"It’s about asking what AI can allow you to do that couldn’t be done before"

Hello! Welcome to AI Gamechangers, your regular guide to the people putting artificial intelligence to work across the games industry.

Each edition, we sit down with founders and technologists who are doing more than just streamlining pipelines; they’re rethinking what interactive experiences might become with AI’s help.

This week’s guest is Japhet Asher, a writer, director and producer. From launching MTV’s Liquid Television to pioneering interactive content for children at the BBC and developing character-led projects at Meta, his career spans the full evolution of digital storytelling. We discuss AI as an architectural tool for storytelling, why the “all-in vs. cautious” adoption debate is a false dichotomy, and how to build fiction that audiences don’t just consume but inhabit.

As always, scroll down for the full interview, plus our curated round-up of the week’s best AI news and tools. And feel free to browse our catalogue of interviews from the last two years.

Japhet Asher, Polarity Reversal

Meet Japhet Asher, who has one of the most inspirational CVs you’ll encounter. He’s an Oscar-nominated and Emmy-winning producer known for MTV’s Liquid Television (which gave us Beavis and Butt-Head and Æon Flux) and the BBC’s children’s division. He spent two years at Meta, training and shaping characters and narratives using AI and augmented reality. Now he’s a consultant with a fascinating blog and tons of insight into interactive narratives.

In this interview, we explore AI as a tool for narrative design, the realities of working across multiple frontier models, ethics and guardrails, the unpredictability of real-time AI characters, and why studios should focus less on cost-cutting and more on opening entirely new creative territory.

Top takeaways from this conversation:

1. The real value of AI for studios isn’t cost reduction or efficiency gains alone. It’s about amplifying what creators can do and opening up entirely new kinds of games and experiences that weren’t previously possible.

2. AI characters tend to collapse into archetypes unless you build in conflicting drives, backstory and friction, just as you would when writing for any other medium. The craft of characterisation still matters.

3. Relying on frontier models for real-time game characters is risky: every model update can change a character’s behaviour. Studios should consider smaller, locally controlled models for functions where consistency is critical.

4. Creators working with AI are designing spaces for players to have personalised story experiences rather than controlling every word. This reframes what IP development looks like in an age of interactive fiction.

AI Gamechangers: You’ve had an amazing career. Please give us a sense of what your mission is today and how you got here.

Japhet Asher: I’ve been on a fairly long journey through storytelling in various contexts. I started out as a documentary maker a very long time ago, and had some significant success in that area, but then took a complete pivot and got involved in animation. I grew tired of the artifice in “packaging reality” and wanted to explore things from the opposite side.

I’m probably best known in that arena for creating a show called Liquid Television that spun out Beavis and Butt-Head and Æon Flux, and was sort of the first real short-form compendium experience; in a way, I was curating a “YouTube playlist” before YouTube was around!

“These tools are changing the way we need to develop IP. It’s possible to build a story world you can explore as a creator and find slices of story you want to distribute. This has come along at a time when audiences are increasingly spread across platforms and different media”
Japhet Asher

I was recruited from that world into American telcos who were developing interactive television about 15 years too early. Tends to be the story of my career! I had an amazing experience developing interactive applications; I got the interactive bug and wound up in charge of interactive content for BBC Children’s.

My journey continued, exploring new interactive experiences at BBC Children’s. I got the bug for augmented reality; I wrote the world’s first AR-powered novel and ultimately started developing ideas for my own startup combining AR and AI. I then started working on a contract with Meta, developing characters and interactive narratives, combining characters and exploring games with real-time AI.

What sort of storytelling projects are you working on today, thanks to AI?

I write an AI-themed blog called The Context Maker, where I share a lot of the experiments I’m doing, like The Face Tracer, a quest for your historical doppelgangers, and Quantum Teapot, an exploration of personalised interactive narrative. I’m interested in how real-time AI allows you to not just role-play, but enjoy interactive fiction in a broader sense.

