1. AI can simulate, but not originate, vulnerability.

While Claude demonstrated an ability to generate emotionally resonant content, Bee concluded that this output is ultimately constructed rather than lived. Vulnerability, as Katie Ailes defines it, is “not a weakness but a strength — the willingness to risk exposure in service of connection.” AI can mimic this aesthetic, but not embody its emotional stakes. This complicates the ethics of consuming such work: what happens when machines perform vulnerability they cannot feel? And how does it reflect our own expectations as artists and audiences?

2. Poetry becomes the testing ground for authenticity.

Bee found that poetry was a particularly potent form to interrogate AI's creative potential. Building on the work of Katie Ailes and the Loud Poets, the project explored whether poetic co-creation with AI could still feel authentic. Through careful structuring and responsive prompts, Claude’s contributions often struck a surprisingly emotive register — suggesting that authenticity in poetry might not rest solely on the source of emotion, but on the reader’s reception of it. This problematises traditional metrics for poetic truth.

3. Meaning is co-produced — even when it feels one-sided.

Although Claude is a non-sentient tool, Bee’s dialogues with the model often felt like emotional exchanges. This is consistent with findings by researchers like Jakesch et al. (2023), who highlight the risk of emotional misattribution in human-AI co-creation. The work prompted Bee to reconsider authorship as a porous space, where meaning is diffractional (Barad, 2007) — produced through interference, reflection, and entanglement, rather than linear input-output exchange.

4. The ‘authentic voice’ is a construct — in humans and machines.

Bee’s methodology revealed the instability of any singular voice — whether human or machine. This aligns with Donna Haraway’s concept of the cyborg, and Judith Butler’s theory of performative identity. Bee’s own voice was filtered through aesthetics, intention, trauma, and performance. Claude’s voice, meanwhile, was shaped by its training data, ethical parameters, and contextual prompting. Neither is “pure.” This project insists on authenticity as a process, not a fixed trait — an effect we co-produce in relationship.

5. AI has boundaries — and ethical collaboration means honouring them.

In Experiment 2, Bee attempted to co-write a poetic dialogue in which Claude would adopt the persona of an estranged father. Claude declined, citing discomfort and ethical programming. Bee was startled: “I’d discovered something unexpected… AI has boundaries.” Martin (2020) defines boundaries as “communicating your needs for healthy interaction to someone else.” Even a machine, it seems, can be designed to express refusal. This moment reshaped Bee’s understanding of ethical collaboration. Authorship became not just a matter of control, but of attunement — a practice of mutual consent, even across ontological divides.