AI is already turning the world upside down. Robin Moore, Innovation Consultant with Media Cymru and Director at Shwsh, sat down to discuss innovation in the Welsh media industry, the place of AI, and some of the support available for people developing innovation. Watch the interview on YouTube or read the transcript below.

Transcript

Why is innovation so important to you?

I was very lucky after leaving university, and not being particularly employable, of discovering a new technology, which at the time was the World Wide Web. I fell in love with this idea of using new things and often for youngsters it is the way to a new career. I was quite lucky to then develop, both websites for the BBC, and find myself developing any new technology that came along and how we should utilize it for content creation. It’s very exciting because you can be very creative in the technical sector, and work with other sorts of visual creatives, performance creatives and storytellers, while being at the cutting edge.

It also is important because it allows us to control what that future will be. If we tend to try to keep within our old skillset and use the ways we’ve already worked, we often find the audience moves on without us. They will take new technologies on, they will change and we are forced, often by someone else doing it, to try and keep up. I think it’s much better to take hold of innovation and be in control of that future. And that means being aware of what technology can do, utilizing it and seeing how you want to use it, both in terms of how you create your content, but also in terms of ethics, and in terms of what’s actually good for the audience.

How is the Welsh media sector exploring AI innovation?

Much like most other media sectors we’re following the big fish. A lot of AI is delivered by the really big technology companies. It costs an awful lot of money. You need a lot of server infrastructure, and most importantly, you need access to a lot of data, because most of the AI we’re seeing at the moment is basically machine learning. You need data to train the machine in order for it to do something useful and usually mimic something that humans can do, either because it’s cheaper, or often because the machine can do a lot better. An average AI model might have millions of parameters. Humans when they’re making decisions are using perhaps five to seven things at once, so an AI can often see things that we can’t see, which can be extremely useful.

In terms of the sector here we’re mainly, like anywhere else in the world, following those big fish. Often they will create AI systems that are meant for one purpose, but actually turned out to be useful for other things. And companies here can spot that and they could do it at two levels.

Either they can just use what I call commodity technology, when an AI becomes part of a Software As A Service platform that you can get a subscription to e.g Midjourney, which is a stable diffusion model that allows you to generate images. So for instance, there’s this company locally, Storm + Shelter, who’ve shown how they can use Midjourney to do their concept development for animations. It’s not replacing the person who’s doing the development, it’s just speeding up part of their workflow and a human is still making the key decisions.

“[AI] is not replacing the person doing the development, just speeding up part of their workflow”

At the other end, we can try and create AI, so you’re actually generating systems, if you’ve got enough data. But you do need to have a lot of data for this, so it does tend to be just the bigger companies, but we’ve got companies in the region that have looked at deep search, using natural language processing on how they can deliver content. We have companies that are looking at audio and video search. So if you’re managing the archive for a large broadcaster, for instance, you can do that better if you have AI tools that can analyze the video footage. You have not just the simple things like transcription – which we have to remember is AI, because a lot of the tools we use are AI but we forget that – but also it can work out who’s in which scene, where the scene cut points are, maybe certain dynamics and spotting certain objects in the scenes, which help people then when they want to use the material later.

But generally we tend to be more at the level of trying to spot the opportunities in the technologies that someone else is creating and create something new and novel – the innovation – out of the way it is applied to the particular problems we have here.

There’s quite a bit of work around the Welsh language because that’s one of the idiosyncrasies of our particular markets in Wales. For instance, I’ve been looking at a project around doing lip syncing, and seeing how the generally trained models worked for Welsh. You can have a piece that is being dubbed and you also move the lips on the people, so it looks a lot more natural. But there are lots of other technologies that really we’ve been looking at for the last couple of decades, but particularly around language translation, subtitling and access.

What’s the most exciting AI development in our sector?

In terms of the media industry at the moment, there are near term things that we really can use now. So people are looking particularly at tools like ChatGPT – not necessarily ChatGPT, there are other models around – but the ability of those language models to speed up processes in terms of writing.

Now, obviously there are some tensions in that area. It can be very helpful to speed up certain bits of the process and it can open up completely new forms of content e.g. you couldn’t have a writer write live dialogue for you in a game. Some things it can do that writers clearly couldn’t do. There is a concern that it takes the authorial voice off the writer, and that actually it will put particularly writers lower down the pecking order out of work. I believe this is already happening and there are challenges in the journalism space around this.

But it’s definitely an area lots of people are looking at, particularly the idea of being able to engage people or deliver really niche content you wouldn’t normally have a budget to pay a writer to do. And the AI is actually very good at editing content as well and doing services around that, which can be interesting.

The other side tends to be generative AI around vision. Mainly at the moment it’s stills. And so some of the still creation systems, like DALL·E, Midjourney, Stable Diffusion, they become very good at making almost photo quality. And in fact, with lots of Midjourney images, it’d be pretty hard to spot whether it’s real or not. In fact, you tend to spot the difference because it’s too studio looking. It’s actually been trained to do studio lighting so much that it looks false because it’s hyper real.