The challenge there, if you’re working with a model (in those cases, it’s ChatGPT), is creating the right kind of system prompting to keep your vision as a creator, while still making it a collaboration with the user. The work I do in that space really leans into the things that I’ve learned through working on model training at Meta in particular. I look at what is and is not a strength of a particular model. I find the things that might in some circumstances be problematic from a model become quite interesting in a storytelling or artistic sense, once you identify them.

I think that people on average tend to go to AI chatbots for utility, to tell you how to cook sausages, or to tell you about tax law, or give medical advice, though some of these things are probably a little dangerous, honestly! So what is the role of a personality in that context?

We have the ability with most of the major models to adjust personality. But I find that AI characters tend to form into archetypes with models, because models have familiar pathways. So, in order to get truly interesting characters, you have to think about creating conflicting drives, just as you would as a writer. You have to give any character, even a utility character, a backstory, and a place to find friction in its own responses, so that it’s not entirely predictable – so that it’s not homogenous. And that’s where you start to get interesting engagement.

What was the moment on your journey where the lightbulb went on and you realised AI was going to be important to storytelling?

I was an early adopter of ChatGPT. In general, when I see a shiny new toy, I’m going to play with it and explore what it can do. I’d also started to look at Midjourney and some of the other initial [generative art models].

I was working on the startup project I mentioned, which was combining augmented reality with machine learning and optical character recognition. The idea was that any physical book could have digital overlays if you used selected printed text as an “anchor” for where the digital material should be displayed. We built a prototype, figured the tech out, and got some money from Innovate UK. As that was happening, I was starting to understand that the emergence of smart glasses would be important for this work (and what time frame that would take!).

We had audio and video links coming through, and related text and comments on a page, turning a physical book into a “book club”, with digitally annotated notes. And I started to think, “How do you combine an audiobook with its physical book in a really useful way?” I realised you might have a chatbot as a reading companion, and started to explore what that might look like.

Then, of course, ChatGPT was released in November 2022, and finally, I could say, “This can happen the exact way I thought it might!” That sent me down the path.

Technologies like virtual reality have a lot of potential, but often the public is slow to buy in – perhaps because of expense. At the same time, millions of people use ChatGPT (and have adopted it very quickly). What does that tell us?

It’s a different thing. ChatGPT isn’t a repository of all knowledge. It’s an interface; there’s a predictive algorithm, and it presents text. But it allows you to use your writing, and now your speech, to communicate with computers, which most of us weren’t able to do well before. That’s something easy to grasp. What is difficult to grasp is how to use it effectively! But getting some kind of response from it that can delight or intrigue you is very straightforward. That’s a big part of its success.

It’s always about the interface. Take AR, for example. As long as it’s on your phone and you’re holding it and walking around trying to look at things through a little phone screen, it’s always going to be limiting. VR even more.

The combination of AR and AI excites me massively! If there were one thing I would most like to do next, it’s to work on that. Some of the work I was doing at Meta is probably now leading down that path, but I’m no longer there. But I’ve had a glimpse of what real-time AI with AR might be – where that’s going is exciting.

Let’s talk more about combining technologies. What excites you about AR, AI and interactive fiction together that just isn’t possible with linear storytelling?

On one level, I’m just interested in seeing how people use story worlds that I make. From a creative perspective, it’s entertaining to read transcripts or see how people engage and where they take things.

“As a society, we need to think through how we introduce AI into kids’ lives. Children themselves need to have a voice. Kids are using AI now, and it’s going to be integral to their existence. Teaching critical thinking will be super important”
Japhet Asher

There are a couple of pieces on my Context Maker blog about what one learns and notices, and that’s fascinating in itself. Maybe that’s a selfish creator instinct. In the end, we’re moving into a world where your role as creator is more as an architect. My blog is called The Context Maker for a reason: it’s the idea that you’re creating a context for people to have narrative experiences. That’s what excites me.

I think some writers are writers because they like controlling every word, and they like controlling every element. But you aren’t really in control, because people bring themselves to the material. Stephen King describes a wonderful thing about the telepathy of writing – he describes in prose a table in his office, and by doing so he’s teleporting it to you wherever you are when you read it, at a different time and space, and you, of course, will imagine it differently. It will be a different table, though its essence will be the one teleported. And I think that’s true when you think about creating fictional contexts with AI engines. You can create environments where the user will experience something akin to what you intended, but it will inevitably be a marriage with their own intent. And I think that’s kind of wonderful and interesting.

