Joel Bailey – The Role of AI in Social Media Management

Joel Bailey – The Role of AI in Social Media Management

Podcast Power of Ten AI Service Design

My guest in this episode is Joel Bailey, product and service director at Arwen.ai, a four-year-old marketing technology startup that uses AI to help brands to manage and moderate their social media comments at scale.

Joel brings 25 years of experience working in service design roles, leading change and building new stuff for a diverse range of organisations. We talked about his journey into Service Design, Service Design and AI, navigating AI and social media, the future of design and much more.

Note: Given the fast pace of developments in AI, I should point out that we recorded this in June 2025. I had a big backlog…



Timestamps

00:00 Introduction to Service Design and AI
04:50 The Journey to Service Design
10:00 Understanding Service as an Outcome
15:08 The Role of AI in Social Media Management
20:11 Challenges of Moderation on Social Media
25:32 The Evolution of Business Models
26:36 The Role of AI in Startups
29:13 Navigating AI and Content Moderation
30:20 Understanding AI Bias and Ethics
31:43 Engaging with AI in Social Media
35:11 The Future of Design and AI
40:21 Creativity in a Fast-Paced Environment
44:56 The Importance of Service in Design

Transcript

Note: This transcript is machine-generated and may contain some errors.

Andy Polaine (00:09) Hello, welcome to Power of Ten, a show about design operating at many levels of zoom, from thoughtful detail through to transformation in organizations, society and the world. My name is Andy Polaine, I’m a design leadership coach, service designer, educator and writer. My guest today is Joel Bailey, product and service director at Arwen.ai, a four-year-old marketing technology startup that uses AI to help brands to manage and moderate their social media comments at scale.

Joel brings 25 years of experience working in service design roles, leading change and building new stuff for a diverse range of organizations. He says if you slice him down the middle, he’s 30% total service nerd, 30% AI and design fanatic and 30% family man.

Joel, welcome to Power of 10.

So that 30%, that’s like a sort of slice of Brighton Rock, is it? There’s different sort of…

Joel Bailey (00:57) Good to be here.

Yeah, do you know what? That is exactly the image. Although I think my family would have something to say about only getting 30%. But it depends. Maybe that’s time-based. Obviously not love.

Andy Polaine (01:13) Okay, very well done. look, tell us, I’ve talked a little bit in the intro about your background, we actually don’t know how we know each other, but we’ve been in each other’s orbit certainly from the service design world. Tell me a bit about that journey to where you are now.

Joel Bailey (01:32) Yeah,

I was trying to remember exactly how we connected. I think it’s one of those ambient relationships that just came out of the ether, right? As you say, in similar orbits. I think it is around service design. So I was very lucky, I think, to stumble across service design very early on in my career. And I’m not a trained designer. I have an arts background. ⁓ I didn’t do any sort of

qualifications or certificates in design. But I’ve always had a creative lean. And I think when I started working and I was working in organizations where I was basically being asked to solve problems, which is what lot of people get asked to do. If you’re not doing admin, you’re solving problems. And I was pretty frustrated with the tool set that was available, which was very kind of expert led, best practice, best best practice that. And I think it was…

in one of those roles that someone said to me, what you’re going towards is like service design. Because I come from a digital background and I remember, I called myself an information architect at one point. I used to work across A2 size bits of paper mapping out websites and things, right? And then I started looking at user sensitive design. If anyone remembers the Rosenfeld book with the polar bear on the front, the information architecture, but legendary stuff.

Andy Polaine (02:57) I was about say publisher.

Joel Bailey (02:59) All right, exactly. So now we’re feeling old, right? So these books were quite seminal to me, and it was very much making it up as I went along. And then I got to work in, I went to work as a contractor consulting in government in the UK, and I started doing what is now known as user-centered design. But back then, we were just winging it. It was like the logic of the project I was doing, which is working on one of the big government super sites, as they were called.

This one was businesslink.gov.uk for anyone who cares. It’s basically a website by the government to businesses. And I learned that the best way to create the outcomes you wanted was by asking users what they needed. And the worst people to ask about what a small business owner needed was a civil servant at that point, because they were like, well, they need to out Form 20348. And it’s like, well, do they really, you know, they might have to by law, but that’s certainly not what they want to do. So that’s when I first got into user-centered design.

And then we fast forward, I guess, a few years and somewhere along that road, someone said service design. I remember going to the Wikipedia page and it was like two lines, know? it was like, service design is applying user-centered design to a wider scope of organizational activities, all different channels, multiple touch points. I was like, okay, that’s kind of what I’m doing. Because we had telephone lines we were dealing with, we were dealing with email, we were dealing with webpages, PDFs, all that sort of stuff. So it was like, okay, so this is a good space to be in.

And I guess the rest is sort of history in that I got very invested in it. I really got excited about service design. I’m really excited about two things, the design approach, which I’m sure we’ll talk about, which I still use, and service and what services, which I’m, as I said, a bit of a nerd about service and what it is and how it works in my spare time.

And now, if I come all the way to now, all these things kind of come together in this very unexpected way. So, you know, I’m running a business, I mean, but they call it product, but it’s a service business. It’s got a big digital product in the middle that uses AI to detect and remove toxic commentary from social media. But it also uses the same technology to, you know, have conversations with social media users on behalf of brands. So there is service happening there.

We as an organization are providing a service to our clients and we have to design it continually because our users have very high standards. Anyway, that’s very short, a long answer to your question, but that’s kind how I found my way. And along the way I’ve had quite formal service design jobs. worked at LiveWork for a while and everyone who knows service design knows LiveWork. So was very grateful to be to work with Labyrinth and Ben and the others over there. ⁓ And I’ve led service design.

in large organizations and small, but now I’m not really doing service design as a job for the first time. I am applying it in a startup, which is so different.

