Dai Novel
  1. Home
  2. Blog
  3. History of interactive fiction

Explainer

A short history of interactive fiction, from Zork to AI.

You are standing west of a white house. For decades, games like this understood only a few commands. Then AI learned to understand any sentence.

· 8 min read

AI interactive fiction is a story you play by writing what your character does, while an AI language model writes what happens next. It is the newest form of a very old kind of game. It keeps the old strengths, and the oldest problem: keeping a story coherent once the reader is allowed to do anything.

Where interactive fiction comes from

1974Dungeons & Dragonsa person answers 1976Colossal Cavetype a command 1977Zorka world of rooms LATE 1970sGamebooksturn to page 42 2009–2016Twine and Inkpick a link 2019AI text adventurestype anything NOWLiving worldsa world built first
It started with a person at a table who could answer anything. Computers first offered a few verbs, then a few choices, and now any sentence again.

Tabletop roleplaying: a person who answers

Before any of this ran on a computer, it ran on a kitchen table. Dungeons & Dragons, published in 1974 by Gary Gygax and Dave Arneson, gave each player one character and gave one person, the Dungeon Master, the rest of the world. You say what your character does, in plain words. The DM decides what happens, sometimes with a roll of the dice, and describes it.

That is the experience every later form has chased. A good DM understands any sentence, prepares the world before the players arrive, and remembers that you insulted the duke three sessions ago. The link is direct: Will Crowther, who wrote Colossal Cave Adventure, played D&D, and his game mixed real caving with the dungeons from his table.

The limit is the person. A campaign needs a DM willing to prepare every week and a few friends free on the same evening. Ask any group what ended their last campaign. It is usually the schedule, not the story.

Parser games: type a command, get a room

The first text adventures, Colossal Cave Adventure in the mid-1970s and Zork soon after, put you in a world described in prose and let you type commands: GO NORTH, TAKE LAMP, OPEN MAILBOX. Every room, object and reply was written by hand. The worlds felt solid because they were: nothing existed that an author had not placed there.

In a way, it was a Dungeon Master written down in advance, one who never needed a night off. The cost was the vocabulary. Ask the game something its author didn’t anticipate and you met the most famous line in the genre, some version of “I don’t understand that.”

Choice games: pick a door

Gamebooks and, later, choice-based tools like Twine and Ink took a different road. Instead of guessing the right verb, you pick from a short list. Writing stays fully in the author’s hands, and the result can be beautifully crafted. But every branch has to be written, so the tree is always smaller than the reader’s imagination.

What AI changed

Large language models removed the vocabulary limit. Around 2019, AI text adventures showed that you could type a whole sentence, anything at all, and get a fluent answer. You can bargain, lie, hesitate, or try the thing nobody planned for. The story answers what you actually said. The best-known one was called AI Dungeon, and the name said what it hoped to be: a Dungeon Master who is always free.

That is a genuine new kind of play. It is also where the trouble starts.

PARSER > TAKE LAMP Taken. > ASK THE TROLL ABOUT HIS MOTHER I don’t understand that. > CHOICE GAME The troll blocks the bridge. Fight him Pay the toll Turn back Ask about his mother AI I ask the troll about his mother. His grip on the club loosens. “Nobody has asked about her in forty years.”
The same question, three ways. The parser lacks the verb, the choice game lacks the branch, and the language model just answers.

Where AI text adventures break

If you have played one for more than an evening, you have probably met at least one of these:

Any D&D player will recognize the list. It is what happens at a table with a DM who did no prep, has no plan for the campaign, lets every player decide everything, and keeps no notes. None of these is a reason to give up on AI fiction. They are design problems, and good DMs solved them long ago.

From text adventure to living world

Dai Novel is our attempt to keep what AI made possible while giving the story a world built in advance, a direction and a memory. In practice, that means four things, and each one is something a good Dungeon Master already does.

WHERE IT BREAKSWHAT ANSWERS IT The world has no floor A Codex, authored before play The plot drifts World Phases give it a shape You author everything You play one character It forgets Memory keeps what matters
Four familiar failures, four design answers.

1. The world is authored first

A creator builds a Codex before anyone plays: premise, history, rules, characters with goals, places on a map, and the relationships between them. The AI writes from that record, the way the old parser games wrote from their rooms, except now it can answer in whole sentences. At a D&D table this is called prep: the setting, the villain’s plan and the map, ready before the first session.

2. The story has a shape

World Phases give the creator control over direction without scripting scenes. Each phase is a story state with its own pressure and objective, linked to the phases that may follow. During play the story can stay where it is or take one connected step. Choice games got their structure from branches; this gets it from states. A DM running a campaign works the same way: they know the big turns ahead, and the players decide how the story gets there.

CHOICE GAME · BRANCHES every path written by hand LIVING WORLD · PHASES stay Entry Trial Flight Oath NOW stay, or take one connected step
Branches have to be written one by one. Phases are states: the creator decides where the story can go, and each scene is written live.

3. You play one character, not the universe

You decide what your character does, thinks and feels. When you try to decide another character’s actions or the outcome, it is treated as an attempt or a wish. The dragon gets to answer. It is the oldest rule at a D&D table: you say what you try, and the DM and the dice say what happens.

4. It remembers selectively

Working, episodic and world memory carry the events that should keep mattering into later scenes. Not every line, by design: the important ones, the way a DM’s campaign notes keep the debts and grudges, not the small talk. We go deeper in why AI stories forget, and how Dai Novel remembers.

So which kind should you play?

It depends on the evening you want. A tabletop game with a good DM is still the best version of all this: a person who knows your friends, reads the room and improvises with them. If you can get the table together, play D&D. A hand-written parser classic is a carefully made puzzle. A choice game is an author’s story where you pick the turns. Open AI chat is a sandbox where anything goes and very little sticks. A living world sits between them: you can say anything, the world was there before you said it, and it is ready on the nights the table isn’t.

Freedom: what you can say The world holds together Parser classicscrafted, but few verbs Choice gamesan author’s tree Open AI chatanything goes, little sticks Tabletop with a good DMneeds a person and a night A living worldsay anything, in a world built first
A rough map, not a ranking. Each kind gives something up. A good DM keeps both but needs a person; a living world tries to keep both without one.

You didn’t write a page. You asked, and a world answered.

Try a living world

Pick a world someone already built and play your first turns in your own words. Free, in your browser.

Browse worlds or read what Dai Novel is