It also naturally leads me to another question: the nature of authorship. It’s interesting to describe yourself as an artist or architect. You’re building worlds, context and characters. But unlike, let’s say, a novel, you’re not shaping the whole narrative. It’s a relationship between you and the interactivity of the participant, and also the AI is in there as well. Is AI an author of this experience somehow?

No! AI doesn’t do anything unless you ask it to. So I don’t perceive AI as an author, and I don’t really think that people should get really het up about it. “AI wrote this!” No, you asked it to, with a prompt. It generated something, and you said, “Yes, that’s exactly what I want to say.” Therefore, you are the author. If it wasn’t what you wanted to say, or it wasn’t said the way you wanted, you need to change it. You can change it by re-prompting it, or rewriting it, or you can treat it as inspirational and write it entirely yourself. But either way, if you’re saying, “This is my voice”, the fact that you’ve used AI as part of that process doesn’t excuse you from authorial responsibility. That seems really straightforward to me.

However, in interactive fiction, you are writing with your audience, because they are going to personalise and change it. It’s obviously possible to use an orchestrator model to create a set of story beats that have to be met based on certain moments and drive a narrative in a specific direction. Even in interactive fiction, you can stick an ending on if you choose to, and that’s part of the work of structuring behind the scenes.

“They say AI won’t be original. That’s idiotic. People are original, and we’ll use these tools in original ways, to create original things – if we experiment with them, explore and try things”
Japhet Asher

That’s difficult, though, because a problem I’ve found so far with AI models is that they tend to anticipate the ending in the middle! They’re very bad at holding on to secrets. I’m sure we will solve that, but at the moment, it is definitely problematic. If you upload large knowledge uploads with lots of backstory, sometimes I find that models just obsess over some particular element, and it becomes much bigger than you intended in the narrative. That’s very irritating. There’s a level of unpredictability you have to roll with.

Have you experimented with many models? Is the OpenAI model definitely the best for interactive fiction?

Yes and no. It was good. The trouble is that every update changes everything that I’ve done! I originally made The Face Tracer with GPT-5, and I noticed that 5.2 has changed it quite a bit. This is one of the issues with creating things relying on a specific model. And now OpenAI is finally removing access to 4.0 and 5, which will upset many people and undermine stuff I have made that leaned into the characteristics of those specific models.

Everything I do now, I tend to use both Claude and ChatGPT for different things. I work with Claude and set up projects there. Brainstorming and exploring ideas are good things to do, though Claude constantly brings up things I could add; so then I use ChatGPT as my editor, without any of the knowledge of those conversations.

I use Gemini when I’m working on things that are multimodal. I value Nano Banana Pro for precise image outputs. For a while, I was building with Character.AI, but I’ve stopped doing that, only because I don’t really have an outlet for them.

Everybody needs to explore different models and find what they’re good at – for them. Sometimes people say, “Claude is great at writing!” That doesn’t mean Claude definitely is great at writing. It means that they have found Claude Sonnet 4.5 to be a good match for their kind of voice.

“ChatGPT isn’t a repository of all knowledge. It’s an interface. It allows you to use your writing, and now your speech, to communicate with computers, which most of us weren’t able to do well before. That’s something easy to grasp”
Japhet Asher

One tip I can hand out is this. The work I did learning how to create personalities and distinctive character profiles, with Meta and on my own, I’ve now applied to the way I create writing styles in Claude. I’ll pre-prompt via the writing style training tool as if I’m prompting a character, with backstory, behaviours and dialogue examples, and then that character is effectively the author, which seems to work well in terms of creating characteristic writing.

As you say, when a tech company updates their model, like OpenAI updating ChatGPT, users often find its personality changes. That could have huge implications for videogame studios creating characters that run on these models. Do you have any advice for people on how to roll with changes like that?

