CHOOSE YOUR OWN ADVENTURE: FROM BOOKS TO GAMES TO AI

When I was ten, I kept two fingers wedged in a Choose Your Own Adventure book at all times. One at my current page, one at the last decision point. Insurance policy. Because those books had a mean streak, and I didn't trust the author not to kill me for choosing the left tunnel instead of the right one. That habit, that suspicion that your story could end at any moment if you pick wrong, stuck with me. It followed me from those paperbacks into Zork, into King's Quest, into The Walking Dead, into AI Dungeon. The same impulse has been driving an entire lineage of entertainment for almost fifty years: people want to control the story.

What's changed is how much control you actually get, and what you give up to get it.

The paperback revolution

I've written about CYOA books in more detail elsewhere, so I'll keep this part brief. Edward Packard invented the format. Bantam published The Cave of Time in 1979. By the mid-80s, the series had sold over 250 million copies. Kids were learning about narrative branching at Scholastic book fairs, which is a weird sentence to write but it's true. The books were simple: read a couple paragraphs, hit a decision point, turn to one page or another. Most paths ended in death. A few ended in triumph. You could "beat" the whole book in twenty minutes, but mapping every branch took hours.

The thing those books did that nothing else could do at the time was activate your imagination in a way that felt personal. There were no visuals. No sound. Just text and whatever your brain conjured up. The Cave of Time sent you to ancient Rome or prehistoric jungles or the far future, and you rendered all of it yourself. That's a superpower that gets overlooked when people talk about the history of interactive fiction. A kid reading a CYOA book in 1983 was having a more immersive experience than they'd have with most video games of that era, because imagination has infinite resolution.

The UK produced its own version of this idea with the Fighting Fantasy series, which bolted a dice-based combat system onto the gamebook format. Suddenly you weren't just choosing paths. You were fighting goblins with stats and hit points. It was a solo RPG disguised as a book, and it filled a gap that Dungeons and Dragons couldn't fill for kids who didn't have a group to play with.

But all of these books shared one fundamental limitation: someone had to write every path. Forty endings in a CYOA book sounds like a lot until you realize that a human being had to plan and write each of those forty threads, and they all had to be at least somewhat satisfying. The branching tree gets expensive fast. Double the decision points, and you've more than doubled the amount of content you need to create. This tension between player freedom and authored quality is the thread that runs through everything that followed.

You are in a maze of twisty passages

Computers solved part of the problem. If you let a machine track the state of the world, you can create branching narratives that would be impossible to manage on paper. Zork, released in 1980, understood this intuitively. You typed commands in plain English and the game responded. "Open mailbox." "Take sword." "Kill troll with sword." The parser was clunky by modern standards, but the feeling was magical. You were having a conversation with a story.

Infocom, the company behind Zork, went on to release dozens of text adventures across every genre imaginable. Comedy, horror, mystery, sci-fi. Douglas Adams co-designed The Hitchhiker's Guide to the Galaxy and made it one of the most hilariously unfair games ever created. You could die in the first five minutes by failing to grab a towel, which is exactly the kind of thing Douglas Adams would find funny.

What text adventures added to the CYOA formula was a sense of spatial presence. You weren't just choosing between two options at the bottom of a page. You were navigating a map, collecting inventory items, solving puzzles that required you to combine objects in clever ways. The trade-off was that your choices were constrained by the parser. You could only do things the game recognized. Type "pet the dog" and if the designers didn't anticipate that, you got the dreaded "I don't understand that" response. Freedom of expression, but only within the bounds the programmer anticipated.

This is where the pattern starts to emerge. Each step in the evolution of interactive storytelling gives you something and takes something away. Text adventures gave you spatial exploration and persistence, but took away the unlimited imagination of a book and replaced it with a vocabulary list.

The golden age of graphic adventures

Sierra On-Line and LucasArts took the next step in the late '80s and early '90s. Instead of typing commands, you pointed and clicked. Instead of imagining the scene, you saw it. King's Quest, Space Quest, Monkey Island, Day of the Tentacle, Gabriel Knight, Full Throttle. If you grew up in that era, these names are burned into your memory alongside a lot of unresolved frustration about moon logic puzzles.

