‘Progressive’ AI dialog tools are a Trojan horse

Ubisoft’s recent announcement of an AI tool that generates dialog for Non-Player Characters in video games was greeted with supine equanimity in most sections of the gaming media.

Headlines like “New Ubisoft AI Tool Could Help Scriptwriters Create More Immersive Game Worlds,” and “Ubisoft Developing An AI Ghostwriter to Save Scriptwriters Time,” ticked the company’s core messaging boxes for the product, called Ghostwriter.

Most of the ledes I read were obliging enough to parrot Ubisoft’s assurances that the tool is a time-saver, designed specifically to help its workers make more creatively productive output, and not in any way to, say, diminish the cost, size and power of videogame narrative teams.

Ubisoft’s video demo of Ghostwriter makes the tool seem useful enough. Human writers enter details of an NPC’s roles, personality, and parameters. Ghostwriter churns out a bunch of potential verbal responses (aka “barks”) which the writers approve or delete, adding their own edits and perhaps a few tonal instructions for voice actors.

Ubisoft claims that Ghostwriter is designed “to help [narrative teams] complete a repetitive task more quickly and effectively, giving them more time and more freedom to work on the game’s narrative characters, and cut scenes.” For consumers, the claimed benefit is “NPCs that are interesting, consistent and just chatty enough.”

Elimination of opportunity

But the notion that Ghostwriter portends a better life for narrative designers, and a diminution of a tedious part of their lives was not greeted with the same optimism by some of the people whose working lives it claims to aid.

Writing on Twitter, Sam Winkler, lead writer of Gearbox’s Tiny Tina’s Wonderlands referenced a bunch of negative outcomes from wide adoption of products like Ghostwriter. For a start, it eliminates a core element of basic training for new narrative designers for whom “barks, ambient dialogue, and descriptive text is a lot of writers’ first billable (and résumé-able) work when they are starting out”.

Those writers’ newfound reliance on a Large Language Model (LLM) AI tool makes it less likely for them to generate “interesting work” that might give the game unusual charm, and help them get noticed as promising writers. To put it another way, ticking boxes and making minor edits in an AI suite is probably not going to attract ambitious young writers, nor will it help them develop their skills.

It’s simple editing work that could easily be farmed out to lowly-paid and outsourced, contract workers.

Winkler’s most devastating critique is that “if Ubisoft finds itself needing more barks than it can pay actual writers to write, they have WICKED overscoped. I say this having just assembled a team of talented external writers to assist on something my core team didn’t have the bandwidth for. It’s fucking worth it.”

Alanah Pearce, who works with the narrative team at Sony Santa Monica made a similar case: “As a writer, having to edit AI-generated scripts/dialogue sounds far more time consuming than just writing my own temp lines I would far prefer AAA studios use whatever budget it costs to make tools like this to instead hire more writers.”

Why would Ubisoft spend more money on creating a tool to do a job that humans can manage at relatively low skill and wage levels?

The most likely explanation is that LLM AIs will improve rapidly in the next few years, and that they will begin to compete with more experienced, and expensive writers whose work focuses on lore, player-character dialog, and plot.

Labor impact

A recently released research paper on “the labor market impact potential of Large Language Models” claimed that up to 80 percent of U.S. workers would find their work affected by AIs in the next few years. The researchers clarified that “affected” is not the same as “eliminated” and it’s sure to be the case that AIs will sometimes bring benefits to workers and consumers.

The paper released a list of jobs that would least likely be affected by AIs. They were all specific manual labor jobs that require a range of training, from lowly dishwashers, to skilled carpenters. None of the jobs mention working with words or language, such as script writers, lawyers, or journalists. Nor did any of those jobs touch upon programmers, artists, animators, and other creative roles in digital entertainment.

The author and screenwriter Michael Marshall Smith (disclaimer – I know Michael and have socialized with him on occasions) posted a thoughtful Substack editorial on the moral problem at the heart of stuff like Ghostwriter.

Smith writes a lot of pitch-deck presentations for TV execs. He makes use of online photography and art to help his thinking process, even though “I’d never have the budget to commission artists or photographers”. Now, he’s begun to use AI image generator Midjourney, and he finds it more useful than his previous practice, while acknowledging its current limitations.

“It’s immediately clear that Midjourney has an in-built aesthetic, an undercurrent discernible regardless of tweaks you make, a result of the reference material it’s been fed,” he writes, before making the point that these products are only going to get better.

Like most people earning a living from creating commissioned artistic work, it is not commercially viable for Smith to commission original artwork every time he knocks out a new pitch. But he does understand the implications of shifting from human creations to AI creations, because there are plenty of organizations that do have such budgets, and which spend them on the useful employment of skilled artists.

“The thing is… I’m not an artist,” he writes. “I’m not seeing my entire livelihood suddenly at risk. I’m not watching aghast as a commission I might once have been paid hundreds of dollars for — money for food, rent, my kid’s education — is suddenly fulfilled by a computer operated by a non-artist.”

Foolish notion

At Game Developer Conference last week, anecdotal evidence suggests that anything AI-related attracted the biggest crowds. Writing on GamesBeat, Dean Takahashi noted that a Generative AI social meetup had “450 people on its RSVP list and another 500 more on the waiting list.”

Game companies spend a lot of money sending employees to GDC, and they usually have a focused agenda on what is most important to them in any particular year. There are certainly more products like Ghostwriter being developed.

Only a fool would swallow the notion that those products are being designed to make the lives of narrative designers better in any way. Nor, for that matter, are these products guaranteed to make the games themselves any better.

Drawing from her experience in videogame language translation, localization specialist Julia Gstoettner wrote on Twitter: “Machine translations have been exactly this for years in the localization industry now, that’s why there’s such a drop in general quality in the industry. There’s more and more money to be made from games, and yet, costs are cut in ALL the wrong places.”

It’s a bitter irony that Ghostwriter carries the same name as an actual human profession. The tool’s creator Ben Swanson gave a talk at GDC, which showed many of his creation’s useful features. But he was at pains to stress how his work is all about making life better for video game writers. According to a conference report on Game Developer, Swanson worked with writers to help them in their work.

The report stated that “Ubisoft has decided that its Ghostwriter tool is best-used to support the process of writing different kinds of barks, background lines, and text for user interface,” adding that “Ghostwriter is explicitly not being used for game cinematics or lore.”

But that doesn’t mean future iterations won’t incorporate these features. It certainly doesn’t stop other companies from pushing AI’s capabilities as far as possible. And even if some human input is always going to be necessary, the entire point of AI is to find ways for computers to do, for little expense, what humans currently do for greater expense.

Videogames are rarely created by individual artists, but by large teams made up of specialists. These are exactly the workers most vulnerable to advances in AI. If one person, and an AI, can replace five people, you can bet that’s what will happen.

A research paper that Swanson helped write, back in 2021, made the following conclusion. “We are particularly excited to see [LLMs’] impact on the fine arts. In particular, we see great potential in tools built with this technology when there is a human, in this case an artist, in the loop [my italics] to complement the natural deficiencies of a simple but powerful text generator that lacks editorial control and responsibility.”

This is how creative industries cut themselves off from developing long term talent, in exchange for short-term cost savings, and long-term power dynamics that are to their own advantage. As we have seen in recent months – and repeatedly down the years – game companies have no compunction about reducing headcount as often as they like, and without much in the way of compassion or even logic.

Ghostwriter itself might never cost a single person their job. It might turn out to be an absolutep boon to its users. But its ilk will absolutely cost many, many people their livelihoods and their dream careers.