It’s difficult. Depending on what you need from a character, you have to ask yourself whether you need one of these frontier models – frankly, for many functions, you could use smaller models, locally, that you can control.

I think people have to be pretty careful about how they engage with the larger models, and also recognise that it’s very difficult to create guardrails. Really consider that for any game in an environment where that’s a risk.

I find real-time AI interaction really enjoyable and interesting, but it’s also the most challenging part of AI in creative practice. I’m much less concerned about the ability of AI as a tool for generating images or video. Even if it’s text that you curate yourself that a player never actually engages with, it’s something you’re ultimately responsible for. You take responsibility for what you put in front of people.

With real-time AI, it’s like you’re setting something free in the world (within a framework) to see what happens. Every time, it’s an experiment! You have to keep that in mind. Having said that, obviously, there are more small, containable models which you can train around very specific functions and pathways. And I think people just need to make sure they don’t over-arm their characters!

You were on a panel at PGC London where the topic was about whether it was wise to go all-in on AI or exercise caution in your business. Where did you land in that debate?

I said that it’s a false choice! As I’ve said, there’s a big difference between using real-time AI versus using AI tools.

But also, I don’t know that you can tell anybody who’s new to AI to go all-in on it. If somebody isn’t yet using AI, they probably never will. That’s a valid view. But for most of us, it would be crazy not to explore what these tools can do for your studio, for your pipeline – not because of efficiencies and cost reduction, though those have merit. But because of the ways it can amplify and give you opportunities to tell stories or make games that you couldn’t have made before. People are very focused on how AI can save money and time, or reduce staff needs. But actually, the real question should be, “How can this allow me to do things that I couldn’t have considered before? How can this amplify what I do? How can this open up new territory?”

People talk about AI and say, “It can only repeat everything we’ve done before, because it’s trained on everything we’ve done before.” They say it won’t be original. That’s idiotic. People are original, and we’ll use these tools in original ways, to create original things – if we experiment with them, explore and try things.

So you know, it’s not about being “all-in” – it’s about asking what this can allow you to do that couldn’t be done before. And sometimes that might be streamlining a process so that you can use resources somewhere else, but sometimes it might be making a different kind of environment, exploring world models and what they mean. Taking a different approach. I think every maker, every studio, should at least have some portion of their time set aside for exploring different tools and seeing what they could potentially add to their creative production capabilities – not to replace anything, but to see what could be different. Everybody will find their own answer to what works for them.

You’ve worked for many different audiences, including a very young audience when you were at the BBC. As responsible creators, do we have to bear in mind what children might use AI for? What trust and safety issues do we need to consider?

We have to consider this. I will admit that I’ve been struggling to think through a piece I’d like to write about children and AI models, because it’s so very tricky: kids have to learn to use these models and will interact with AI throughout their lives. As a society, we need to think through how we introduce AI into their lives.

And children themselves need to have a voice. Kids are using AI now, and they won’t perceive it the way we do; it’s going to be integral to their existence. It just is going to be there. Teaching critical thinking, teaching how these machines work, making sure kids have both the creative faculty to use them well and the critical faculty to use them cautiously – these will be super important.

“You are the author. If it wasn’t what you wanted to say, or it wasn’t said the way you wanted, you need to change it. You can change it by re-prompting it, or rewriting it. But the fact you’ve used AI as part of that process doesn’t excuse you from authorial responsibility”
Japhet Asher

On top of that, people are making default decisions at the minute. There’s a great debate going on in the UK about whether to ban YouTube and other social media for people under 16. And I can see the same thing happening with AI. There’s age gating to a degree already, but at the same time, loads of kids are using AI. Most kids are using it for their schoolwork! I did some research for my blog, and I believe it’s something like over 80% of students in high school in the UK are using AI for their homework. That’s dramatic, and I don’t know that anybody has put the necessary thought into what that means.

My particular solution for things like using Claude in the classroom is that kids should be turning in their conversations, not turning in just their output. Show how you work. It’s a lesson I learned many years ago when I succeeded in a maths exam despite getting the answer wrong – they saw my scribbles, and they could see I was actually on the right path.