The Sierra games, designed largely by Roberta Williams and her team, were gorgeous and brutal. King's Quest V would let you walk past a crucial item with no indication that you needed it, then trap you in an unwinnable state eight hours later. They inherited the CYOA death-around-every-corner philosophy, except now dying meant losing hours of progress instead of flipping back a few pages.

LucasArts took a different design philosophy. Ron Gilbert declared early on that their adventure games wouldn't have dead ends or arbitrary deaths. You could click on everything, try everything, and the game would respond with humor rather than a Game Over screen. Monkey Island let you do things like challenging a pirate to insult sword fighting, and losing just meant you got a funnier insult. The puzzle difficulty was still real, but the game was on your side.

Both approaches were valid, and they produced genuinely great storytelling. Grim Fandango is one of the best-written games of its era. Gabriel Knight: Sins of the Fathers has a story that holds up against any crime novel I've read. These games proved that choice-based interactive fiction could produce real narrative art.

But the choices themselves actually got narrower. In a text adventure, you could type anything. In a graphic adventure, you clicked on things. The number of verbs went from infinite (whatever you could type) to maybe nine (look, talk, use, push, pull, open, close, pick up, give). The story got richer. The visual presentation got better. The player's expressive range got smaller. More was authored, less was imagined.

Japan takes the wheel

While Western developers were building puzzle-heavy adventures, Japanese developers were inventing the visual novel. The format is exactly what it sounds like: a novel with visuals. Screens of text, character portraits, background art, music, and periodic decision points that branch the narrative.

Visual novels leaned hard into the "novel" part. Where Western adventures were about puzzles with story wrapped around them, visual novels were about story with choices dropped into them. A typical visual novel might have you reading for twenty or thirty minutes between decision points. The choices felt weightier because you'd invested so much in the narrative by the time they arrived.

Steins;Gate is the one I always point people to when they ask what a great visual novel looks like. It's a time travel story that starts slow, builds methodically, and then hits you with emotional moments that land precisely because you spent hours getting to know the characters. The branching system ties directly into the time travel mechanics. To reach the true ending, you need to make specific choices across multiple playthroughs, essentially traveling through alternate timelines until you find the one that works. It's the CYOA concept elevated into something structurally brilliant.

The visual novel format also demonstrated something important about choice in fiction: sometimes fewer choices hit harder. A CYOA book might give you forty decisions. A visual novel might give you six. But those six decisions in a visual novel can reshape the entire story in dramatic ways, because the writers invested enormous effort into making each branch feel distinct and complete. Quantity of choices turned out to be less important than quality.

Telltale and the illusion that worked

Telltale Games figured out something clever in 2012 with The Walking Dead: you don't actually need to give players real choices. You just need to make them feel like their choices matter.

I know that sounds cynical, but stay with me. Telltale's games tracked your decisions visibly. "Clementine will remember that." Other players' choices were shown at the end of each episode as percentages. 73% of players chose to save this character. 27% chose the other one. The social element made you question your own decisions in a way that the actual narrative branching didn't fully support, because most Telltale storylines converged to similar outcomes regardless of what you picked.

And yet it worked. The emotional impact was genuine. Choosing who to save when both options feel terrible produces real anguish, even if the long-term plot doesn't change much. The Walking Dead's first season made me cry at the end, and I knew the major beats were probably the same for everyone. It didn't matter. The feeling of having made those choices, of having owned those decisions, was enough.

Life is Strange refined this approach and added the wrinkle of time manipulation. Max can rewind time and see the immediate consequences of a choice before committing to it. This sounds like it would remove tension, but it actually increases it, because you're making an informed decision and you still don't know what the long-term effects will be. Seeing that one option makes your friend happy right now but the other might be better later, and having to choose anyway, is agonizing in the best way.

