AI Dungeon’s creators are launching an experimental AI-powered game platform

Latitude, the company behind the text game AI Dungeon, is launching Voyage, an artificial intelligence-powered game platform. On Friday, the firm revealed the closed beta, as well as a waitlist for current AI Dungeon players. It’s the next stage for a firm that started as a university hackathon project and intends to one day assist others in creating their own games using trained AI models.

OpenAI’s GPT-2 and GPT-3 text generation algorithms are used in AI Dungeon 2, which was released in 2019. To begin, compose your own adventure setup or generate some opening text. Then you can type in any command you like, and a virtual game master in the style of Dungeons & Dragons will improvise some prose to describe the result. It’s strange and entertaining, but it lacks standard game elements, more akin to an interactive fiction engine.

The games in Voyage are more structured. Medieval Problem is a Reigns-inspired experiment in which you play as the ruler of a kingdom and issue freeform text commands to your advisors, with the results reflected in success ratings. It’s still a lot like AI Dungeon, but with a clearer framework for what you’re expected to do and a system for determining success – however, after playing the game, that approach seems a little too lenient and random.

Pixel Meanwhile, this is a party game in which one player enters a sentence, the AI constructs a pixelated image of it, and the resolution of that image slowly rises until another player guesses it. It’s a combination of the art app Dream and a Pictionary-style mechanic.

Voyage, according to Latitude CEO Nick Walton, is a natural progression for the company. “AI games are kind of resuming at the beginning,” according to the business, with text adventures akin to Zork or Colossal Cave Adventure. “Now we’re getting into 2D graphics, where you can see some visuals.” AI Dungeon, which is included in Voyage, recently includes Pixray image generator-generated AI-produced photos.

The ultimate goal for Voyage is to include game production tools, not just games. “Our long-term ambition is to enable creators to create things that are dynamic and alive in a way that present experiences aren’t, as well as to develop things that would have previously required a hundred people in studios,” Walton says. There is no set timeline, but Latitude expects to work on the system throughout the first part of next year.

LATITUDE WISHES TO ADD TOOLS FOR GAME CREATION, NOT JUST MORE GAMES.

Voyage could be able to find a long-term business plan with the use of creative tools. The GPT-2-powered features in AI Dungeon are now free, however, access to the higher-quality GPT-3 algorithm requires a subscription. Following the Voyage beta, Latitude intends to launch a subscription service for the game.

However, Voyage’s new games lack the diversity and replayability of AI Dungeon, and they’re still plainly the work of a business trying to crack machine learning games. “This technique,” Walton says, “is one of the things I think will be incredibly advantageous in terms of being able to iterate and find out what experiences people prefer.” “You can kind of take current models and make a game that you’re pretty sure people would love using traditional games.” But this space is so distinct that it’s difficult to say.” The question is how much people will be willing to pay to be a part of it.

Latitude will certainly need to be cautious with OpenAI’s application programming interface as its purpose grows (API). GPT-3 initiatives are approved on an individual basis by the organization, and they must follow content criteria to avoid misuse. Since AI Dungeon allows users a lot of latitude to design their own stories, Latitude has battled with these constraints in the past, which led to some users generating disturbing sexual scenarios, which upset OpenAI. (It’s also dealt with user command security issues.) Before establishing a deal in which some user commands would be routed to a non-OpenAI algorithm, the business spent months working on filter systems that mistakenly filtered more trivial fictional content

“IF YOU TRY TO MAKE SOMETHING SUPER SERIOUS WITH AI… IT WILL BE DIFFICULT”

Although Pixel This and Medieval Problems are more closed systems with fewer visible moderation dangers, offering creative tools risks undermining OpenAI’s governance of GPT-3, which could cause its own set of problems. Latitude plans to transition more of its games to other algorithms in the future, according to Walton. “We’ll have more structure and systems in place so that it’s not just consuming the [OpenAI] API straight.” At the same time, I believe that the majority of our models will be ones that we host ourselves,” he says. This includes models based on new open source projects, which have struggled to compete with OpenAI’s efforts in the past but have progressed since their inception. “I don’t think that gap will last very long,” Walton says.

Most video game “AI” is a very simple set of instructions, and many games use procedural generation, which remixes developer-created building blocks to produce massive amounts of material. In contrast, a business like Latitude uses algorithms that have been trained to generate text or images that suit a pattern from a data set. (Think of them as a super-advanced version of autocomplete.) Outside of gaming, other businesses like NovelAI have used text creation for creative work.

Latitude, on the other hand, is still figuring out how to create mechanisms in which participants can expect fair and reliable results. For example, text production algorithms have no built-in sense of whether an action succeeds or fails, and systems for making those judgments may differ from human intuition. Image generating algorithms are fantastic for making strange art, but in a game like Pixel This, players can’t always tell how identifiable a given picture would be.

For the time being, Latitude’s response is to embrace the chaos. “If you try to develop anything extremely serious with AI where people demand a high level of coherence, it’s going to have a hard time,” Walton adds, “at least until the technology gets better.” “However, if you embrace that component of it and let it be crazier and wackier, I believe you can create a wonderful experience where people are delighted by those surprises.”