If you grew up playing games in the late 90s and early 2000s like me, then congratulations, we have been overwhelmingly spoiled about how quickly games can evolve. During that time period, when developers were finally starting to get a handle on early 3D tech, each year brought *wildly* new experiences from AAA studios. Games at the time would look radically different from games even five years earlier. So, if you were a gaming enthusiast, or even just had a passing interest in the medium, you were in for radically new experiences pretty much every year.

The 2010s have…not been like that. Games from this generation look and play pretty similarly to games from last generation. There are still incredible games being released every year, and the indie scene is more vibrant and creative than it’s ever been, but we aren’t getting too many of these technologically innovative, genre-defining, titles on a year-by-year basis. And, to a certain extent, that’s okay. It’s not like every year we’re seeing formally and technically revolutionary works of film or literature, certainly in the consumer sphere. Once a medium reaches a certain level of technological maturity, technological creativity is less enticing. You already have the tools to do most of what you want.

But with games, a medium that is perhaps the most closely linked with new technology, it does feel slightly disappointing. While I’ll probably be playing new Bioware games until EA eventually shuts down the studio, I’m not looking for Mass Effect 5, I’m looking for an experience that evokes what it felt like to play Mass Effect for the first time. Figuring out what Mass Effect did well is pretty straightforward; figuring out how to recreate the experience of playing it for the first time is much more difficult. It relies on some level of technological and design-focused novelty.

So that’s what makes AI Dungeon 2, my favorite game of this year, so strange. Like Mass Effect 1, it’s a fairly janky experience, but in all other areas, from writing to visuals to tone, it is absolutely nothing like Mass Effect. But the novelty of its approach to narrative evoked how I felt playing Mass Effect for the first time. But, unlike Mass Effect, AI Dungeon…truly is unlike anything else I’ve ever played. It’s like playing Zork with a dying computer. It’s like playing D&D drunk with your friends. It’s like going to an improve comedy show that is heavy on audience participation. It’s…well, it’s like playing Mass Effect for the first time. All of these experiences gesture in the direction of AI Dungeon, but don’t fully capture it, because the game is so unique that it eludes comparison. And that, more than anything the game does in its own right, is what makes it so exciting to me. I am so excited, not just for this game itself, but for the genres of games that could be built around it. For the bits of its tech other genres could steal. Like playing season one of The Walking Dead, playing this made me imagine what others could do with this template. But as excited as I am for the future, I have loved my time just with what we have now.

So. Playing AI Dungeon made me feel like I was playing Mass Effect for the first time because it defined a new (or, at least, new to me) style of interactive narrative. What is that style? The style is the honestly unparalleled possibility space of the system. AI Dungeon’s machine learning model will respond uniquely to almost all player input, creating an experience that, for the player, is functionally infinite. Repetition does happen, but there are always new system states for the player to explore.  As such, this is the closest a game has ever gotten to the “go anywhere, do anything” promise, but it’s worth acknowledging that that promise is often fickle. Sometimes, everything will click and the system responds perfectly to what the player writes. However, it’s very easy to break. Repeating lines, loops, crashes, or the system just not getting what you want it to do. For example, in a recent run my friends and I did, we were under attack from the CIA, so we called Bernie Sanders, who we had just made president, and asked him to abolish the CIA so the attacks would stop. We had to repeat the request multiple times with limited responses, and even after, the game didn’t understand in a systemic sense what the CIA was and that it was abolished. The narrative was mostly in our head. And this is where AI Dungeon actually does have a sort of progression curve, though it’s very different from those in other games. Instead of learning how the systems work and learning to conquer them, you’re learning how to work with the AI to generate the best stories. You learn what types of phrases to avoid, what types of requests the system is more comfortable with, and when to bail because it looks like you’re headed into a loop. These skills make AI Dungeon less a game in the traditional sense, with explicit win and lose states, but rather a collaborative storytelling platform. Your collaborators can include other human players, but the AI is your primary storytelling companion. And that’s pretty unusual for games. I keep pulling from comparisons outside the world of games, because video games are really bad at this type of loose storytelling. In the majority of games, everything has to be pre-programmed, so players never learn to think this creatively, they learn to figure out what the designer wants them to do. A system can only have so many states, right? So, in terms of pure storytelling structure, AI Dungeon isn’t much like Mass Effect at all, with its finite system states and limited reactivity. The experience I’ve had that’s closest to AI Dungeon is doing an improv comedy set with a partner without any prep time. You can’t pause the story and talk about where you want to go next, you’re both just flying by the seat of your pants, trying to signal to the other what to do, but mostly replying with, “Yes, and” to everything they say. As a result, AI Dungeon doesn’t really have win and lose states. You can die and get a game over screen, but that’s pretty rare and easily reversible. The game isn’t about winning, it’s about telling great stories. And that’s an approach to creativity I would love to see more of.

Going into a new decade, it’s tempting to wish for dozens of games using AI Dungeon’s model. But, right now, it simply isn’t profitable. It its current form, the game costs around $10,000 per day to run servers for. No one is going to be making money off that any time soon. But I would love to see games try this more expressive storytelling, because new technical improvements that primarily benefit storytelling are pretty rare in this medium, especially in the last two generations. The potential for new permutations is quite literally endless.