Hot off the Algorithm: Prototyping AI-generated artefacts from the news

A rapid prototype for a Master's module testing whether AI-generated artefacts can deepen engagement with current events. I published v1 as a GitHub Page and rebuilt the pipeline in n8n with outputs to a Notion gallery. The system transforms a headline into an abstract painting and then a sonnet, producing a public stream of paired images and poems.

News products optimise for speed, not reflection.

• Readers skim headlines, rarely forming a durable mental model or emotional connection.
• Need friction to prompt slower, more critical engagement without overwhelming users or adding editorial labour.

An automated loop runs on a schedule to transform headlines and publish results.

Converting a headline into layered, non-literal artefacts-first a painting, then a poem-will encourage pause and reflection.

• Headlines - Select a top headline from major sources.
• Painting - Generate an abstract image via DALL-E 2 guided by theme and mood.
• Poem - Use GPT-4o to produce a sonnet inspired by the painting.
• Publish - Post to a GitHub Page (v1) or a Notion gallery via n8n (v2).

Qualitative analysis revealed thematic and emotional resonance. With tragic events tending to result in darker, more subdued paintings and sonnets and political upheaval often manifesting into fragmented, chaotic output. However, rudimentary quantitative analysis of twenty outputs - comparing headline emotional tone (arousal), image entropy (visual complexity), and sonnet emotional tone - revealed low correlations:

Taken at face value these weak relationships indicate that the prototype may not sufficiently preserve the underlying emotions or themes presented in the source media, though it may be the case that further quantitative analysis would reveal correlations as yet unseen.

• Headline arousal vs. Image entropy: ~0.106.
• Headline arousal vs. Sonnet arousal: ~-0.023
• Image entropy vs. Sonnet arousal: ~0.106