Jimi Hendrix’ back from the dead? - Computer-Generated Content Creation
In an age where computers have been doubling their processing power yearly for the past 40 years, where Artificial Intelligence is a reality, one question arises: can computers create inspired media content?
As depicted processing power allows it. We have access to farms of computers; we can parallelize tasks to millions; we can crack the DNA… Don’t tell me we can deal with 12 plain notes? Algorithms can today also mimic the rational thought. Let’s take music for instance. It is all about frequencies and harmonies. All of these are clearly depicted by physics and mathematics.
We can explain scales, rhythms, even weight the ‘human factor’ in playing a musical instrument. But creativity is more than rationale. A catchy melody is more than just following a pentatonic scale. Yet music programs claim they can not only accompany you if you provide the chord progression of your song, but they can also create ‘solos’ with the touch of this or that famous artist… You want to jam with Jimi Hendrix? Go for it! Well they mostly mimic the sound, copy some catch sentences, but it takes more than that to genuine song writing / composition.
Media creation is an iterative process, an empirical way where we “digest” and “mix in our own way” what we’ve heard/seen in the past with our own personal ingredients to come up with something “new”… something “us”. When one says that fashion is a constant come-back, that’s the reason why.
So inspired creation requires ‘knowledge’ and ‘history’. The “partial” reminiscence of all the parts of songs you heard and that caught your ear & your heart; that your brain will collate to “create”. That’s exactly the promise of “neuronal systems”, with deep learning, where “self-learning” machines are force-fed with information and a tag/result/learning the system should ‘output’ out of these ‘inputs’. That way, a specific neuronal system will be created to repeat such behavior.
In other words, by showing a zillion picture of chairs and stools to a computer, with the correct tag attached to each picture, then the computer will be able to tell from any new unknown pictures you will show if it is a chair or a stool. The difference lies here that in regular IT, you would ‘program’ the system to reach to the fact that ‘if you see a ‘back’ to it, it is a chair, if it is just legs and seating, it’s a stool”. Here the computer has made its own “definition”, its own program to decide whether what is presented is a chair or a stool.
This is one of the secrets how Google computer could defeat the world’s best go player. But if one can explain that ‘this is a chair’ or ‘that is a winning move in go or chess game’, how can one teach a system what is a “good song” from a “bad song”? Charts ranking?
Tastes are as numerous as inhabitants, while go/chess are driven by the same rules across the planet. In music/movies, it’s what makes you vibrate, what puts a part of you in resonance that will trigger passion and love for a creation.
So we may have the computing power and the cloud services we sell to our customers for processing, we may have algorithms that learn by themselves like we use for Traffic Prediction, Failure prescription and other AI/ML topics, we can copycat existing contents/artists with the latest music programs, we still face the same mountain to climb: unless we have a “feeling” of what we like or not, unless we have a vibrating string somewhere deep inside, it is going to be very difficult to computer-generate any inspired content. The next wave of research will be around this “human” part, this “emotion”, this “feeling”, this ability for discrete entities to react differently to the same stimulus depending on their own specificities. If all computers are alike, every human being is different.
Copy we do it more or less, invent we’re far from being there yet. Check soundcloud or Avid Cloud Collaboration for Digidesign/Protools users: they clearly show the complementarity of silicon and flesh. Computer-assistance stops where actual creation starts.