Serene

concert pianist / technologist / etc.

Algorithmic Flattery

serene | april 2nd, 2017

In which I’m regularly accused of being Martha Argerich, Sviatoslav Richter, and Claudio Arrau -- by robots.

As an internet-enabled pianist, I often upload video recordings to the internets. The vast majority are unlisted or works-in-progress, as I prepare my recording setups and test things out. Yet, due to what is presumably the over-eager YouTube automatic-copyright detection algorithm, Content ID, I receive copyright notices even on the unlisted pieces, everytime.

This is of course silly, and also bizarrely flattering. Perhaps these are excellent indicators on even my practice renditions of Beethoven’s last Sonata Op. 111, and other delicious pieces by Bach, or Ravel, or Liszt… After an amount of obsessive practice, my pianism is tolerably world-class to draw the consistent false-positive ire of our robot overlords, and thus, I should rejoice.

But actually, probably not. We have much to be worried about, which I’ll articulate further below.

Given the relatively smaller size of the classical music market, its training data for YouTube’s content fingerprinting is impoverished. As a result, it’s bad at distinguishing boundaries between the same classical piece interpreted by different artists. (Yet another area which humans do a better job than machines, for now.) Of course, even this could be premature praise, as it assumes that the content fingerprinting system is a thing. It may simply be that this system itself is generally bad and inaccurate, enforcing moronic convention baked into and enforced by law. A quick look around indicates that YouTube’s Content ID system has provided headaches for quite a wide spectrum of users. In any case, here on the YouTube of 2017, if you’re a classical musician, expect false positives everytime.

The juxtaposition of ineffective legal policies and incompetent robots can result in hilarity at times. For me, these automatic copyright claims are currently of little consequence, a minor nuisance, straightforward to dispute if I care.

However, it can get much worse. And this won't be about the specific ways in which DMCA and related systems are trash fires actively harmful for human civilization regardless of the intent behind their existence. (This has certainly been covered with great depth and thoughtfulness by others, so I won't replicate their work.)

Rather, I'd like to point your attention to how we have on our hands the normalization of automation combined with misguided and harmful law. Beyond the world of classifiers on relatively obscure classical repertoire, a larger issue is the normalization of reliance on poorly trained "machine learning" algorithms, aka glorified matrix math - and then using duct tape to hook it up to the legal system. When these systems are blindly built with giant assumptions and lack of insight into how the training data actually works, then applied within self-reinforcing bubbles, we get... egregiousness. Left unchecked, this will literally ruin lives. Bad math backed by guns.

It doesn’t take much imagination to extrapolate this into a Brazil -style dystopia, as reality frays and seams widen into cracks, and people start falling through. This trend is already underway with such exciting nonsense like the great Australian Centrelink debacle, and within a few years the phenomena will likely far exceed our expectations. Our current levels of unnecessary human suffering can easily be amplified into ever broader pandemics by automated ineptitude.

As for YouTube’s Content ID, further hilarity can be observed when noting meta-squabbles such as how the automatic copyright-enforcement algorithm is itself under copyright dispute .

What a waste of time for everyone. Since I fully understand and disagree with the incentives and reasons for its development, it is painful to imagine the engineering resources and talent wasted crafting this actively harmful code, along with the amount of time wasted for anyone dealing with its inaccuracies and fallout. A vast squandering of human hours. Many, many lifetimes worth -- an indirect genocide, salved by the illusion of "protecting creators" while doing the opposite.

Despite all this nonsense, in any case, I will now go back to practicing. Life is too short to not create beautiful music. And when I’m not creating beautiful music, I will continue to study broadly, and hack on things which increase our freedom on the internet, because the alternative is lunacy.

Also, I’m glad I got some writing practice in, and another blog post from these scenarios.

Thanks, robot overlords!

return to writings

Last Update: 2022.03.21 | Thoughts my own. | © 2013-2022 Serene