How AI enables and limits creativity with futurist Tomas Haffenden

The creative industries are undergoing a rapid change due to AI. Non-creatives are using Large Language Models to generate images, video, copy, and music. So what can marketing and social media professionals do to stay one step ahead, regardless of whether they're a creative, strategist or generalist marketer.

In this episode of Soc•AI•l Skylines, creative strategist and Tomas Haffenden demonstrates to Gary and Christina how creativity and AI isn't just as simple as "just getting AI to do it." In this episode, Tomas talks about:

• The importance of linguistics in prompting.
• What are super tokens.
• How to generate the outputs from AI that match what's in your head.
• Where AI ends and humans have to take over.
• Where to start if you're looking to upskill on AI.

Credits:

• Soc•AI•l Skylines is a Social Hills production.
• Hosts: Christina Vetta and Gary Andrews.
• Producer: Peter Northcote.
• Videographer: Imogen Mabin.
• Guest: Tomas Haffenden.

Transcript

SPEAKER_03 0:08

Hi and welcome to Social Skyline, a Social Court Production. I'm Christina Miller.

SPEAKER_00 0:13

And I'm Gary Andrews. In each episode, we'll be cutting through the noise around AI. We'll be speaking to professionals who are using it day in, day out to get their insight. And this episode promises to be really fun and fascinating as we take a deep dive into the world of creativity. We're speaking to senior creative producer at Telstra and futurist as well, Thomas Haffenden. Thomas, thank you for joining us. Thank you for having me.

SPEAKER_04 0:37

So you're a regular voice on LinkedIn about the capabilities of AI. So what got you started working with this kind of technology?

SPEAKER_01 0:47

Shouting into the void of LinkedIn, yeah. I think I have always been attracted to disruption. And I think as this technology developed, I recognized how disruptive it was likely to be to creativity in general, as well as resurfacing some really interesting, I think, kind of philosophical questions around what do we mean by artificial? What do we mean by intelligence? What do we mean by all of these different things? Um and so I just started talking about them and then didn't shut up. And then people seem to be interested in some of those opinions. So that's I think where it came from.

SPEAKER_00 1:22

Nice. Um, you know, talk about disruption. Obviously, the creative industries are an industry that are getting seriously disrupted at the moment. Um, is AI gonna replace creatives entirely?

SPEAKER_01 1:33

Uh I think it will replace creatives who don't use AI, I think is what it's likely to do. What AI is capable of is supercharging what's already been being done, but already what's been done, the creative process in every element. But I think there are some inherent difficulties and challenges in there because whereas we had experts in each field, we're now having this real blurring of who can do what, or perhaps more who should be doing what. And so people who are not designers are suddenly designing, and I'll use that term fairly loosely. Um, and I think that creates, I think it kind of comes back to what I was saying about how are we defining these terms, and it's uh yeah, I think it's a really exciting space to be tracking.

SPEAKER_04 2:16

So let's get a little bit deeper into AI and creativity. So, in your mind, how do the two coexist together?

SPEAKER_01 2:26

Um, I think if you look at the way in which also let me make it more personal, the way I would create something is go from some random idea that is in my imagination, so it has no real tangible way for me to share it, other than kind of I guess linguistically and try and describe it to you. Um, but to get from that idea into some sort of realization, something that's been executed, um, normally requires it to come out of you onto something. And I think what AI, particularly kind of large language models and the ones that allow you to generate imagery and stuff like that, allow you to do is to bridge that gap, which might seem really simple, right? Like cave paintings, you know, like drawing pictures in the sand, everyone has the ability to move from imagination through to something that's a bit more tangible. Um, but I think that the skill gap is huge. I know people who literally can't draw at all. They can't, you know, they're like like this. And I'm like, that doesn't look anything like it. And so what what this technology I think is allowing people to do is to, or those that want to use it, is to make that jump between what's in your head in terms of ideas, imagination, into something that you can share with other people and interrogate and explore. And that um speeds things up. I think the I guess the inherent danger is that it it can it doesn't mean necessarily that the things you're about to produce are of any quality. I've used the phrase in my writing quite a lot, tsunami of shit. Um which I think is better than AI slop. Everyone start using the sea of sameness, yeah, or the sea of sameness, right? Because the the power of this technology is its ability to iterate at volume, but I think the thing that will always be uniquely human, which is why I don't think it ever replaces creatives, is the curation. It's looking at the hundred things it's managed to spit out for you in a few seconds and saying, right, those ones are shit, that one's quite good. Let's go a little bit further. Um, and I don't think it has that ability to do that, it doesn't have that selection quality.

SPEAKER_00 4:25

I feel that what you're saying about not being able to draw, there is hope for me in pictionary yet. Um my elephants might not look like cats.

