Economics PhD student Prashant Garg performed a fascinating analysis of Bob Dylan’s lyrics from 1962 to 2012 using AI. He detailed his project in Aeon:

So I fed Dylan’s official discography from 1962 to 2012 into a large language model (LLM), building a network of the concepts and connections in his songs. The model combed through each lyric, extracting pairs of related ideas or images. For example, it might detect a relationship between ‘wind’ and ‘answer’ in ‘Blowin’ in the Wind’ (1962), or between ‘joker’ and ‘thief’ in ‘All Along the Watchtower’ (1967). By assembling these relationships, we can construct a network of how Dylan’s key words and motifs braid together across his songs.

The resulting dataset is visualized in a series of node graphs and bar charts. What’s interesting is that AI is able to see Dylan’s work through a new lens, something that prior scholarship may have missed.

…Yet, when used as a lens rather than an oracle, the same models can jolt even seasoned critics out of interpretive ruts and reveal themes they might have missed. Far from reducing Dylan to numbers, this approach highlights how intentionally intricate his songwriting is: a restless mind returning to certain images again and again, recombining them in ever-new mosaics. In short, AI lets us test the folklore around Dylan, separating the theories that data confirm from those they quietly refute.