Lately my YouTube feed has been overrun with AI singers — whole tracks, start to finish, without a single human voice in them. For someone as tone-deaf as me, it was a humbling education. There's a whole genre of it now: feed a few lines of lyrics into a tool like Suno and out comes a song in a smoky, lived-in voice that lands every note dead on pitch. One of them gave me genuine goosebumps. I sat there listening and thought: singing, as a profession, is about to be redefined.
I know nothing about music and I'm in no position to judge singers. But singing is hardly the only trade AI is quietly redrawing. The one I actually know something about is my own — academia. And the fight that has raged hardest there these past years turns out to hinge on the very thing a singer lives and dies by: pitch. Or, in our language, Key Performance Indicators — KPIs.
The academic high-note contest
So what is an academic KPI? Academia is an industry like any other, and industries need numbers. A professor has to publish a certain quota of respectable papers a year to clear the bar the university sets. Universities especially want you in the top journals — Nature, Science — because a large slice of the global rankings comes down to exactly those papers.
None of this is abstract in Malaysia. Sin Chew ran an investigative series on it not long ago, "The Numbers Game in Academia," and it dug up some strikingly concrete thresholds. At Universiti Malaya, for a senior lecturer to make associate professor, their H-index (a measure of scholarly impact) has to reach 6 and their research funding has to top RM 550,000; to climb one rung higher to full professor, the H-index has to hit 11 and the funding has to accumulate to RM 1.2 million. Those numbers are the discipline's pitch line.
And the pressure runs all the way down. PhD students need enough papers to graduate. Undergraduates get pushed to publish so they can apply for PhDs. Even high-schoolers, angling for a spot at a top university, have started racing to get published. So everyone on every rung is clawing toward the tip of the same pyramid, grinding out metrics day and night. Picture a whole music industry crammed onto one stage, every last singer straining for the highest note and running their trills up into the stratosphere, all just to make a judge glance their way.
We insiders have a darker, private name for it. In old Chinese folklore there's a practice called raising gu: you seal a jar full of venomous creatures and let them devour one another until only the single most poisonous one crawls out alive. That survivor is the gu king. Anyone who has fought their way to a post at a top school is, more or less, a gu king who made it out of the jar.
Push the belting far enough and voices crack. The same Sin Chew series turned up the ugly numbers: in a late-2023 Nature analysis, Malaysia's paper-retraction rate ranked sixth in the world, and our name keeps surfacing near the top of unflattering lists — "extreme publication output," "abnormally high self-citation." Years ago, Universiti Malaya's engineering faculty was caught red-handed handing staff an actual KPI to cite one another's papers, a few colleagues per article, purely to fatten the ranking figures.
Nobody sets out to sing badly
Everyone who walks into academia arrives carrying some version of a dream. And round after round of filtering tends to grind that dream down into a machine for producing papers.
It's a lot like singing. Every singer starts with an artistic dream too, but the moment they step onto the competition stage, if they don't belt a high note or throw in some vocal acrobatics, not one judge turns a chair. Plenty of people lose themselves somewhere in that long scramble to be seen.
Let me be clear: I have nothing against KPIs. In a Sin Chew interview a few years back, I called myself a "KPI dove." A KPI is a bit like pitch — a singer who can't hold a note is a non-starter. Every year I get pulled into some televised singing contest, and while I can't resist griping at the screen, I'm reminded each time that you can't skip the fundamentals. I played this high-note game hard myself back in the day, so none of what follows is me lecturing from the cheap seats.
The paper itself I have no quarrel with. Writing one forces you to take the fog in your head, compress it, sharpen it, and send it out into the world (regular readers will recognise "compression" from an earlier column). What I object to is treating the KPI as the only ruler there is, and measuring a whole person against it.
The moment AI opens its mouth, the contest is exposed
In the AI era, this problem gets blown up to the point of no return.
I've argued in earlier columns that AI is still a long way from AGI. But if all you need is to clear the academic KPI bar, that's trivial for it. Today a reasonably capable researcher, with AI riding shotgun, can turn out in a week — a day, even — a paper that's nothing earth-shattering but has "nothing obviously wrong with it" either.