It’s very interesting in terms of how we might use stock imagery in the future, and the idea that anything you can imagine can be created. And it’s actually quicker than trying to search a stock imagery database. It definitely is the end of clip art. But again, it’s kind of challenging for illustrators because you can copy people’s illustration styles. So there’s some ethical dynamics there around whether you would use a live illustrator’s work to prompt the AI to create content, which is kind of stealing their style, basically, which is definitely an issue. Because they obviously don’t get remunerated for you doing that.

“The ethical dynamics around using AI to create content… is definitely an issue.”

But also it could be starting to be used for video. This is very nascent at the moment. It’s very early days, but you are seeing companies starting to set up AI film arms to look at how would they make something that’s completely generated in the virtual space. Personally, at the moment, and I would say for few years to come, AI doesn’t really have that authorial voice. It doesn’t actually have intent. So you do have to govern it a lot. And while it can do some of the tasks that we think of as being part of that creative process, usually you still need someone steering it all the way through that process.

Finally, a technology that’s particularly interesting at the moment because we’ve had a lot of interest in the sector here in virtual production is a technology called NeRFs, neural radiance fields. This is a system a bit like photogrammetry, if people are familiar with that. Photogrammetry allows you to take photographs of 3D scene, and then the AI tries to generate a 3D model of that scene. And it’s been used a little bit, it’s particularly used in pre-viz, for instance, to generate scenes to look at how you’re going to change them with a set for a drama. And then when you actually turned up at the place, you know you’ve got everything right. And you can look at what sort of shots you might want. And you’ve done that before you’ve even gone to the location.

NeRFs take it a step further in that it’s a similar process, you take photographs and the AI generates it, but the AI is generating a concept of how the light is bouncing around the space. This allow you to get a shot from between the different pictures you may have taken, so you’re taking stills at different points, but you can move a film camera between those points. It’s filling in the gaps, if you like, between the information that was captured. So you get to basically control the camera to go anywhere potentially, which is interesting.

But also NeRFs, because they’re dealing much more with the light and not trying to model the physicality, they allow you to capture glass, capture chrome and reflective surfaces, and so you can actually do it in a lot more realistic scenarios. You also can relight the spaces, which is incredibly exciting.

People are starting to look at how this could be used alongside virtual production, whether it be using LEDs or green screens. So that we would go out with the team and maybe in half an hour, they use a drone to capture a space. The drone is just taking video, going round in set loops. And then, back in the studio, you can actually position the virtual camera anywhere within that core space and get a reasonable shot. Obviously if it’s slightly outside where the drone has taken, you’re going to be in a mess.

But at the moment, from the models I’ve seen, it’s starting to get a very exciting way of working. And it could really have some environmental benefits, in that we’re not having to travel to some of those places that we might have actually had to travel to do a shoot. And you can bring the spaces instead into the studio, into the system that you’re using to do your visual effects.

How do we manage the impact of AI innovation wisely? How can we balance innovation with also having wisdom in terms of keeping that human element and being aware of the impact that has socially, economically, careers wise, on people?

People should definitely be concerned. It’s kind of healthy. And hopefully that concern turns into taking a real interest, because we have to, in order to take control of the future, actually be really engaging with what AI can and can’t do, and particularly in terms of our own careers,

“Very fast change [due to AI] is going to be very painful… the best way to slow it down is to learn to integrate it within our roles”

For me there are a couple of different ways you can take it. You could say, “I’m going to go out and strike” and if a company is forcing you to use AI in a way that really would put you out of a job straight away, I could completely understand. That’s the biggest challenge is very fast change is going to be very painful. But more generally, I think that if we can slow down that change, and the best way of doing that for me is actually using the AI as it comes out, so we learn to integrate it within our roles.

AI and most new technologies, historically, don’t tend to just replace jobs. They initially replace tasks. It’s very rare that a technology will just outright replace a whole set of jobs in the short term. What it does is it allows people to do more work, or we actually find expectations of quality go up or of volume go up. And then slowly over time you find certain roles will shrink. So, if you’ve got five people doing a role, perhaps, and you’ve got a whole set of tasks that speed up because of AI, obviously you might find you need three people doing that job in the future. But AI doing that job on its own at the moment is not very likely.

In the distant future, yes, probably we’ll be using AI a lot more to do editing jobs and those sorts of things. But at the moment, and in order to slow down that rate of change, because it’s change it’s really scary, we just need to make sure that we’re utilizing AI, building into our work, taking the benefits, the advantages, the creative opportunities, that time-saving out of it, and finding new things to make. Because usually what happens when a new technology is introduced is the market actually creates more need for the product. So we find the appetite actually goes up to match that increase in supply.

It is very rare that something completely cuts things out. And usually it’s actually older technologies it cuts out. So, for instance, a great example of the effect AI is going to have is the effect photography had. So artists all complained at the time of photography that’s it wasn’t an art form – now of course we do see it as an art form – and that it was going to get rid of all portrait artists. Well in fact only really rich people could afford portrait artists. They carried on wanting portrait artists for some time. They sometimes wanted photography as well as a gimmick.

