SORRY - LONG TEXT ALERT
As a person who worked for years generating data to AI models, I will try to write few words about it.
Every AI need to be fed with data.
Almost every concept of AI exists since the 70's. However, the computational power was far from the minimum requirements and a effective neural-network predictive model started to be really funtional at late 90's / early 2000's.
Music (and arts in general) always had been on the debate when the subject is AI because of the multiple "unpredicted" behavioring (aka human factor).
What happens is that, Music have an importart role into the Generative AI because the machine can learn in every little detail based on this "unpredicted" behavior and apply (escalate) it to other areas such as medicine, text generation in the style of someone, or who knows, discover who is playing in a ramdom song and check if the credits are corrected. (connex rights)
Now let's play a game.
Lets use our beloved MK as an example.
Every MK song is categorized by many things, per example:
- General aspect of the lyrycs
- Storytelling
- Background (bibliographical)
- Song similarity (comparison)
- Harmony
- "Plagiarism" or Influences (other artists similar songs)
- Voice tone
- Number of words (small words, dialects, big words, slangs...)
- Vocal performance nuance (volume, space, tone, screams...)
- Guitar tone (phrasing, music theory, style, performance, nuance...)
- Groove
- Band members
Every item have a new tree of nodes. Every item is repeated by each band member. Every guitar tone\lick\phrase is compared with similar existing nodes... and so forth.
And each item changed based on wich MK era we are selecting.
We can use(prompt) only the MK from the first 2 DS albums per example.
Let's do an exercise and imagine how we can do a AI version of Eagles's Hotel California sung by Dire Straits in 1979.
If we could back in time and be with the band, as a musical director, on a studio, to work that song. What we gonna do?
Remember that we can travel in time knowing everything until 2025.
Very basic mental exercise:
- We know the groove
- We know the guitar tone
- We know how the lyrycs should work
Now try to categorize every aspect of wich are the DS songs that influenced the singing, licks, grooves... and put it into organized data (would be a LOT of data) and then know how to "talk" with the machine and refine it (More data). You could deliver a relative decent job.
You need data and how to deal with data to have descent AI results if you wanna do complex experiments. And that is the word. Experiment.
AI is a tool and people are discovering the limits of it.
Of course Chris comments are valid and absolutely necessary on how the consuming world runs. However AI unlocks another side of creativity (for good or for worse) that will be possible for everyone.
Classical Music was over because the composers tryed almost everything in terms of harmony and beauty that time. After that came Stravinsky, Messian, Mingus and modern harmony concepts.
It happened with Bepop, Jazz, Rock, Classical...
Speaking about guitar, who would imagine, 20 years ago, a guitar player like Tosin Abasi? Style, Phrasing, Tone, Technology, Concept...
A person who goes for a Jim Campilongo or a Nels Cline concert wants a real experience.
I believe that, in some years, we will live a balance about it.
Until then, many things could appear and the real artist always will find a way to emerge and be noted.