Andy Polaine (05:57) Yeah, well let’s come to why it’s different actually in a second. mean, it’s interesting for me, you know, talked about the information architecture thing. I find sort of people go one way the other and sort of encounter services. One is they’re working into interaction design and UX and they’re like, there’s sort of more than this. There’s a larger ecosystem or I want to be more involved. If only they had come to me earlier.

in that process and I can input it in the strategic part. And so they kind of do that zoom level out. I mean, this is why the show is named after that Eames powers of 10 film of the different levels of zoom. Or they go the other way, which is, I’m sort of systems thinker. I’m a kind of ecosystem thing. And I think the IA thing is, well, it’s actually a good mix of the two, but there’s definitely that kind of overview of how does all this kind of hang together and what is the kind of.

the view of this and you sort of gone that way. So I can kind of see where it meets in the middle.

Joel Bailey (06:49) Yeah,

there was a third option, was, I’ll just stay in the website UX space with my headphones on, looking at screens. There were a few people that did that and I did think about that at one point. But I’m just too curious and I think ask too many questions. And then once you start asking about the call center, like what happens when people can’t do this thing on the website? Then you end up going to the call center and talking to the manager and asking them questions. It’s just very organic. But that systems theory there.

It’s fascinating to me and I’m not a systems theorist. I’m a systems thinker in the very basic sense and I’ve read enough to be dangerous probably. it’s increasingly, you spend longer and longer in organizations of any size. You have to understand systems theory and you have to understand that you can’t change one thing without some ramification elsewhere. So I’m very sympathetic to the systems world. It’s complicated.

hard to deliver.

Andy Polaine (07:49) It’s complex, isn’t it? It’s interrelated. We just said the thing about you know, you’re working there and they call it a product, but I call it a service. you know, my endless rant that everyone’s bored off by now is that, you know, everything’s a service and certainly anything that’s digital is. mean, I always say if you’ve ever used turned on Slack and the, ⁓ you know, the service is not working or it’s, you just get a load of placeholder boxes, how you know.

You are using an avatar to a service and anything else same goes for your banking and almost everything that is digital.

Joel Bailey (08:26) Yeah, and the way to frame it in my mind, I remember doing presentations of FS, Financial Services audience, and it was all about, we trying to create better mortgages or better homeowners, right? What are we trying to do here? And in our organization, the product, sorry, the service we’re after really is we want to move our clients from one place to another, from a less good place to a better place. And the better place is

They’re just more engaged with their community, right? That’s what R1 is all about. Our mission is to make your social media more social. So that is all about more engagement. The product, one of our products, is a tool that auto-moderates all the horrible stuff, the spam, the racism, of which unfortunately there is still a bit too much of it on social media.

So for me, it helps to think about service as an outcome. And it’s about movement from A to B for anyone ⁓ in life. And I take it very seriously, whether it’s service leadership and how I’m trying to work with others to get them from A to B in their lives and careers, even with my family, right? It’s like, what can I do to serve this person to move them forward? With our clients at Arwin, it is very important.

to not think that we’re a rubbish removal business, because we could do. But that really limits your opportunity for that relationship with that client. When you say, yeah, we do rubbish removal over here, but actually our service is about getting you to better engagement, whether it’s on your ads or on your organic. As a service for me, it’s such a useful framework for that.

Andy Polaine (10:00) Yeah, think, know, nine is it’s a broadening of view of the ecosystem because it includes more than just the product. Which also obviously every time you kind of broaden the lens, I mean, there’s a danger you lose yourself, but there’s also this idea of, there’s a different way of, we could come at this from a different angle as well. And that’s often where some of the innovation lies.

Joel Bailey (10:19) Yeah, you know, as service designers, we’ve all been invited to do innovation as much as just improvement, but you can’t innovate, you’re right, without a bigger ambition and service.

Andy Polaine (10:31) But I mean, but conversely, can’t, know, services I had got, I think it’s probably for the reputation of some of the agencies that certainly are the big consultancies that I might have worked for, of delivering sort of concepts and, you know, and journey maps and blueprints. And then the clients run out of money, go, here you go, good luck with that. And obviously, you know, the real thing, reason why our book is called From Insight to Implementation is because

You know, how things are implemented makes such a massive difference. I mean, you’ve talked about businesses filling in a form or something. You hear this stuff about you. Whenever I explain what I do to people and that’s a whole thing in itself or what service design is, I, you know, I then immediately get this barrage of, you the other day I was trying to do X, Y and Z and this, and it’s, it’s tiny things. It’s always the gaps and the transitions where things go wrong.

and screw people up. And you mentioned that sort of flow chart of, know, person does this and then they go to that. the arrow, innocent arrow or the hairline between those two boxes on the flow chart contains a whole lot of horrors when it goes wrong.

Joel Bailey (11:41) Yeah, and think you and I, we both had a very similar visual metaphor for this, I think yours is crossing the chasm or something like that. Yeah, the crevasse. Now I had a blog post a few years ago about crossing the chasm, I think I called it, which is that that line between two boxes on a process map actually goes across cultural boundaries or organizational boundaries or team boundaries or geographical boundaries. And in that, all sorts of things get lost, know, lots of tacit knowledge.

Andy Polaine (11:48) Yeah, the Cravat.

