Whether in robotics, medicine or the stock market, artificial intelligence (AI) has already begun to transform the world, displacing numerous jobs and outperforming humans in many disciplines. However, a significant flaw of AI is that it lacks creativity, as AI itself is a human creation. Nevertheless, is it possible for AI to produce music at a level indistinguishable from humans’ capabilities?

Most AI software runs with a neural network, a system of interconnected nodes that send signals to each other. According to IBM, “artificial neural networks are comprised of node layers, containing an input layer, one or more hidden layers, and an output layer,” each with “an associated weight and threshold.” Each node resembles a neuron in the brain, as it responds to inputs and generates outputs if the value of the information reaches a threshold. However, increasing the layers of a neural network only benefits the software’s analytical abilities and doesn’t affect its ability to process inputs that are not pre-determined by humans. For example, if someone has never seen a car brand, they won’t recognize the logo. Comparably, if an AI finds a musical pattern it hasn’t seen before, it won’t know how to classify it unless it recognizes multiple recurring copies and creates a new class. Humans can write example inputs for AI to process, but AI won’t react suitably to inputs outside of its programming.

Most importantly, because AI can’t understand factors outside of its focus, it can’t adjust depending on the societal preference of specific songs or genres: it can’t keep up with the trends of modern-day music. AI is fundamentally used to find patterns: although it thrives in its primary task, AI can’t anticipate change and can only adapt to changes belatedly. According to an article published in Big Data & Society, Anton Oleinik classifies “the capacity to trace linkages between heterogeneous and previously unconnected elements as a distinctive human social activity.” On the other hand, AI would need humans to program a link between the new input and a node in the neural network; humans would have to give updated information constantly to the AI, which defeats the whole point of the ‘artificial’ in AI. 

However, AI can still be used in music, even if it cannot write new songs entirely. For example, AI’s pattern identification abilities allow it to recognize similarities between songs and develop recommendation systems, such as those currently used by Spotify. The ability to detect correlations also enables AI to complete tasks like sorting music into specific categories, such as by key, instruments, genre or even the gender of the voice! Its use in the music industry is crucial to organizing all the new songs being released constantly into easily accessible and recognizable categories.

Although AI has remarkable potential in many fields of music, it probably won’t write new, inventive music anytime soon.

Example of AI-produced music:

Franz Schubert – Symphony No.8 in B minor, D.759 (“Unfinished”) finalized by Huawei AI