Detroit: Become Human went maximalist. Three playable characters, branching paths that can diverge wildly, characters who can permanently die midway through the game. After each chapter, the game shows you a flowchart of every possible path, and seeing how many nodes you missed is both satisfying and maddening. Detroit probably represents the upper limit of what's feasible with hand-authored branching content. The amount of writing, acting, and production that went into all those paths is staggering, and even then, some branches feel noticeably thinner than others.

The Telltale era (and its descendants) showed that emotional investment matters more than branch count. A game with three meaningful choices can outperform a game with three hundred hollow ones. But it also showed the ceiling of authored interactive fiction. Even with massive budgets, there's only so much content a team can create.

AI blows the doors off

AI Dungeon launched in late 2019 and completely broke the model. Instead of choosing from authored options, you typed whatever you wanted and a language model generated the next part of the story. No limitations. No predetermined paths. No "I don't understand that" parser failures. If you wanted your character to open a portal to the moon and start a cheese factory, the AI would roll with it.

I played a lot of AI Dungeon when it first came out, and the experience was unlike anything I'd had before. The highs were genuinely creative. The AI would pick up on threads I'd started and weave them into surprising directions. A throwaway mention of a mysterious letter would become a whole subplot three turns later. When it worked, it felt like collaborative storytelling with an infinitely patient co-author.

When it didn't work, it was a mess. Characters would forget their own names. Plot threads would evaporate. The tone would lurch from serious drama to slapstick for no reason. The AI had no real understanding of story structure, pacing, or payoff. It was generating text that sounded like narrative without having any of the intentionality that makes narrative satisfying.

NovelAI came along and addressed some of these problems with better models and more user control over style, tone, and content. Character.AI took a different approach entirely, focusing on conversational AI characters that you interact with through dialogue rather than traditional narrative. The experiences these platforms create are genuinely new. They're not quite games, not quite stories, not quite conversations. They're something that doesn't have a name yet.

The fundamental trade-off is the same one that's been operating since 1979. AI gives you infinite choices, infinite freedom, zero constraints. But it takes away authorial intent. Nobody designed the story you're experiencing. Nobody carefully placed that plot twist for maximum impact. Nobody spent months revising that emotional climax until it hit just right. The AI is improvising, and improvisation is exciting but inconsistent. A great improv comedian can make you laugh harder than any scripted joke. But you'd never replace a tightly written movie with a two-hour improv show and call it equivalent.

What gets lost, what gets found

Here's what I keep coming back to. Each generation of interactive storytelling traded something for something.

Books gave you infinite imagination but limited interactivity. Text adventures gave you spatial exploration but constrained your vocabulary. Graphic adventures gave you visuals but narrowed your verbs. Visual novels gave you deep emotional storytelling but reduced your agency to a handful of pivotal choices. Modern narrative games gave you cinematic production values but often disguised limited branching as meaningful choice. AI gives you total freedom but no quality guarantee.

None of these are objectively better than the others. I still reread Fighting Fantasy books. I still play parser-based interactive fiction from the Infocom era. I still think Grim Fandango has better writing than 99% of modern narrative games. And I still fire up AI Dungeon sometimes when I want that feeling of pure, unconstrained possibility, even knowing that the story quality will be a coin flip.

The people making the next generation of interactive fiction are going to have to figure out how to combine AI flexibility with authored quality. That's the hard problem. Give the player real freedom, genuine unpredictable choice, while still delivering a story that feels intentional and crafted. Some developers are already working on this, using AI as a tool to fill in the gaps between authored content, generating minor dialogue and side interactions while keeping the main narrative beats human-written.

Whether that hybrid approach works remains to be seen. But the desire that started all of this, the kid with two fingers wedged in a paperback who wanted to control what happened next, that's not going anywhere. It's the same impulse that drives someone to boot up Detroit: Become Human or type a wild prompt into AI Dungeon. We want stories that respond to us. We've wanted that for fifty years, and every generation of technology has tried to deliver it in a new way. The best version probably hasn't been built yet. But the lineage is clear, and every step taught us something about what choice in fiction actually means.