Joel Bailey (12:09) And that’s what service designers do is they’re trying to stand in the user’s point of view and say, how am I being carried from A to B? It looks so simple here from a business analysis point of view, but the reality is I’m suddenly being handed to someone or another part of the organization who doesn’t care about me as much as that first bit. Like sales, they really cared about me. Onboarding, they didn’t care about me so much because I’m now just a product to be processed through onboarding. Anyone who bought a new utility. Yeah, yeah, you’re done. We’ve won you.

And we’ll come on and talk about startup and how it’s different, but ideas implementation is obviously massively ⁓ shrunk down in my world. We can have an idea and implement it in hours, which for much of my career was just not possible because the raw material of my work was other people. So I’d have to work with other people to develop conceptual ideas.

and then they’d have to work with other people on committees to make change happen within the organization to make that concept a reality. We obviously don’t have that. We are 15 people and we work very quickly to take an idea into reality, however quick as we want to or not.

Andy Polaine (13:19) I, like you, probably have had large enterprises say, treat us like a startup. We want to be like a startup. Well, so what does that mean? And a startup, one of the things is it’s not like, we’re really innovative. And so that’s kind of what they’re hoping. But one of the things of that is a startup has often key people are wearing many hats, right? Or at least a couple. You have such a small

group of people, the communication lines are very, short. Literally, I’m the person next to me or whatever, even if they’re remote now, there’s only a handful of people and those people are empowered to make decisions, right? Because it’s probably their thing anyway. And so when you say enterprise client, you want to be treated like a startup, are you happy to get rid of most of your decision making and governance process that gets in, well, not governance, but kind of ⁓ gatekeeping process that gets in the way of that.

Can we just have direct lines to someone who can make a decision? And, you know, can we try stuff out and be risky? And they’re like, yeah. So we’d like to be a startup, but you know, not those things. Yeah. And so then they always fail. So I really like to kind of come back in a minute. I’d really like to talk about the design and AI thing, but just go back and tell us a little bit more about what Arwin does. Because I think it’s really interesting. I want to ask a few questions around that.

Joel Bailey (14:42) Yeah, so I mean a lot of people just kind of sigh a little bit when they hear the words AI at the moment and we are in that high cycle dip, right? And I get that. First thing I’d like to say is that Arwin, we have been AI from start four years ago. So we’re pre-chat GPT. So it makes us really old, makes us wise like Gandalf, right? We’re pre-chat GPT. We’ve been doing this for quite a long time in AI terms, not obviously.

mean, AI is 30, 40 years old, right? But anyway, what we do as Arwen is probably easiest from the origin story from where it began. So the euros which are being played now, if we go back four years ago, we remember the racism that was thrown at the players who missed the penalties. It was kind of that point that Matt and David, CEO, CTO got together and said, we’ve got to solve this thing. We can solve this thing. Let’s solve this thing.

I came on board a bit later. And ⁓ what we built is a platform that pulls in data models and algorithms from all sorts of different providers. And then we plug this thing called Arwin into your social media profiles in an authorized way. We’re authorized by the different networks.

And then Arwin scans every single comment that you get on social media. And if it finds something that breaks your rules, and we set those with each client, so everybody wants to get rid of spams, or porn bots, and crypto bots, and all the fraudulent stuff, which can just grow.

Andy Polaine (16:14) 90 % of the internet

Joel Bailey (16:16) My name is certainly is then. The

networks obviously have it worse than others. And obviously if you’re a brand and you’re trying to connect with your audience, either on paid or organic, that stuff really gets in the way. It’s a pollutant, pure and simple. So we remove that and we do it by studying both the content and we have probably the world’s largest database of spam items in the world. And we do fuzzy matching. So we don’t do like exact keyword. We don’t use keywords at all actually.

We do fuzzy matching. Basically, Arwen’s saying, how close is this to every other spam item I’ve ever seen? It looks at the comment and it looks at the profile. So it can look at the profile’s avatar. It can look at the different characteristics of the profile to determine and score how likely that is to be a spammer’s profile. And then it looks at the content itself and makes a decision. It does all of that in under a second. And then it hides it if it goes over a certain threshold. So it’s…

That’s the simplest way of describing it. Now we do that for spam and that’s like our biggest seller because everybody wants to get rid of it. But then we also do it for unwanted content. So that is what we call in the UK lawful but awful. Okay, so the networks are pretty good at getting rid of illegal content and that stuff is really the unpleasant.

Andy Polaine (17:32) Because they have to.

Joel Bailey (17:33) Yeah, because they have to in different regions in different ways. And that’s a colossal job and they do a very good job of it because we don’t see things like child abuse on social media. So they are doing a very good job. The awful but lawful bit, because it’s bit loose and fast in ⁓ legal frameworks globally, even our online safety bill in the UK is a little bit fuzzy about how to deal with that. That’s where we step in. So a brand basically says, so we are a, we’ve got a number of Formula One teams who have very premium sponsors.

very large audiences, increasingly a female audience that they want to engage with because they’ve watched maybe Drive to Survive, the Netflix program, they’ve really engaged in this sport in a new way. And we’ve worked out through our analysis that these female fans are a small, they’re about 25 % of Formula One fans. They contribute about 80 % of the vibes to the community. They’re really positive. So, and yet they get all the toxicity. So then we start helping them to hide misogyny.

sexism, sexual aggression, all of which occurs in sporting environments, unfortunately. Gaming environments, we work in a bit as well. So it’s all about, and I go back to my earlier point about we are trying to improve engagement. The first step to improve engagement is to put some rules in place. We use the analogy of a bar, right? You walk into a bar and there people smashing bottles in the corner, shouting abuse, racist abuse, or ⁓ trying it on in really offensive ways with women, whatever it is.

