I’m a Noted Music Critic. Can A.I. Do My Job?

When TIDAL asked me to write an essay about whether Artificial Intelligence could fulfill the role of the music critic, my first thought was “Can I get A.I. to do the piece for me?” I turned of course to ChatGPT, the program that has stirred up a lot of, well, chat recently. (It’s currently available to the public free of charge, the idea being that users’ interactions with the chatbot will help its developers at the company OpenAI improve it.) I made the instructions as simple and straightforward as possible:

Write an essay in the style of music critic Simon Reynolds that expresses skeptical views about A.I. taking over the role of the music critic.

Within seconds, the program served up 200 words of disconcertingly clear and well-organized argument. The effect was at once mind-blowing and underwhelming. Although millions of my own sentences can be found on the internet, the program proved unable to duplicate any stylistic mannerisms. The argument itself struck me as averagely intelligent, making entry-level points about how A.I.-generated prose is necessarily deficient in empathy and nuance, and how it would lack the unique and personal perspective of a human critic. (I had to give the chatbot points for self-awareness, at least.) Similar formulations on the same topic, substituting the names of music journalists with highly individual prose voices, produced equally neutral and characterless results.

The program is addictive, with tremendous scope for time-suck. You keep thinking that if you bang away at it, trying out ever more outlandish approaches or finely tuned commands, it’ll suddenly and dramatically improve. But I found that the “voice” remained consistent: earnest, plodding, attuned to bland generalities rather than arresting specifics, and irritatingly fair-minded. Not promising attributes for a critic! 

The even-handedness is probably the most significant defect, when it comes to the prospect of a chatbot usurping people like me. Asked to compose a fierce critique of a particular record, it produced the prim rejoinder “I am not programmed to write negative reviews …  my primary goal is to be helpful and supportive.” This is not the primary goal or proper role of a critic: brutal honesty, to the point of unkindness, is closer. But the truth is that ChatGPT is just as incapable of writing a rave review. Instead, it surveys the range of existing viewpoints and gathers a balanced array of pros and cons. It sits on the fence.

Some pundits suggest that writers will use A.I. as a sort of sous-chef, a labor-saving subordinate that handles the grunt work of research and pattern-finding, freeing up the human head chef to lavish their skill on the delicious dish of prose itself. But when it comes to the basic procuring and preparation of ingredients, this particular program is far from infallible. Asked to discuss the political resonances of “Still Ill” by the Smiths, it referenced a non-existent video for the song (which was never a single). More worryingly, it confused Nia Archives, a young Black British singer-songwriter-producer who weaves her voice and melodies into drum & bass rhythms, and Nia Andrews, a Black American singer-songwriter-producer. Attempts to get it to critique Bob Dylan’s The Philosophy of Modern Song resulted on one occasion in a review that treated the book as if it were a record: “At the heart … is Dylan’s signature voice, a raspy, expressive instrument that has always been one of his greatest assets … delivering a series of powerful, emotive performances that are equal parts confessional and confrontational.” When, inevitably, I asked for commentary about myself, I learned that I had authored The Blissed Out Guide to Trance, a non-existent book. A separate inquiry about “rockism” revealed that the concept had been invented by Simon Reynolds. Not true, but for a moment I was half-convinced the program was clumsily trying to butter me up.

No, I would never rely on it as a research tool.

Spending time with this particular A.I. entity often reminded me of situations where someone you know has recently read a long, in-depth article on a subject they otherwise know nothing about. (And by “someone you know” I mean “all of us.”) They, or you, will regurgitate what’s been ingested about, say, atmospheric rivers, or cryptocurrencies, but there’s no extra level of perception or insight: It’s all second-hand, and the chances are some of it has gotten garbled in the memory.

Professional writers might not have a lot to worry about as things stand, but teachers have more reason to be concerned: A.I. can effortlessly produce a B-level paper on any subject, with better grammar and spelling than most students in their late teens and early twenties can manage. But the flair, the zesty flourish of real enthusiasm that pushes an essay into the A-zone, is beyond it. 

