老司机传媒

AI was one of the hottest topics last year, as big companies started to invest in efficient Artificial Intelligence programs while others were creating programs to detect whether AI was used on something or not (ahem… TurnItIn?). Anyways, AI is still being developed nowadays, and despite prohibiting it or not, one likely would have already asked: can AI replace a human?

While AI can definitely replace humans in some tasks, mainly repetitive or mechanical tasks that do not require as much supervision, it has not developed enough to do complex tasks–like writing a good Student Movement opinion piece–or to emulate emotions. Concerning art, Artificial Intelligence has not developed the capacity to create original work at the level of human work, or at least not without taking existent art products and remixing them. Examples of this are the struggle of AI with drawing hands, the use of banks of images to train AIs without the permission of the original creators, and other details that can pretty much reveal that AI did something. However, as it is rapidly expanding and developing, some have used AI as a bank of inspiration or as a base to start with, and especially some writers have found it quite helpful, such as Akutagawa Prize’s winner Rie Kudan, who admitted that AI wrote 5% of her work, which coincidentally talks about AIs. Besides all of that, is AI’s work really comparable to a human artist’s work? 

For this article, I tried to find out whether AI could make something comparable to a human’s work in music by trying to make an instrumental track. I used two AI programs found on the internet for music-making:  and , one that only creates instrumental music (claiming to be royalty-free), and the second one that tries to make music with lyrics and even a singer but with minimal customization options. After some experimentation, you can come to your own conclusions on whether AI is definitely winning territory to produce music or whether humans’ touch on music production is still unbeatable. 

I am someone who cannot live without listening to music (almost) all the time, and one of the genres that give me the energy to work and get things done is K-Pop. This genre has diversified and westernized more with time, but the primary roots are based on hip hop, R&B, pop, jazz, electronic music, and some Afrobeats, all with some Korean touch and lyrics. With this in mind, I took myself to the SOUNDRAW website, signed up, and started to make some choices on which the AI would make the music track, with the goal of creating something similar to K-Pop’s instrumentals.

SOUNDRAW will give you a chance to choose a duration, tempo (the speed of the song), genre, mood (the vibes of the song, to say it in simple words), and theme (a more specific topic of your song, such as AD/jingle music, sports, etc). All of these sections have predefined options, so you cannot write a concrete idea outside of what the AI is programmed for. For the first track, the selected setting was 3 minutes, between normal and fast tempo, hip-hop, with a heavy & ponderous mood, and with a motivational and inspiring theme. The idea in mind was a track similar to songs by the K-pop groups Stray Kids, Ateez, Itzy, G-Idle, and others. 

The program itself would provide several options and songs based on your choices, and it also gives you a chance to edit them a little bit since the instruments in each song were samples based on intensity (represented by the color brightness) and categorized in 5 sections: melody, backing, bass, drums, and fillings. It also gives the option to change BPM (Beats Per Minute), the instruments used, the key (the tone/scale of the song), and the volume of each track. 

For this first try, the structure of the song that I had in mind was Intro - Verse - Chorus - Bridge - Verse 2 - Pre-chorus - Chorus - Bridge 2 - Chorus - Outro (quite a classic structure of a K-pop song). After some time deducing how to use the program and creating the song structure, here is the first result:

The song probably does not have the best transitions between each section, but it could work as a draft for a song, probably for a summer commercial or even for a full reggaeton song (in any language). Personally, it gives me some vibes.

For the second song, since I wanted to make a more specific and complex structure, the setting was 4 minutes, a fast tempo, with genres between hip-hop, electro, and D&B, a heavy and ponderous mood, and a dramatic theme. The structure was Intro - Pre-Verse - Verse 1 - Pre-chorus - Chorus - Post-Chorus - Verse 2 - Pre-chorus - Chorus - Bridge 2 - Long Chorus - Post-Chorus - Outro. The result was:

Although I am not well experienced in music production, from my musician’s perspective, this track, with some extra touches of effects and changes in some of the instruments plus vocals, could work as a b-side for a K-pop album or for any other hip-hop artist. This is probably the best result I have gotten from this AI so far, and I have to say that I was not disappointed. 

Finally, I went to the other program, songR, to give it a chance and see what this AI could offer. I selected the hip-hop genre and gave it a prompt: a K-pop song about fighting for your dreams. It later gave me the lyrics and the option to choose one of the three singers they have, and I selected one of the female ones. The results? Let the song talk for itself…

While the backtrack is generic, the voice was not in the genre's style, and the words' pronunciation can barely be understood. Besides being a very short song, it had a vague structure. In conclusion, this song is not at the level of a published one and does not have potential. 

Although this experiment was not big enough to give a general statement for all music AIs, it is clear that some programs are becoming more sophisticated in creating something from zero. In general, AIs do not represent (yet) a risk that musicians have to face, and the fact that music needs the human touch to be pure art is restated with this small experiment. We will see what will come in the future, depending on where the trends lead to and how music tastes evolve (or, some would say, “decline”) through time.