SPEAKER_01 4:33

I'll tell you what, that is an absolutely fantastic idea. That to someone someone take that up because the idea of having some sort of AI pictionary where the game is for you to not be allowed to use the exact word in the prompt that you give AI to create the image. That's fucking great. That's right, that's my or at least our idea. I don't think that's an absolutely great that's a what a fantastic idea that would be. Because that kind of brings onto something that is a real interesting limitation, which I know we we talked that we were going to talk about, is linguistics.

SPEAKER_02 5:04

Right?

SPEAKER_01 5:05

This this stuff is incredibly powerful, and you're absolutely right. There's nothing to stop me as a non-coder coding an app. If my brief is tight enough, if I understand what I'm asking the AI platform and the model, whatever model I'm using, to do. And I've got a great example of this, right? Is that I would consider myself reasonably well read and reasonably well um kind of positioned to create things. But I had this idea that for a uh it's probably about a year ago now, so there's other ways you could have solved this. But I wanted to create an image with sunglasses evenly spaced on a square thing. So maybe 20, 30 sunglasses, all different, evenly spaced. Now I'm sure if you conjure that and you and it's a top-down white background, very simple. Hopefully, anybody watching or listening to this is like, yeah, cool, I've seen those, right? When they do the backpack ones, Belroy, you know, and that everything's beautifully, neatly. I don't know who unpacks like that as opposed to emptying it out, but whatever. Um, neatly put out. And I was like, right, I want to do this. So I said, blah, blah, blah, evenly spaced, blah, blah, blah, all different sunglasses. Came out, they weren't evenly spaced. And for me, I was like, no, no, no, no. I need it, it needs to be evenly spaced. Anyway, I don't know, 20, 30, 40 minutes maybe of going round and round. I'm like, for fuck's sake, this is this is mental. I could have, it would have taken me less time to get on Photoshop and do it myself. Um, and then randomly, probably a week, maybe two weeks later, I happened to be looking at something and I was like, oh shit, someone's done the thing. And I discovered it is called knolling, right? K-N-O-L-L-I-N-G. Now, why I think this is really interesting is I went back to that system and said, I want a uh I want sunglasses, 20 different pairs of sunglasses on a square thing, nolling. Did it in the first time.

SPEAKER_04 6:50

Whoa.

SPEAKER_01 6:50

Wow.

SPEAKER_04 6:51

Everyone out there, nulling.

SPEAKER_01 6:53

You're very welcome. Right. Now, what I think that illustrates, which is a real is the strength and limitation of this lit large language models and and this way of communicating with them, is that even those of us who consider themselves reasonably eloquent are limited by our understanding of language and linguistics and how we use it. And and in particular, as famous Wittgenstein thing is, you know, it's not in the meaning, it's in the use, because the way we use language isn't necessarily what language means. Prime example is the myriad uses of the word shit to mean something good or something bad, right? Depending on how you use it. But these large language models are trained on an increasingly large history of correct languages. So for we could spend all we could spend the next hour trying to dissect the difference between beautiful and stunning. But the reality is, I would go as far as to suppose that we probably use them interchangeably. I don't think I can think of an example where I'm like, well, that's not quite right. I I I can use them interchangeably.

SPEAKER_04 7:56

I feel like stunning is that just that step further than beautiful, though.

SPEAKER_01 8:00

Do you think stunning's a bit more? Okay, yeah.

SPEAKER_04 8:02

But I I get what you're saying.

SPEAKER_01 8:03

But the reality is that the language models know the exact differences. And so I think that's where our inherent limitation is being able to ask for what we want. Yeah. Which feels really counterintuitive, the idea of saying, the idea of saying this system can basically do anything you want, from code to paint a picture to create a video. There was something I looked at today, Kling O One, and released, I think their USP is the fact that you can use that same approach to edit videos. So you can give it a video and say, I want to do the following things and you see from behind. It has the context.

SPEAKER_04 8:42

So we could punch this whole thing into write the whole thing out.

SPEAKER_01 8:45

Done. It shouldn't be underestimated the requirement of an understanding of linguistics to get it to do what you want. I think it's very easy for us to do it at a base level, you know, write me a hundred-word story about a dog for my kid. Easy. But but to push through to a to a to a level of specialism, the nulling example, requires something different. And I don't think we're I don't think necessarily we've found a way to to bridge that gap. So nulling and prompting what like what is the difference between is specific uh is is I guess could be referred to as a a su a super token. Okay. Right? So everything you w when you're prompting something, each word you said give uh loosely, each word you put in is a token. Okay. Whereas there are certain tokens, because of the way the words use, is is highly specific. So you could see it as a super token, it's almost worth more than the word the because you're you could probably write sunglasses nulling and ignore all the other words and it probably would be able to do it. Nolling's the style is the specific style of laying things out with that exact spacing viewed from above on a blank background. Wow. And so whereas the prompted but the prompt is that that's the challenge, isn't it? With the prompting, is that you know a lot of people's you know, shit in, shit out is you hear a lot of that. But I think that's too simple. The reality is I think the where people are struggling and where people are quite rightly being frustrated is they're like is they're thinking, I can describe this task to my intern, for example, and I know they can bridge the gaps to get me the thing. Whereas I'm using the same words and I'm not getting there. Um I think the result the result of that will be net positive because I think it's forcing all of us to say unlike a an intern situation or potentially even you know a friend or a colleague or whatever, where they make a mistake, you might then be like, Oh, that's fine, I'll compensate for that. Where you're using this technology and the iterations are so fast, I think it is forcing us to be a bit more self-reflective, to be like, right, it's only me and the thing. So I'm probably to blame. Why didn't it understand? What am I missing? The knolling example again, which means that if you're capable of doing that, I think you can start to become a better, more developed, more intelligent human being if you can get past the frustration of it not doing the stupid glasses picture that you wanted in the beginning.