We come in, I often think of that film Roadhouse, coming like Patrick Swayze, the old Roadhouse, we come in like Patrick Swayze, and we just tidy things up. Some people do get blocked from the community. We automatically identify those people. But everyone else just finds that you can’t say these things anymore. We help our clients write a clear policy, put it on the door. And then within three months, they have got at least on average 21%, 22 % increase in engagement.

and the number of toxic comments has gone down by around 70%.

Andy Polaine (19:33) Just because it’s just not a horrible place to hang out.

Joel Bailey (19:36) horrible place and no one wants to be you know there’s a real sheep mentality.

Andy Polaine (19:40) It’s the whole Nazi bar thing, right?

Joel Bailey (19:42) Yeah,

exactly. So you denormalize it. We talk about denormalizing toxicity, which is why we invest in speed. Let me give you an example about we don’t got a lot of behavioral analysis on this. So the behaviors of ⁓ a hater, a troll, and the behaviors of the sheep that follow them are quite obvious when you look at the data patterns. Someone will, so let’s say client post comment, post a beautiful picture of a football ball being kicked into a goal by a female football player.

someone comes in and comments and says, get back in the kitchen. We’ve seen all these, so I’m not describing anything. These aren’t my words, let’s be clear. So that someone posts that as a comment, as soon as that is seen by other people, it creates permission for other people to do it. So they come in and start saying, yeah, yeah, you shouldn’t be doing that. All of that. Because we get rid of that first comment and we hide it in seconds, often under a second,

Andy Polaine (20:29) I’m going to some football.

Joel Bailey (20:38) then it doesn’t create that wave of permission and we shorten the life of a potential toxic event considerably as a result.

Andy Polaine (20:46) I’ve got a question for you, which I would imagine like everyone else is thinking. Yeah. Obviously, you know, you do this for brands, right? Because whatever Mercedes, AMG, Formula One team don’t want to have that stuff next to their brand. And in fact, that’s one of the most sort of best ways to kind of pressure brands to shift the way they think about certain issues is to highlight that. Why don’t the social media platforms do this themselves? Why is…

X and everyone else Y or know, Substack for that matter, talking of the Nazi Bar thing. Why aren’t they doing it? Why? I mean, they haven’t hired you yet or is it problematic at the scale they’re at?

Joel Bailey (21:30) Well, they’re partly at scale. So it’s important to think about pre-publication moderation and post-publication moderation. Again, I would say they do an excellent job of pre-publication moderation on illegal content. Very good at finding illegal content and stopping it from getting out. That in itself is a colossal data management job, huge amounts of data processing going on.

Awful but lawful stuff that I’m talking about. There’s a few reasons why it gets really hard for them to do. First of all, they aren’t legally obliged to do it. Section 230 of the Communications and Decency Act, which some people will be familiar with in the US where they’re all based, basically says they’re not an editor. They don’t have to edit. They’re not responsible for what their users put online. And there are all sorts of challenges going on on both ends of the spectrum to that. Some states in the US are saying,

You can never moderate anything pre-publication. It’s against the First Amendment or right to free speech. Others are saying you have to to protect children. So this is playing out and has been since we started the business. that legally, if they start pre-publication moderation of the comments that I’m describing, then they are basically saying, well, we are an editor. And then they’re leaving themselves open, I think, to a challenge of like, well, if you’re going to do it for that thing, you’ve got to do it for these other things that I care about.

And then we get into the really hard bit, which is one person’s really toxic comment is not another person’s really toxic comment, unfortunately. Now, some are. I mean, I think I’m not going to do them on your lovely show because it wouldn’t be right. But there are some comments that I see that no one in the world would go, that’s great. Yeah, there’s nothing wrong with that. But unfortunately, you get into gray areas, which is why we don’t set any standards on our clients. We work with them to work out what is your graphic equalizer to profanity to insult.

And it’s different for every client because they’re different brands and they have different communities and they engage. So sport is going to be very different from children’s entertainment, obviously, different demographic. And you can’t expect one network to do all of that on behalf of everybody globally. It’s it’s nigh on impossible. And I put it, I posted about this the other day, actually. So I think, I think there is, there are a number of factors that prevent them from doing it. I also think that.

If you do go into the pre-publication moderation world, you really take on a responsibility. Clearly, as a network, you’re making your money out of attention, the eyeballs that see the ad. If you’re having to put a human set of eyeballs on every bit of content, every comment before, the business model just won’t stack up. You’ll be spending so much to check everything that you won’t be making enough out of the ad eyeballs. Human eyeballs, one side, human eyeballs on the other side, they’re both costing the same.

So I am very sympathetic to the networks because I think the other thing I would, I’m going to go one step further and say, we do expect social media to be a utopia when the rest of the world isn’t. Okay. Now I know we can look at the algorithms around ranking, although Metra has done some great changes that our clients benefit from, that positive interactions get higher, ranked higher than negative interaction. So they are working. But I think we’ve all seen that the drama.

is what the ranking algorithms go for. And drama is often where toxicity lives. So people are pushed into places where there is toxicity. we can’t, know, a lot of our clients, if you go to a store or an event or you hang out with the CEO, they probably all got security at the front door. They’ve got security in a drive that drives them around. If they fly or do something,

They have extra levels of insurance and care and protection on themselves and their buildings. ⁓ And really, that’s no different on social media. ⁓ So I think we’ve just got to be careful about the networks. They’ve got to make the human race just much nicer to each other. It’s really hard.