Ultimately, it will be about that kind of transparency, both in terms of education and learning, and in terms of usage. I think all people should be transparent about using AI. There have been studies done where an audience will love something at first, and then they find out it’s done with AI, and then they won’t love it anymore. But that moment will pass. I can remember the same thing happening with computer graphics...

Back in the 1980s, the original Tron was blocked from the Academy Awards because it was felt that using computer-generated imagery was not true craft. Now you’d struggle to find a movie that doesn’t use CGI in some way! Technology becomes accepted and part of the creative toolset over time. Do you think AI is on a similar path?

In the creative industry sense, I feel that’s true. Obviously, generative AI is vast in its implications for all kinds of areas – and artificial intelligence, more generally, for society and the world.

“The combination of AR and AI excites me massively! Some of the work I was doing at Meta is probably now leading down that path. I’ve had a glimpse of what real-time AI with AR might be – where that’s going is exciting”
Japhet Asher

But in terms of the creative industries, I think the switch from hand-drawn to computer graphics is a relevant comparison. I’ve been around long enough: I started out making films using a stand-up Moviola with a foot pedal – it felt like driving a car. You’d work with yards of real film, razors, tape and all the rest of it; it was very visceral and tangible. You had a physical relationship with the medium. Gradually, we’ve abstracted and abstracted [the process], yet the viewer still responds in emotional and indeed visceral ways to the games and films on screen.

The medium is not the central issue here. It’s the message, as ever. And I think people should remember that: these machines have no message. We have to be the message to each other.

What are you working on now, and what’s your next creative project?

I’m doing some consulting – I like using these tools and exploring what’s possible and helping others do that too. If anybody wants to explore that with me, do get in touch!

Strategy is one thing, but, equally, making things is another. I have a couple of things of my own that I’m exploring creatively. I’m still very much focused on telling stories, and I still regularly write and make without using AI at all. But I’m also thinking for myself and others about how these tools are changing the way we need to develop IP, because it’s now possible to build a story world you can explore as a creator and find the slices of story you want to distribute. And this happens to have come along at a time when audiences are increasingly spread across platforms and different media.

There’s an interesting opportunity here to rethink what story development is like. You could train a model on your storyworld instead of just building a bible and pitch deck: you could create a place that is not only for users to visit but for the creators and potential funders to explore. You add to it as you develop more storylines, art elements and so on. You may find new avenues of narrative, and you’ve got something to consistently come back to and build from for all the different iterations you might make for the diversity of platforms that have become touchpoints for your audience.

Subscribe now! It’s free, and you’ll receive interviews with industry insiders and studio leaders in your inbox. Coming soon: Unico, CALM Consultancy, GameBot and more.

Further down the rabbit hole

Some useful news, views and links to keep you going until next time:

• MapleStory Universe developer Nexpace has announced an investment into South Korea generative AI game platform Verse8 (we interviewed its CEO in the previous AI Gamechangers).

• Roblox has launched its 4D generation technology in beta, using the Cube Foundation Model to enable players to create interactive, functional 3D objects through simple text prompts, including cars that drive and planes that fly.

• Saudi-backed game publisher Scopely confirmed using generative AI to create a Star Trek Fleet Command advert that drew criticism from gamers and pundits.

• Is it time to be worried? In the same week that OpenAI research scientist Zoë Hitzig quit, so did Anthropic safety researcher Mrinank Sharma, writing an open letter saying “the world is in peril”.

• ByteDance unveiled its new AI video model Seedance 2.0, prompting performers’ union SAG-AFTRA to issue a statement about “blatant infringement” of actors’ voices and likenesses.

• Gaming experts, including Rebecca Liao and Oscar Clark, weighed in on what Google’s AI world-building tech Project Genie means for games.

• Japan revealed plans to support the games and content industries. Measures will target global distribution, anti-piracy efforts, and AI-driven translation tools.

• Bitmagic (who we chatted with back in 2024) revealed that its AI development tool has been able to create a version of the classic game Civilization (“Idle Civ”) along with other titles. You can play it on the official site.