SPEAKER_04 11:07

I just have one one more point on that, um, that I just want to get your opinion on. So, do you think that it's kind of like a pattern recognition thing? The more you do it, the more you start to understand what works, what is what the system is responding to the stronger?

SPEAKER_01 11:25

Definitely. I think that sort of the ch the challenge is as well, I I guess yes, yes, to answer it. I think the more you use a model, the challenge is we're it's nice. A lot of people are saying AI, right? As as this umbrella term. And again, the reality is there isn't AI, it's not one thing. Even AI itself, I think, can be pulled apart and is a difficult thing. The reality is if you're using OpenAI, ChatGTP versus Gemini Nano Banana with your creation of images versus you know Notebook LM or any of those other ones, they all have different characteristics, strengths, and weaknesses. And that, I guess, more if we zoom out, is another potential limitation of where we are at the moment is that you need to invest the time to keep up to date with the hundred new things that are released almost daily and recognize the strengths and weaknesses of each, which is kind of means you end up building a I guess if I think about my own kind of day-to-day use, I've kind of got a team of AI, is how I would probably see it, where I'll use one for one thing, then I'll move it across to something else, then I'll move it into something else to do a slightly different task to kind of try and get the best of all worlds, I think.

SPEAKER_00 12:40

I mean, it's that's actually segues quite nicely into what is going to be our final question, which is yeah, a lot of people are probably still in the very early days. Some people like yourself are a lot more advanced. So where would you say, you know, somebody who is just going, right, I still need to start upping my knowledge of AI? I I've kind of got those ideas in my head that are starting to be creative. I realize that I can get to a point where, you know, those I I can't design, but I've now got that ability to do it. Where should they really get started? I know that's a very big open-ended question.

SPEAKER_01 13:17

Okay, so I'll answer this in two ways, right? Firstly, I would say practically. There's uh AI Advantage is someone I would absolutely follow. It's a guy called Igor on uh YouTube. He does uh his sort of tag is AI You Can Use, right? So he's talking very specifically on demonstrations of how to utilize the technology as opposed to oh, there's nano banana point five or whatever. Um another guy's uh Futurepedia, another guy on YouTube who I would also suggest following. Um the reality is if you want to keep up, I probably invest 30 minutes to an hour, two, three, four, you know, sometimes every day, trying to keep up, and I don't think I am. So I would say let yourself off the hook trying to keep up. More practically, I think the biggest demonstration is in your exploration of it and and committing time to explore. I think the best way of doing this, so this is my little tip for anybody next time you're in the car uh having a drive, get Chat GTP out, um, turn on the voice interaction part of it and have a conversation. Which sounds really rudimentary and stupid, but you will be blown away by how quickly you you develop a rapport, what feels like a rapport. Now it's all synthesized, obviously, but how quickly you can start to explore a problem. Oh, what I I'm not sure what I want to eat tonight, for example. Start with something super simple, and you will have a full conversation with this technology that starts to allow you to explore things that perhaps you wouldn't do. I think sometimes it can be a bit frustrating to go in with the idea of a single task and then feel frustrated that you don't you can't get it to do the stupid sunglasses thing. Whereas I think kind of starting more informally, like a relationship, um, and just having a bit of a chat is really interesting and and and the developments in that that voice interaction is will blow you away.

SPEAKER_00 15:10

Wow. I mean, there's so much to unpick into there, and I think everybody who's watching this will get a hell of a lot out of the ideas that come from this. So if they want more insight from you, or just you know, pictures and of random sunglasses. What's the best place to follow you?

SPEAKER_01 15:29

Uh LinkedIn um is probably the best place uh to follow me, I would say. Um uh or just reach out to have a chat. I think I'm always interested in in you know the direct contact with people and how they want to use it. Um I'm always if I've got the time, I'm always willing to talk about it.

SPEAKER_04 15:44

So thank you so much, Thomas. That was that was awesome. Quite a lot. Noling will be looking more into that one. Thanks everyone for tuning in today. So you can find more episodes of Social Skylines on all podcasting platforms, Social Hills YouTube channel and the Social Hills website, and our LinkedIn page, of course. So don't forget to follow, like, and subscribe.