Andy Polaine (25:32) Yeah, I think it probably says the thing you’re talking about eyeballs says something more, I think around, I would suggest broken, certainly, or toxic or difficult business models. Right. You know, that’s the problem as you know, the cardinal sin of the internet is that it lacked micro payments and that created advertising or ad tech. Right. So then everything kind of follows on from there.

I’d like to move on a little bit. You talked before about something I want to circle back to. You said, know, obviously when I was doing other stuff, it involved a kind of group of people and the pathway from idea to trying something out was much longer, know, weeks, months. in some cases, lots of expense to an infrastructure. And now you are saying, you know, that’s much, shorter now. We can come up with an idea and try it out very, very quickly.

How much is kind of that because of the AI aspect of it or how much is it, know, what role is technology playing in there and how much is it to do with your culture as a startup?

Joel Bailey (26:36) Well, I mean, definitely the culture of startup helps. It’s all experiment, right? The whole business is an experiment live in the world. But if I look at any department, I say department, we don’t really talk about departments, but like we have, you we’ve got engineering and product, we’ve got marketing, sales, it’s all ongoing experiments. And the first thing you learn

when you’re working in a startup is that most of the books are wrong or they were right for a certain context, but they’re wrong for your context. And so you quickly realized that you needed it. The most important thing you need is an experimental framework. So I’m working with Tushar, who works in our marketing team on a range of marketing experiments, which is where I guess the AI also comes in, not our own AIs, but other AIs out in the world that we’re working with. So we’re working with a couple of different AI

products that allow us to not just to take our kind of experimental DNA of a startup, bring AI into that, and it then accelerates the experiment itself. So things like research, secondary research, primary research, less so, but people keep trying to tell me that avatars are going to do primary research. I don’t think trust is there yet for qualitative research, but hey, but quantitative research, it’s amazing.

I can get an AI to go and find out lots of facts for me very quickly and compile them. You’ve obviously got to look out for hallucination, but the thing you’ve got to remember in a startup and in any business is all you need is a hunch. It’s a Rory Sutherland thing here. He says, we’re not doing quantitative research for like Elsevia or some university or something. We don’t have to be 100 % right. We just have to be right enough or righter than the next person. And one of the things that

that I was blessed with as a service designer in previous roles was lots of good data, like qualitative data and quantitative data. I could go out and speak to customers. I could go out and do research projects that last weeks. We probably can’t do that so much anymore. I things have changed budget-wise. But that was a real luxury. In a startup, you’re very data poor, very time poor. ⁓ So the experimental framework plus AI allows you to overcome the time piece, but also bring in the data.

And you just accept the fact that you might get errors in the data because you’ve just got to be faster at getting to the hunch and then sitting down and going, it’s good enough, we’re going to push the green button and we’re going commit to this as an approach.

Andy Polaine (29:08) You’re not generating content, is where, you know, that can be…

Joel Bailey (29:13) Quite

well. We have a second prod. We’ll talk about that separately.

Andy Polaine (29:18) Well, putting content out in the world, that’s where it’s pretty problematic, particularly with AI as you know, you’ve seen this kind of weird thing going on where companies would fire an employee, a customer service employee if they said that kind of thing in a chat thing, but obviously they just kind of plug a chat bot in and let it go crazy and with some crazy results too and not great results for them.

But you with the moderation thing, can also, you know, this is a very well known thing that part of the problem that happens with moderation is if for example, a person of color is talking about experience of racism that they’ve had, gets flagged up because they’re using the words that the racists are using and it gets flagged up as content. And so those marginalized voices get further marginalized. How do you deal with that aspect? Because there is a, you know, it’s still quite, it’s very, very complex, obviously. I think there’s a lot of the…

there’s just a black boxy, we don’t really know what’s going on. Even the people have kind of put it together, you know, once it’s got up and running. But how do you balance that?

Joel Bailey (30:20) You’re absolutely right. You know, this is one of the risks of AI’s who’s built it how they built it on what data has it been trained etc I mean we just distinguish between LLMs which will come on to And how they work and how they’ve been trained versus the various models that we’ve been using Pre-charge GPT and continue to use really effectively They have all been built and trained by experts in this field, right? So they are detection algorithms and they are benchmarked for bias and error rates

So around, you know, white supremacism, bigotry, racism, all these things. Okay. And I could go into quite a lot of detail about why we’re confident in how they make decisions. So, so clearly that’s something to keep an eye on. It’s, think when you get into the LLMs bit where it feels like a bit more of a black box, I can tell you how the other ones work and I can tell you how they were trained and how they were built.

Andy Polaine (31:12) So

there is some rickets behind that.

Joel Bailey (31:14) Providence and rigor. LLM is, it feels like, you know, they’ve just created something and it’s just gone, it’s just gone out to the world and hoovered everything up and it’s coming out with something quite useful, but is it entirely true and right? So we do use large language models and we do use generative AI. So I’ve talked about the moderate product. We have an engaged product. So like I said earlier, our service mission is to make our clients more social, make their social media more social.

So once you’ve got rid of all the bad stuff and that bar is now great and everyone’s hanging out, having a great time, now you want to talk more to your community. It’s safe, it’s lovely, it’s got all the people in there that you want in there. The most surprising thing to me about social media, so $270 billion got spent on social media last year, sort of globally. Someone came up with that number. It’s really big. And it roughly sounds about right when I hear about some of the social media advertising budgets.