What it really cut was the other technologies. So very quickly you saw postcards go from using etching to photography in just a few years. But that wasn’t replacing all those jobs. It was the etching companies who were taking on the photography instead. Similarly people used to get silhouettes drawn of themselves. We don’t even see that any more as a thing. That’s something that did just completely disappear, but those studios had to introduce different things.

Now that change is always going to be painful. It’ll be painful for some people today. So we need to try and work out ways that we slow down that change. And for me, as a person who’s interested in innovation, my view is grab hold the innovation. Be one of the places that is starting to use that change as soon as possible, so that you’ve got a longer period of time to change your skillset.

I did say there were three ways. There’s kind of a, “we’re not going to do this, and we try to limit it in some ways”, which for some things is completely appropriate. There’s “we will take it on board and we will try and integrate it in what we do”. And for some people, the third option is simply you spot that AI is going to be a massive new market. They will need people to train AI. They’ll need prompt engineers to help it make great imagery or great video. And so there are whole new careers that it will create as well. So depending, perhaps on where you are in your career, you take a different approach.

What’s your advice on exploring and perhaps embracing AI? What can people within the sector do to innovate?

Well, firstly, it’s really awareness, being aware of the change, trying not to get scared by a little bit of technical language once in a while. Because usually people like myself, we’re just communicators, we don’t understand it all either, and a lot of journalists, they’re just trying to interpret it. As long as you can get past the jargon, you usually find that actually there is a story under there that you can understand and you can see why that tool or that technique might be useful to you.

Particularly start with the things that are useful to you in your day job. At the moment, tools like ChatGPT, various Large Language Models, there are lots of ways I’m seeing people just using them very practically. If you’ve got to send out lots of different social media, well just write a general description, give it to one of the chat engines and it will write rewrite it in different forms for you. You can then tweak it. You can get it to write lots of different versions. So, you still find yourself doing quite a lot of work. It’s not like it’s a single button press and it’s done it all. But just starting to get used to that idea that we’re working with it.

And I think people will do this very quickly. Microsoft has started to bring these tools into their core products. Adobe have just launched a Stable Diffusion-like image generator inside Photoshop. And so we’ll find people just start using this very quickly and before we know it, they won’t even be calling it AI, because we’ll just be using it as a tool. So first step is making sure you are doing that.

The second thing is starting to understand where it might be going in the slightly longer term. So being aware of what’s being launched, what new tools are coming out. There are sites like Future Tools or Product Hunt that help you be aware of anything new that comes out, which is quite cool. For instance, we’re recording this podcast using Descript, which is a tool I first discovered on Product Hunt a few years ago. And it’s that sort of thing that is really enabling for you to be aware of, and that that’s the opportunity and where you might want to change your skillset and your working practice in the next few years.

 

Finally though, the really simple thing to do is come to some of the Media Cymru events. We do events that are about inspiring you, about introducing you to some of the new technologies, sometimes getting quite deep under the hood to see if it is useful for you or not? Is it worth your time understanding it? And then we can, through demos and trials, actually let you get hands-on with things, so we kind of, de-risk the innovation for you and hopefully make it a bit less scary.

What are the Innovation Surgeries?

So the surgery sessions are really kind of a step on from coming to our usual events. So we hope we’ve inspired you. Often people come who are already kind of in the space where they want to innovate. They just need a bit more support. They need a bit more helping de-risking that innovation. So the surgery is really a case of flipping the dynamic, usually we’re telling you about all the stuff that’s cool. What we want in the surgeries is you tell us what you actually need.

And we start by doing a bit of horizon scanning, which is simply looking for the technologies that might be relevant to what you do. And we then think about ways you could apply it to your business and the different benefits that you could get from the technologies. Those benefits could be new ways of reaching the audience, could be new ways for more deeply engaging the audience. It can be cost effectiveness or time saving in terms of your own production process. Or it can be just simply being more creative, giving you completely new creative options you’d never thought you had. Or we go through those.

Often it’s a case of matchmaking. I don’t know everything, so it will be a case of me saying, well, at the higher level, here’s what might be of interest. Oh, and if you’re interested in that perhaps you should talk to this person, they’re a specialist in this. And we’re also help people particularly start up businesses with their business support. So my partners, Town Square offer Startup Club, and they can offer support off the back of the surgery sessions. The surgery is kind of a dialogue and we often find that we ended up pushing people into our funding rounds. I then off the back of the funding rounds, they’ll come back for surgeries if they get funding, to look at how they actually develop their R&D projects. And one of my specialisms is rapid prototyping and how to do live pilots. So often we help people around thinking about, how do you test something as efficiently as possible and really learn from your research experience. Because it’s all about that, how do we learn something new that we can apply to our businesses.

Thanks Robin for taking the time to talk to us. Check out the latest Creative Collective Innovation Spaces events, including Innovation Surgeries and StartUp Club. Sign up for the Media Cymru newsletter for all our latest events and opportunities.