Think of it this way: belting high notes and running impossible trills used to be the mark of a rare, gifted singer. Now AI opens its mouth and the pitch is flawless, the high notes come free, and it never gets tired. So here's the problem. When an undergraduate applying for a PhD is already holding a dozen papers — more than plenty of junior faculty — the paper as a currency has simply been devalued.
It's the same logic as a schoolkid doing homework with AI. If everyone can score full marks with a chatbot, what do full marks prove anymore? The question just shows up more starkly under academia's finely calibrated system of metrics.
So what's actually worth competing on?
This bind isn't really new, and it certainly isn't AI's fault. AI has only magnified it, blown it up until you can no longer pretend not to see it. The fix is simple enough to state: once the basic KPI is cleared, in the end you have to look at the person. The word I reach for most these days, when I talk with students and when I pick them, is taste. Scholarly taste.
It sounds vague. It isn't. People who can write a competent paper are thick on the ground. Every year I sit on a selection panel and look out at a sea of straight-A national top-scorers and Olympiad gold medallists, and we joke among ourselves: by this standard, how did I ever get in?
Because, as with singing, technically competent performers are everywhere, and the ones who can actually move you are rare. That missing sliver is taste. In research, the questions you could ask are endless and the papers you could write are endless, but not every one of them is worth writing — the ones you could dash off without thinking should just be handed to AI. Whether you asked the right question in the first place: that is what taste is.
How do you tell whether someone has taste?
I'll admit this can't be made fully objective. But here's the interesting part: long before AI, top universities admitting students never cared all that much about how long your publication list ran. They cared far more about a face-to-face conversation, because a few minutes of one tells you whether there's a real person in there.
Not long ago I helped judge a book prize in Malaysia, and as an interviewer I went in hard — every question with no clean answer, things like "Do you think AI is destroying secondary education?" What I found was that the people who could actually catch the ball and keep the rally going were few.
This matters more than ever in the AI era. Singing has its masters and apprentices, and what a good master ends up passing on isn't technique — it's the ear, that fussy, discriminating ear. Same with my PhD students. On coding, AI is far better than I am; they stopped needing me to hold their hand a long time ago. The one thing I still have to give them is a question I believe is genuinely worth asking.
Give two students the same AI and watch what happens. Some get stronger by the month, because AI is a fine tool — it can hold you up while you reach for problems that used to be out of reach. Others rush to pad their output, and the more they lean on it the more they lose themselves. In the AI era everyone has ten times the firepower. But that firepower shouldn't go into writing ten times as many papers. It should go into asking questions ten times as hard.
Can we actually turn the ship?
Anyone with a shred of idealism left about scholarship would nod along to all of this. But turning the ship is easier said than done. "KPIs above all" won't vanish overnight, and to an outsider, more papers simply look better than fewer.
And yet a real singer was never in it for the judges' scores — she's in it for the singing. Someone who does scholarship ought to have a little of that in them too. There's an old Chinese poet, Tao Yuanming, who resigned his official post rather than, as he put it, "bow for five pecks of rice" — grovel for his salary. That kind of purity sounds naive, and, honestly, it used to be a luxury: skip enough KPIs and you'd forfeit the right to stay in the game at all.
Which is where AI offers a strange, almost hopeful angle. When papers can be mass-produced by AI and scores stop meaning much, the KPI that has been clamped around everyone's throat might finally loosen. And when it does, we could be forced — liberated, really — out of the habit of counting papers and back into the older, harder work of appreciating a person's taste. Seen that way, AI isn't necessarily here to take our jobs. It might be here to hand a little freedom back to the people who just want to do the work.
Coda
My mother's favourite refrain these days is, "AI is doing so much harm — why on earth do you still study it?" The trouble AI brings really is vast and tangled, and it really has squeezed young people's room to breathe another notch tighter. But on this one question, at least, what it magnifies may not be all bad — because it forces us to ask, all over again, what is actually worth anything once the KPI bar has been cleared.
So how do you grow taste? The old-fashioned way. You read.
It's just that in the age of AI, one has to wonder who is still willing to sit down long enough to finish a book.