. Is that not almost generative? It's a new form. It's not something that people were creating beforehand.

[00:36:06] Audience: But it had to be based on photos or something beforehand, that somebody else, a human, put work into and didn't get compensated for.

[00:36:20] Rory: Yeah, I'm purposely being antagonistic.

[00:36:23] Audience: But back in the Dark Ages, not dark, 100 years ago, 50 years ago, the goal was to create automation so that we would have more time to do creative arts. Most rich kids don't go into labor jobs, they don't go into line management, they go into creative pursuits. So if artificial intelligence is taking away the creative jobs, that is killing humanity.

[00:36:55] Rory: Okay, so we're out of time, and I don't want to end on there. There are some good examples here. We have a last comment, so I'll give it over to you here.

[00:37:06] Audience: Yeah, thanks. I just wanted to add my concern, which is actually not on here, but it's related to kill humanity, maybe a watered-down version. I'm worried about adding to systemic bias or deepening divisions, because we don't have the policy in place or the tools to actually vet the data that's being input. I'm worried that we're just running off with it so quickly that it might cause more harm down the road.

[00:37:38] Rory: Yeah. You know the way AI is better at chess, it's better at Go, it's better at all these games, and it's better at strategy games now as well, where you have to bluff against humans. So it can manipulate humans better than humans can manipulate other humans. It's interesting what kind of Pandora's box we could open with that. But let's just go back one slide: what are you excited about? I think there are a lot of cool, interesting things happening. Yes, there are a lot of risks. We'll see what will unfold over the coming months. It seems to be that every six months there's another big leap forward. Thank you for joining. I hope you found it somewhat interesting to hear other people's points of view, and enjoy the rest of the day. Thank you.

Closing: what Rory is most hopeful for

[00:38:30] Host: Thank you, Rory. Don't run away, we're closing, it's the end of the day. It's easy to have visions of doom, just as it's easy to have unrealistic hopes for the use of a new technology, any new technology. Think about it: you've seen Oppenheimer, right?

[00:38:49] Rory: I haven't. Oh God, spoiler. I have a kid.

[00:38:52] Host: Spoiler alert. So what are you most hopeful for from AI in our profession?

[00:38:59] Rory: What I loved about the internet was the opening up of data and the democratization of that, which used to be so siloed. What happened was it did a curve of opening up and then closing again. I think AI has the option of opening up a lot more opportunities to people who don't have opportunities. So my only thing is: what can we do to keep it going like that, instead of it closing into the hands of just a couple of big companies?

[00:39:28] Host: Amazing. Please, let's give a round of applause for Rory.

[00:39:32] Rory: Thank you, thank you.