To everyone studying and working on AI:
I just returned from a Hilary Hahn concert at the SF Symphony. For those of you who do not know her, she's one of the most famous violinists alive today. She played three Bach pieces today.
I mostly had my eyes closed as she played the piece. Somehow, during my favourite violin piece of all times, Chaconne from the D minor Partita by Bach, my mind drifted away. It was because I got shivers when she started playing.
The thought was not exactly about AI, but it was relevant. I was thinking about the piano that recreates historical performances. The Traumerei played by Horowitz that made so many people cry; The panache of virtuosity by Lang Lang. From the recordings and machine learning that maps sound into touch sensitivity, the piano plays itself, using the exact rhythm and dynamics of the most iconic historical performances of masters.
I was thinking about whether I would attend a recreated concert of Horowitz. He was my favourite pianist when I was a child. I love his diabolic deep notes. I love his dramatic gestures. But even then, I do not think I would go and listen to the beautiful Yamaha C7 Concert Grand play itself in the style of Horowitz: there is something missing. Because I came to listen to Hilary Hahn. I was listening to her and it was ecstatic. The fact that I had my eye closed did not matter.
There was the Turing trap. If it was really about being indistinguishable with human beings, Yamaha probably outdoes numerous young aspiring pianists already. Their machine-learning powered C7 can play pieces written by Sorabji that no humans would be able to play.
It is about the thousands of hours the pianist spent to master the delicate leggiero, the interpretations of Liszt's Dante that people only achieve by reading the Divine Comedy numerous times. Don't get me wrong, Beethoven X for me was interesting as a concept. I admire the people passionate about widening the frontier of technology. I think MusicLM is a fascinating experiment.
But ultimately, it is not about the Turing test. It is not that AIs are not good because they are yet insufficient to sound and feel like humans. It is because they are aiming to replace what is uniquely not theirs.
On my way back home, I watched the Oscars. I cried when the presenter choked up while announcing that Ke Huy Quan as the best supporting actor, a first Asian to win the category. While Everything Everywhere All At Once is a fantastic movie with or without the Oscar, his story genuinely moved me.
AI is not about being indistinguishable from humans. It is about aiding humans to do what they do better. It needs to be used with intentionality, and the beauty will always lie with the human effort and passion behind.