Even before embarking on my experiments, I had worked out that certain aspects of music journalism can’t be duplicated by A.I. The live concert review, for instance, requires observational alertness and an ability to process and describe the gestures of performance, subcultural codes and rituals, and any number of other elements that create a sense of occasion. An in-person interview likewise requires attentiveness to conversational inflection and facial expression, a grasp on how details of behavior can reveal personality quirks, a feeling for the ambience of a location, and other intangibles of vibe. Given that A.I. sorts through and synopsizes the existing body of opinion and analysis in the world, it’s hard to see how it could review a new record, much less a brand-new artist. 

But there’s a more profound problem with the idea of the artificially intelligent critic. A.I. does not hear the music it comments on: It has no ears, no body. This is especially problematic with pop music, the domain of desire and dance. But all music works on us through our sensory and kinesthetic faculties. Classical music is largely a stranger to funk, but even at its most avant-garde or loftily spiritual it still has rhythm, still pulls at our bodies. Music that seems to transcend the flesh creates that feeling through physically felt sensations of ascent and ethereality. Until a form of A.I. is developed that involves a simulation of embodiment — the complex interactions of hormones, brain chemistry, heart rate, respiration, neurology and more — it’s hard to imagine how it could duplicate that aspect of music writing that leans on a sensual response. Music isn’t just patterns of information; vibrations ripple through us materially. And not just the obvious gut-pummeling forces, like heavy bass or pounding beats, but the subtle and soft high-end frequencies, too. Then there’s the synesthetic effects of sound, the way certain kinds of music trigger visual images.  

The understanding gap also applies to what music is about: love, loneliness, anger, restlessness, despair, euphoria, nostalgia. These have no felt reality for A.I., because it has no experience to draw on. How, then, could it appreciate a rendering of these emotions by a songwriter — let alone be able to differentiate and judge the expressive tools that musicians use to dramatize such feelings?

On an even more fundamental level, the problem is that A.I. has no need for music. It is designed to simulate a kind of detached interest in the subject, or any subject. But music fans — and the critic is essentially a professional fan — have an almost biological dependency on music. Back in the day, you’d hear of people who were such fiends they’d skip a meal to afford a certain record they just had to have. While I’ve never done that, in my youth I’d go without other things that people spent their disposable income on — nice clothes, alcohol, travel — because music was the overwhelming priority. For a fan, the music you love is essential to your being. But you also have a bottomless hunger for sonic discoveries — a constant craving for new things to be in love with. The hollowness of A.I. writing stems from only being able to observe this from the outside. There’s no gut understanding of the way fans identify with stars, or how music can be so unbearably involving and exciting you just have to project yourself inside it by miming air guitar. 

Finally, A.I. has no need to write, either — no deep-seated motivation to put words on paper or on screen. The kind of texts it generates resemble what I think of as “motiveless” writing, like school homework, or advertorial. Proper music criticism, even if done to earn a living, is closer to the sort of willed writing that fills diaries, journals and poems — where the compulsion to write is internal rather than externally imposed. In writing for public consumption, a jumble of urges are at work, among them communication, confession and competition. But there’s also self-pleasure: You are the first reader of your own writing; you see these sentences spring into existence in real time. Just like when you read the work of writers you admire, there’s the thrill of the sudden moment of clarity, the a-ha feeling of recognition and rightness in an image or analogy.

A.I. has no ears to hear music, but it also can’t “hear” its own writing, which is why it exhibits minimal feeling for the musicality and rhythm of prose — the way the deployment of an unusual adjective or adverb can be electrifyingly dissonant. Until a form of A.I. is invented that can simulate the narcissism and exhibitionism of human writers — a chatbot that enjoys swaggering across a page and laughing at its own jokes — then I humbly suggest we of flesh and blood have little to worry about.