You put that content out there, that content on social media will be responded to. You will get comments. Clearly, I’ve talked about the negative ones, the toxic ones. Negative is different. Toxicity, we get rid of it. But you’re going to get negative comments, you’re going to get questions, you’re going to feedback, people going, yay, I love this. The amount of brands that do nothing with those comments just astounds me. You spend all this money advertising essentially is trying to get someone’s attention. You’ve got their attention, they’re asking you.

something, they don’t do anything. It’s like the orphan child. Everyone’s all about, create more blog posts, create more blog posts. So we use large language models to help down at comment level. So if someone’s looked at your ad and says, great, where can I buy this watch? You can’t scale a team where you can, it’ll cost you a lot of money to answer every comment on every social media channel in real time. You’ve got to do it within an hour and to get into that buyer moment.

⁓ Arwen can do that. Now to go to your point about the fears that we all have about hallucinations and stuff like that. So the latest technology that we’re using, which is called retrieval augmented generation, lovely phrase, rag retrieval augmented generation basically says you’re a large language model. You’ve learned off all this stuff, some of it good, some of it bad. But I have a load of content here that I know is good. Yeah. It’s the company policy. It’s the brand voice.

It’s the rules and regulations. It’s the opening times of all the stores. They’re facts. And I want you to refer to those things when you’re making a consideration generating a response. So what Arwin does is it goes to the large language model and says, person, the large language model goes, this person’s asking a question about a watch. ⁓ And then the rag element looks at the rules and says, they’re looking at this watch. It’s this ad for.

this watch, yes, you can buy it here, here’s the URL and it connects them through. Okay. So what we’re trying to do with RAG and it’s a meta originated technology because they’re obviously fighting this kind of, all this crap firing across the internet from AI is like, we’ve got to put rules around it. You’ve to put control around it. And for us, it’s so important because if Engage is going to be having conversations with people online and delivering service, let’s be honest, that’s what it’s going to be doing.

then it has to be accurate or the lawyers will never sign it off because we’ve all heard about, I don’t know what brand it was, a car brand in the US selling a car for a dollar because the chat bot, they managed to get it to do that. No legal team is now going to sign off a chat bot AI. We’re not even rolling it out as fully automated. We’re like, it’s semi-autonomous. It’s a co-pilot. So our AI reads the question, reads the comment, whatever.

and then generates three alternatives and then the manager can go in and just choose which one they want to use. Okay, but the key thing is this rag technology allows you to put guardrails in so that the AI doesn’t just freak out and promise people like a new car.

Andy Polaine (35:11) So talking of co-pilots, I’m not going to talk about Microsoft co-pilot. As we are speaking, and this will probably come out in a couple of months, so as we’re speaking, Figma have just shown all their new stuff. There’s a lot of AI tools in there for design. ⁓ I had a dream, a frequent sort of daydream as a kid that I had a robot me, duplicate me that went into school and did lessons and all the boring stuff. I stayed at home reading comics and eating sweets. ⁓

The question is, it’s a part of me when it looks at some of that stuff that it can kind of very quickly generate. there was some stuff I saw the clip where it’s where you’ve got a whole, instead of having some lorem ipsum text, it just sort of generates a load of actual sort of placeholder text and all that stuff. And obviously it’s going to do a lot and sort of make prototypes really quickly, ⁓ go from some idea to prototype. And part of me is like, you know, that’s amazing. That takes a lot of drudgery away. And then the other part of me is, yeah, but.

Design isn’t just design, as in the end artifact. Designing is a process of thinking whilst you’re designing and actually that process of prototyping of, you know, even sketching on paper actually, there’s a talking to Eva Lottelam in some comments of a thing she did the other day, which is teaching people to sketch wireframes again in paper. There’s a lot of thinking that goes on in the designing.

And I really kind of worry about that. I worry less about the AI taking people’s jobs. So there’s clearly a lot of, there will be a lot of sort of low end product managers who go, oh, this is great. I don’t need designers anymore. I can just kind of write in a prompt and, and it will send some stuff to engineering, which is how they treat design anyway. Rather than I want to have this kind of trio and have this kind of good conversation and understand so that the good ones do that. So I can see the kind of.

the advantages of that, but I do worry about that side of it. What’s your view on it? Talking about that sort of going from idea to prototype very quick.

Joel Bailey (37:03) Well, we’ve both worked in consulting design, well, businesses and agencies. And I think we’ve all seen over the years that what is pushed out as original work is often just a repeat of like, you want a website? Well, we did that with one, we’ll kind of use a version of that. So I think for me, comes down to like, what is the job of inventing a wheel of actual invention?

is going to have to remain a part of the designer’s core skill set. I think there’s lots of reinvention of wheels going on out there and being sold as, well, you want a website, it’s 100,000 pounds. But we’ve done 14 of them. We’re just going to repackage this one. There’s a lot of that going on. And I think, to be honest, look at it, it’s same people sitting down. A website is now, I’m using that as an example, but you could take anything.

is a very well thought out, very well understood artifact object. clearly you want to make it right for that brand. But for me, we’re talking a sort of co-pilot logic here, which is get me to a good decent first draft so it’s all structured and labeled right. And there’s no like stupid mistakes where the site map just doesn’t connect and has broken ends to it. And then let me go in and add in how this is going to work for this brand and this particular use case. I just think that

Design has been doing a lot of recreating of things and treating that as creativity when not necessarily the case, when some of these things are already well-invented, well-understood, and it hasn’t really done us any favors. Equally though, the best briefs are the ones where, right, we’re struggling in this market, everybody’s things look like this, everybody’s things look like this. We need something that’s different.

Right. And again, it comes back to your point about just the client. I used to say, they say we want innovation. I’d say, okay, do you actually mean you want to invent something new or improve the thing you’ve got? Because if it’s improving the thing you’ve got, I do think AI has a role to play there. As I said about testing things, researching things, understanding things, coming up with options, potentially then designing that and the evolution of that thing. But if you want something brand new, like very inventive, I totally agree.

