What can AI reveal about Bob Dylan's songs? Mapping Bob Dylan’s mind (aeon.co)

A researcher fed Bob Dylan’s official lyrics (roughly 500 songs from 1962–2012) into a large language model to perform “distant reading,” extracting pairs of related concepts as nodes and linking them with edges that the model labeled literal or metaphorical and tagged for emotional tone. That process produced a merged network of about 6,000 unique nodes and 9,000 edges, analyzed decade-by-decade with network-science tools (similar to those used in epidemiology and social-media analysis). The pipeline treats each notable concept as a node, records co‑occurrence edges per line, and systematically quantifies motif recurrence, metaphorical density, and sentiment across Dylan’s catalogue. The findings show what quantitative literary analysis can reveal: Dylan’s lyrics become steadily more figurative over six decades—metaphorical edges rise from ~60% in the 1960s to beyond 75% by the 2010s, climbing roughly 1–2 percentage points per album—and metaphorical language skews more negative in tone than literal lines. The model also maps thematic shifts (a protest/political surge in the 1960s, intimate/romantic peaks in the 1970s, darker mortality themes in the 1990s) and tracks how images and biblical references reappear in new contexts. The work demonstrates how LLMs plus network analysis offer objective, scalable complements to close reading—revealing patterns and evolutions in an artist’s corpus that are hard to perceive by human memory alone.