bewildering brain

Bewildering Brain explores whether machines can write like poets — specifically Bob Dylan and Luis Alberto Spinetta. I trained both Markov Chains and Recurrent Neural Networks (RNNs) to generate lyrics in their styles and compared their outputs.

The lyrics were collected using the Genius API and Spotify’s Web API. The goal was not only to see which model performed better technically, but also which one better captured the essence and tone of each artist.

Two modeling approaches were used:

• Markov Chains – predicting each word based only on the one before it, creating semi-random and often surprising lyric chains.
• RNNs (LSTMs) – trained to retain longer-term patterns, producing more cohesive lines that sometimes felt eerily human.

The result was a creative experiment that sat somewhere between data science and digital art. It sparked curiosity across tech, media, and music circles — and showed how machine learning can go beyond analytics and into culture.

For technical details, code, or to remix it with your own favorite artist, you can check out the github repo.

To read the Medium article in Spanish published in Ciencia y Datos about L. A. Spinetta, go here.

To read the one published in Towards Data Science about Bob Dylan, go here.