I wouldn’t say never because I just don’t know how this is going to change. But I’m with the lady who said, and it was a lady, think he said, I want AI to be doing the dishes and washing up so I can do the art, not the AI doing the art. So I have to do the washing up, right? I’m very much in that zone, but I do think we can’t help. But some of the drudge work, not the drudge work, because some people love doing boxes and AI and IA, sorry, and information architectures.

But let’s be honest, when I did information architectures, I always forgot about little bits. I have to go back and do them later. The joy of an artificial intelligence is it generally is complete, right? It doesn’t finish the job till it’s finished. So you’d expect me to be a bit pro at it, and I am quite pro at it. I think that designers have peddled a lot of design that isn’t really creative work. It is just taking something else and putting a slightly different skin on it, on patterns that are very well understood.

which is maybe a spicy take, but that’s the one I’m giving it.

Andy Polaine (40:21) So my response to that would be though, I think that is a systems effect of the environments they’re working in. In that the agile digital product kind of thing has been very, very, it’s been great for so many things, right? It’s really improved the ability of teams to get past all that sort of ⁓ waterfall stuff and big brands, enterprise brands that actually, most of my banks having had really, really awful.

kind of apps and stuff now have kind of reasonably decent ones and things like that. I, you know, that’s definitely because of that. On the flip side is I tweeted or I didn’t tweet because I don’t tweet anymore, but I tweeted the other day and I put it on LinkedIn, which is a sort of AI is kind of the wind tunnel of creativity, right? But you know, there’s some very famous silhouettes of cars out there, you know, small cars, sedans, know, SUVs and…

In the silhouette, they all look the same and that’s because they’ve all been designed in a wind tunnel. And at some point everyone optimizes for pretty much the same stuff and everything starts to look the same. In my experience, whenever I’ve been using a chat GPT or whatever it is, it gravitates towards that sort of mediocre middle and chat GPT therefore sounds always like kind of a management consultant because it’s like very average stuff spoken with amazing confidence.

And what I’m saying about design is I think in an environment where that was a sort of lean and agile thing where speed and velocity without much thought is always held up as the most important thing because it comes from that startup world. in other areas, it doesn’t make so much sense. When everyone is working very, very fast or under pressure to work very, very fast, they’ll lean into what the tools do well. And so they’ll use…

that particular what used to be, know, rave posters all look the same because they all had the same flash shot filter or, ⁓ you know, in the rise of flash and all the vector stuff. It’s all vector stuff because that’s what the tools did well. They all leave their signature. And now, you know, with Figma, you can always tell, you know, this has just been a kind of design pattern that’s been pulled down. And in some cases, obviously, the design system, which the designers are just kind of pulling on. there’s not really that, as you’re saying, there’s not really that much original work being done.

kind of feel like that is that’s not because designers come forward, it’s because they’re under pressure to operate really quickly. I worry that the sort of AI kind of say the AI Figma stuff just kind of perpetuates that and just makes it even worse.

Joel Bailey (42:46) I it will. I think it will. I think there’s two pressures here. One is like a sort of general capitalist, neoliberal pressure. You know, make it as lean as possible and scrub everything out. But there’s also a pressure the other way, which is people like familiarity, right? And do I want the copy explaining something to me about what I should do as a small business to be written by a copywriter? Or do I… It’s just instructional in that sense. It could be written… As long as it’s written in plain English for…

Andy Polaine (42:51) Yeah

Joel Bailey (43:15) the average age of 10, which is the average reading age for the most of the UK. And most corporates don’t do that. And I was surprised when working on copy stuff that people still write copy for people who are like, you know, reading age of like 15, 16, like the economist, you know, and people don’t, right? So you can tell an AI that I want you to write this for the reading age of average UK reader or whatever country you’re in. And it will do that. And it will do that repeatedly. And

Sometimes you don’t want creativity in there. You just want, and I’m talking about service stuff here. mean, marketing is different. Marketing, you’re trying to capture attention. You’ve to have something creative in there. I think creatives are reasonably safe in some of that stuff. I saw the Toys R Us little video that someone’s put out that saw a creative. I think there’s a way to go with capturing attention and delighting the soul of another human via AI. I think there’s a way to go. But a lot of the stuff we’re talking about is

Um, it’s, it’s, it’s just a lot of work that I used to do as service designer is like, well, we just got to get into the terms and conditions and we’ve got to get into the instructions on this page. And do we have to have those there? And if we are, can we make them really much simpler, please? And, I’d have to work with copyrights to do all that. And I’m not having a go at copyrights here, but I think everybody wants to bring creativity to a project that often doesn’t need it. Um, but we, we, we need creativity in the right place, right? I’m not denying that because your point about everything.

All those rayflies were the same until they weren’t because somebody said, hold on, I’m going to do something different. And I don’t think there’ll be any different here. AI already is making everything very generic. And so then something else will have to change. And it will always be a creative decision to make that change.

Andy Polaine (44:56) There

is a lot of very generic, mean, lot of the, I don’t know, I had like three or four websites open for different stuff the other day. And I looked at them and I was like, wow, these, they’re all using the same typeface. They all use the same hierarchy, you know, as it was all. It wasn’t bad. I wrote this essay a little in my newsletter, the last newsletter it was called, it was called On Mediocrity. And you know, it was like, well, it’s just H &M t-shirts, you know, everything’s like H &M and it’s kind of fine.

Joel Bailey (45:07) It wasn’t bad. ⁓

Andy Polaine (45:24) but it’s not very exciting either.

