Matthew Sims
Fig. 1 The study, morning 1/4

Matthew Sims

AI researcher · Counterfactualist · Explorer of imaginary worlds

Once upon a time

Chapter I

Once upon a time

Once upon a time, an English PhD was introduced to machine learning. From then on, he was hooked.

He did an NLP postdoc at UC Berkeley, became a researcher at Netflix, was the first hire at Sudowrite, and started and led the applied AI team at Spotter.

One day, he decided to found a research and product lab called Better Engines, to explore how to augment creativity with new storytelling tools.

Fig. 2

The garden of forking paths

Canon

The path actually taken is drawn in red. Every grey branch is an unexplored possibility.

Chapter II

The engine

Better Engines is a research and product lab exploring how to augment human creativity with narrative AI.

betterengines.ai

§09The black pagecf. Sterne, Tristram Shandy (1759)

Fig. 3 The engine room. Paper in, birds out.
Fig. 4 The premise engine Model 0.1 · hand-cranked

In , wants to , but .

160,000 possible premises.

Reading — genre: temporal noir · tone: bureaucratic · worth reading: up to you

Fig. 5 A world, held together by its threads.

Chapter III

A world that remembers

Arc is for narrative worlds that remember, develop, and hold together across time.

In development · coming soon

Chapter IV

The thread

A brief history of narrative intelligence: a visual essay on sixty years of storytelling machines, from ELIZA to now.

Read the essay
Fig. 6 The library. Follow the thread.
Fig. 7 The thread, 1966 – now

  1. 1966

    ELIZA

    A program that understood nothing made people feel understood.

  2. 1976

    Adventure & Zork

    The machine held a world steady. The player walked in and came back with a story.

  3. 1991

    The quest for drama

    Could a machine improvise a story with a person? A generation spent fifteen years finding out.

  4. 2000

    Emergence

    Nobody programmed stories, only rules. Players found narrative anyway.

  5. 2017

    Fluency

    Machines learned to write. That turned out to be the easy part.

  6. Now

    Worth reading

    The question was never whether machines could write. It was whether they could help someone tell a story worth reading.

    Read The Thread

Epilogue

Currently: building Arc, running Better Engines, and pulling on the thread.

Fig. 8 The study, night. The thread, wound back on its spool.