If you consider Indian classical music I  am sure till now no AI at present can be able to creat  such

I guess when you can hardly tell difference between the artificial one and real one that could be one parameter.

Best regards,

Debra

There are ways to measure complexity of sets of data, which could be used for music. But I haven't heard about any formal definition of creativity that could lead to an algorithmic way to evaluate it, regardless of its human or computer origin

Regards,

Laurent

Focus primarily on the cognitive dimension;

• Domain expertise, which is required for mapping the conceptual space that is to be explored and/or transformed.
• Valuation of the results, which is especially necessary and especially difficult for transformational human creativity.

I´m not sure you could compare "creativity" as such because nobody really knows much about it unless it is being creative. I believe creativity is a state and states can only "be" (experienced). I´m sure there have been many attempts at describing how the creation of novel material, ideas, associations or solutions to problems old and new must be in order to be considered "creative" but still, I thing this is quite a slippery terrain as it very much depends on the expectations, beliefs and previous experiences of whoever is attempting to measure or analyse how "creative" something is.

In any case, if the goal is to create music "automatically" for certain applications, you could perhaps start by making algorithms to represent music that is "generally" considered "good" or "appealing" - not an easy one but perhaps it could be made context-specific such as in composing for an "upbeat" and "joyful" scene where all tags such as joyful, upbeat, sad, cinematic...etc would have their own algorithms attached, made to equal music that was considered a "good fit" by many.

You could then have AI programmed to replicate the most successful algorithms per tag (within a random kind of scenario so as to arrive at novel compositions) or even make it produce music and when the outcome is close enough to an existing algorithm, the program would "know" and select it/archive it...

Tough one !

With  reference to your comment, Laurent, the definition of creativity I use in my book is: "some action or thought that that person has not done or thought before".  I guess in the context of AI, if the software can generate different  tunes from the same set of initial parameters then, to some extent, that might be considered creative.

If a computer program were completely free to compose

it would probably choose a spot far out in the

musical universe. For example, the program would

choose a 97/5 time on a 32 tone scale in ultrasound

at 0.005 beats per minute.

In order to compose music that is pleasant to humans,

a couple of rules should be obeyed that originate

in physiology.

- Tones must lie in the audible spectrum.

- The times denominator should be a power

of 2, otherwise the tones can be very short

- Number of beats per minute must be chosen so

that a human can play an instrument, dance,

or clap with that speed.

- Working memory is limited to about 7+-2 items,

so impressions intended to appear as a whole

should not exceed that number in a group

Aside from such restrictions by nature there

are constraints that a musician imposes on

himself, especially the desire to use certain

musical motifs. Those motifs should not be

distributed arbitrarily. Damien Zanette found

a relationship between linguistics and music

based on Zipf's distribution.

http://www.nature.com/news/1998/040614/full/news040614-11.html

https://arxiv.org/abs/cs/0406015

Regards,

Joachim

Thanks for different perspectives and views to enrich knowledge in the domain.  However, evaluation of creativity or what is good or bad music ?difficult to analyse.

Sometime back i have expressed my views regarding the music which gets also reflected in the present question of music as an AI  to be compared with human creativity .

With this music cannot be placed  down in AI .Music is an art & arts goes with the mind ,brain & intelligence .

It has been observed that students learning for mathematics ,accountancy & such other subject while reading for their study ,they take recourse to music,& listening to their interesting songs.

The same is equally true for fine arts student creating a work of painting & in this line we can not afford to over look the performance of classical dancers who can not create their performance without music .

This is my personal opinion 

I think there is no such thing for evaluation or comparison. There exist a test named "Turing Test" if a human does not pass the test, I assume there will be none, On the other hand, how many tunes do we have? If I recall correctly eight, and how many are we using? Three, I guess. If the music created by an AI contains more than that it is very likely to be coming from an AI with a high probability. Considering that I am an amateur, I dare to recommend to read Godel, Bach and Escher written by Douglas R. Hofstadter.

It is my humble opinion.

Here is some work evaluating code-based music making for CS-based STEAM education: https://pdfs.semanticscholar.org/5b8e/da7fc626e6a9b11e54fb76916d5d89f2b2c8.pdf

Maher has also done a considerable amount of work investigating computational techniques for evaluating human and computational creative products: 

https://www.designsociety.org/publication/32461/using_ai_to_evaluate_creative_designs

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.717.9119&rep=rep1&type=pdf

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.643.2457&rep=rep1&type=pdf

-Nick