From Frankenstein to Farallon

Two hundred years of machines that learned to sing, and the people who taught them.

Prologue: the question

• 1816: Mary Shelley begins Frankenstein during the Year Without a Summer

I. The machine makes sound, 1843–1961

• 1843: Ada Lovelace predicts the Analytical Engine might compose music
• 1939: Bell Labs shows the Voder, built on Homer Dudley’s 1928 vocoder
• 1951: Australia’s CSIR Mk1 publicly plays “Colonel Bogey”; later that year the BBC records the Ferranti Mark 1, the oldest surviving recording of computer music
• 1957: The Illiac Suite, the first score composed by a computer
• 1961: An IBM mainframe at Bell Labs sings “Daisy Bell”, later echoed by HAL 9000

II. The machine becomes a performer, 1969–2007

• 1969: The Archies, a cartoon band, top the Billboard Hot 100 with “Sugar, Sugar”
• 1978: Synthesizers reach pop through the 1960s and ’70s; Kraftwerk releases “The Robots”
• 1983: MIDI gives computers and electronic instruments a common language
• 1985: Jem and the Holograms and Synergy, the holographic computer
• 1998: Auto-Tune goes audible on Cher’s “Believe”
• 2007: Hatsune Miku, the Vocaloid virtual singer

III. The machine learns style, 1987–2020

• 1987: David Cope’s Experiments in Musical Intelligence learns composers’ styles and writes new works from them
• 2016: Google launches Magenta, and AIVA begins composing with neural networks
• 2018: Google’s Music Transformer brings transformer models to music and holds a theme for minutes
• 2018–2020: Pioneers train their own AI: Taryn Southern’s I AM AI, Ash Koosha’s Yona, Holly Herndon’s PROTO, YACHT’s Chain Tripping, Dadabots’ Relentless Doppelganger and Arca’s Bronze instances
• 2020: OpenAI’s Jukebox generates raw audio, singing included

IV. The machine takes instructions, 2023–today

• 2023: Google’s MusicLM and Meta’s MusicGen turn text descriptions into music
• 2023: “Heart on My Sleeve”, with AI vocals imitating Drake and the Weeknd, goes viral and is pulled from streaming
• 2023: Machine learning helps finish the Beatles’ “Now and Then”
• 2023–2024: Suno and Udio turn a typed sentence into a finished song
• 2024: Universal, Sony and Warner sue Suno and Udio over AI training
• 2025: Breaking Rust, an AI-generated country artist, tops Billboard’s Country Digital Song Sales chart
• Today: Google DeepMind’s Lyria 3.5, Suno, ElevenLabs’ Eleven Music and open-source models such as MusicGen, Stable Audio Open, YuE and ACE-Step

AI and creativity: what the research says

A 2024 Nature Human Behaviour meta-analysis of 106 experiments found human–AI teams gained the most on content creation. A PNAS Nexus study of more than 4 million artworks found text-to-image tools raised artists' output by 25% and favorites per view by about 50%. A Science Advances study found AI ideas made stories more creative, especially for less experienced writers, but also more alike. The U.S. Copyright Office says AI-assisted work is protected where a human decides its expressive elements.

We love both: traditional art and new tools. Farallon is made with generative AI tools under human creative direction: the books, themes, lyrics, arrangements, editing and releases are human-led.