Former Stability AI Audio VP, Ed Newton-Rex, Unveils ‘Fairly Trained’ with a Splash of Wit.
“Ed Newton-Rex, ex-VP of Audio at Stability AI, announces ‘Fairly Trained’”
“Ed Newton-Rex, a former VP of Audio at Stability AI, recently took to Twitter to announce the begrudging adoption of ‘fairly trained’ artificial intelligence within the industry.” Oh boy, once again the tech world is just doing ‘enough.’
It’s quite the scene, folks. Our pal Ed Newton-Rex, erstwhile VP of Audio at Stability AI, trots out the term ‘fairly trained.’ A phrase so delightfully ambiguous it could sugar-coat a pepper. Sprinkled atop our usual jargon-encrusted AI cake, it adds an extra dash of confusion to the mix.
Take a moment, kindly imagine ‘fairly trained.’ What springs to mind? A dog that only occasionally pilfers your slippers for its chomping pleasure? A teenager who puts the dishes in the dishwasher…eventually? Well, in the case of AI, it seems it now equates to algorithms that are no longer running amok with unchecked bias. We can’t help but wonder if ‘fairly trained’ is the AI version of ‘doing the bare minimum.’
Newton-Rex further elucidates this premise in his little Twitter confessional. It seems they’ve been working hard – bless them – to remove ‘known biases.’ Notably, no mention of the unknown biases that are part and parcel of the mysterious world of AI.
He plucks an example of bias in voice assistants, claiming they respond better to male voices than female. The solution? The algorithm has been reconfigured to treat female pronunciations as the ‘default.’ Now, there’s a twist. This is like discovering your coffee has been secretly laced with decaf.
Might it be suggested that instead of merely tweaking jackhammer-decibel male bias to default to female, it might be an idea to craft an AI that is—dare we say—equally sensitive to both? An AI that doesn’t need a gender-default, but simply understands human voice with striking clarity, irrespective of the speaker’s gender. Now that might qualify as more than ‘fairly trained.’
But perhaps it’s unfair to expect such leaps from AI. After all, it is just artificial intelligence, duly hobbled by its creators’ limitations. Speaking of creators, it would be enlightening to understand their perception of ‘fair.’ Could it be that ‘fairly trained’ means just enough to slide them out of the line of ethical fire?
Well, Newton-Rex does commendably accept that the tech industry hasn’t quite arrived at the promised land of fairness. Acknowledging that work needs to be done is always the first step, right? Let’s hope the as-yet-unmentioned second step includes more than the mere mitigation of ‘fairly trained.’
That being said, at least they’re trying. Yes, their efforts might just be at the fairly decent level. But as Newton-Rex points out, ensuring AI fair training is more than merely desirable. They’re making moves in the right direction, even if baby steps. Is ‘fairly trained’ industry-speak for ‘we’re working on it?’ We suppose only time will tell. Position your popcorn folks, this is going to be an interesting watch.