Joel Bailey (45:26) No, and I think that’s it, right? And a lot of organizations settle for it’s fine. And lot of managers will and workers will settle for it’s fine because it makes their life easier. If you can do eight hours work in one hour. mean, you know, people do gravitate towards that sort of, you know, decision making. I think that’s why AI is so disruptive inside organizations because you want the quality there and who determines what quality looks like when everybody has access to something.

like this. think AI within the organization and having proper guardrails on that so that people are creating good quality right for this brand, right for this organization, whether it’s internal comms or external customer comms, whether it’s in a service that’s outward facing or inward facing. there’s so many, you know, it takes me back to almost like the early days of the dot com, right? It’s like, this is all new. Where do we start? What’s important? And the perennials are still the same. Like if you’re trying to capture my attention,

You’ve got to be interesting and I’ve got to love something about this. But if you’re just trying to tell me to do something, just give me the facts for you. But in fact, get out of the way. And the promise of AI obviously is to get totally beyond the interface, right? Is that I’m just like, hey, ⁓ AI thingamajig, I need a holiday next year. You kind of know what my family likes. Come back some options. know, like we talk about body doubles. I’d like to send a version of myself off into the world.

as a gopher and it will come back and say, well, I found three holidays and they’re in this sort of price range. That’s very attractive. The question we might come onto is that what does that mean for a designer? Because in a way, I’d like to be a designer. could go off and do secondary research for the day as a body double, come back, tell me what you’ve learned. But I still want to be the person who synthesizes it and goes, oh.

Andy Polaine (47:15) That’s the thing. I think there’s a thing I read the other day, is, you know, what AI doesn’t have is sort of common sense discernment and taste. know, I think one of the things that does, and it’s interesting actually, because I looked up the etymology of intelligence today, because I was to make a snarky comment on LinkedIn. But it was, and it’s that, right? It’s pretty much that. And so, you know, artificial intelligence is such a sort of misnomer. I think one of the things that

what you just described does then highlight is the, differentiates really good ideas from the kind of craft dazzle, right? And I think that that sort of evens that out. We are actually coming up for time. So there is one final question, the one small thing question, is what one small thing is either overlooked or could be redesigned that would have an outsized effect on the world?

Joel Bailey (48:08) Okay, you talk about a thing. I’m going to go right back to the beginning of the conversation because it’s a bit of an attitude really. ⁓ I think it’s an attitude to serve that I’ve designed into all sorts of organizations or tried to, and I’ve certainly tried to do it in ours, which is service and service to others, whether it’s people that you work with, or whether it’s a customer that’s giving you some money for some exchange of value. But for me,

It’s a real anchoring concept. ⁓ I’ve been a service designer, I know service designers, but I don’t think most people really think about service as a concept and what they’re trying to do. But for me, it lifts my entire life, which is why it’s connected with religion. And I’m not religious, but I can see why that connects. So for me, if you’re doing any sort of design, you’re thinking about a service, don’t just think about the service as the noun, right? Think about the verb, the doing of it.

And what is the, what are you trying to move forward here? Is it, it’s not necessarily you’re trying to create a widget or a product or anything like that. It’s like, what is the outcome you want to achieve? Now that’s a bit abstract. It’s not a thing, but it is very meaningful to me that I am trying to serve the world in a better way. And therefore that anchors me down to what does this person need to move forward? I don’t know. I need to find out.

but it’s helped me through my design career. And it just so happens to be helpful because the majority of businesses in Western countries where I tend to work are service industry organizations. And I’ll be honest, they don’t really understand what that means, right? They have forgotten because of their scale that their service organizations and helping them to remind, and everyone knows, like I used to start my workshops and you probably do something similar by saying, right, I just want to go around the room. Just tell me the best services.

Andy Polaine (50:02) Yeah, service.

Joel Bailey (50:03) Or tell me the worst. Yeah, tell me the worst. And everyone knows it. Everyone’s like, my God, and they’ll never forget the worst one. They’ll never forget the best one. But then they go back into the service organization, so they forget it. So I think just rekindling that little flame, and I have it on every aspect of my life. I try to. There’ll be people listening to this going, John, he doesn’t do that for me at all. But I am trying. And I just think it really helps to create an integrity to the design process, to whatever you’re trying to do.

So that’s the one small thing I’d ask everybody to take into account. I think it can change the world, right? I think service is a, I actually heard Keir Starmer talking about it in his speech, a return to service. King Charles did it in his speech, all about service. No one’s really getting it. So I think it’s a communications issue. I got it. So I’m the one person going, yeah, I get it. Because I think it’s really profound.

Andy Polaine (50:53) That’s

very fine answer. Thank you very much. Where can people find you online? Arwin is arwin.ai.

Joel Bailey (51:00) Arwin.ai is where you’ll find our service and what we do. You can find me on LinkedIn. I’m probably mostly there these days because we’re a B2B software as a service business. That’s where we live. I used to have blogs and things. I am on X, I am on Instagram, I am on all these places because it’s my business to know how they work. But LinkedIn, otherwise arwin.ai or joel at arwin.ai.

Andy Polaine (51:22) Thank you very much. Thank you for being my guest on Power of

Joel Bailey (51:25) much. It’s been a pleasure.

Andy Polaine (51:27) You’ve been watching and listening to Power of Ten. You can find more about the show at polaine.com where you can also check out my leadership coaching practice, online courses, sign up for my very irregular newsletter, Doctor’s Note. If you have any thoughts, put them in the comments or get in touch. I’m on LinkedIn, my website, you’ll find it everywhere. You Google, all the links are in the show notes. Thanks for listening and watching and I’ll see you next time.