Something is changing in creative work, and it isn’t what most people think. The issue is no longer whether AI can generate reasonably decent artistic work – it can and sometimes strikingly so. The more pressing question is whether it can develop discernment: the capacity to perceive subtle but significant distinctions – not just the ability to create something but an internalized sense of what separates something good from something exceptional.
The training of a violinist offers a useful way to trace how discernment actually develops – and one whose lessons extend well beyond music.
Serious young violinists begin by immersing themselves in the masters. For me, that meant Heifetz, Oistrakh, and Perlman. This exposure helped me distinguish between good and great playing, emotional depth, and mere sentimentality, passion and pyrotechnics.
Here, AI is able to compete. It can absorb staggering exposure – taking in the output of thousands of masters across centuries and traditions. It can compare styles, detect relationships, and identify patterns humans repeatedly find compelling at a magnitude far exceeding our capacity.
But exposure, however wide, is just the first step in cultivating discernment. Violinists must also learn to calibrate – to adjust their own playing in real time against a standard. Early on, that standard is external: the teacher’s ear, the tradition’s accumulated sense of what is right. Students borrow this external standard while their own slowly develops.
Vibrato offers a clear example – that subtle quivering of a finger on a string that gives a note warmth, color, and emotional nuance. Slightly restrained, it can evoke tenderness or intimacy; slightly exaggerated, it tips into sentimentality. I vividly remember a teacher who could instantly hear that difference and taught me to hear it too, adjusting her guidance to wherever I was that day. For a long time, this felt like something only a human could offer.
But AI is beginning to close that distance. Through reinforcement learning from human feedback (RLHF), models are trained iteratively – learning to provide responses that are attuned to the learner.
That said, AI‘s ability to improve through exposure and expert judgement can develop discernment only so far. There is a third and crucial contributor: the sustained wrestling with the physical materials of one’s craft, and with life itself in its many registers.
A violinist, for example, contends with the resistance of wood, string, and horsehair, learning what the instrument will and won’t give. A painter must submit to paint – shaped, as Annie Dillard observed, by the material rather than the other way around. A poet, or in my case, a namer, works against the simultaneously fixed and mercurial nature of language.
And this wrestling isn’t only with the material. The dismantling of the heart after the loss of a loved one, or any experience that fundamentally alters us, demands its own kind of reckoning.
Over time, these physical and emotional engagements – unavailable to AI – metabolize into sensibility, and with it, the capacity to discern what is truly alive from what is merely correct. Without them, even the most extensive exposure and expert judgement can only carry so far.
So if there is a threshold that cannot be crossed, then what remains irreducibly human is not the ability to generate – a power to which AI may well be a formidable claimant – but the ability to know what is worth generating. Without this, AI will produce work that is accomplished, even impressive – but work that never quite crosses into the realm of truly great and lasting art.
