Max Order is a story inspired by the ERC project
Machines. Creativity is
indeed a research subject on its
own, and the ERC project Flow Machines takes a computer science perspective to look to it– by
that can boost creativity thanks to artificial intelligence.
The key idea of the Flow Machines project is to relate the notion of creativity to the notion of
is what makes an author (composer, writer, painter, etc.) recognizable, “different”.
It seems that Max is bound to copy, just like her father.
Indeed in Artificial Intelligence it is relatively easy to “copy” the style of a composer or of a
writer: but then, if the machine does not go beyond that, the result is going to be just a long
sequence of notes or words without a structure, or a copy of an already existing artwork.
How to escape plagiarism, then? How to invent something new?
Well, for example, the algorithms behind Flow Machines can suggest new creative paths on the basis
of a corpus they learned, just like Flowy inspires Max to find her own style by endlessly repeating
all the Picasso’s quotes he learned by hearth!
In this episode the music is, for once, not generated by AI but it is a reference to the perhaps
most notorious case of musical
"Copyrighted Content used on the basis of 'Fair Use"
George Harrison, My Sweet Lord, 1970
The Chiffons, He's So Fine, 1963
In this chapter we featured two extracts of the two entire pop songs composed by Artificial
The release of these two tracks had a big impact on the media so maybe you have already heard them,
but if you didn’t, here you can find the uncut versions:
And if you want to know how this is done (how much Flowy worked and how much Max created!) please
check this page on
the Flow Machines website.
On the other hand, the visuals produced by a machines are actually pictures modified via the Prisma
application, which exploits deep learning techniques to modify target images.
In the future, AI will have a major role in creativity and production of artistic contents. And
today already, thanks to research, AI can do what Flowy does with Max: for example, suggesting a
subject and a style to be used to implement it, not only in the visual arts but also in music and
For example, the West End musical Beyond the Fence
has been composed with the help of computers, that suggested the plot and co-composed the music.
In the second metalevel of narration, in this chapter you can listen to two songs co-composed by
musician Benoît Carré with the help of his own
“Flowy”: the FlowComposer,
the Flow Machines Artificial Intelligence application that can assist musician to compose in any
You can find other compositions here and follow Max and Flowy on Tumblr!
In this chapter you heard two songs composed by FlowComposer,
the Flow Machines Artificial
Intelligence application that can assist musicians to compose in any style. The first is
“flow-composed” by Flow Machines artist in residence Benoît
Carré in the style of the iconic french
composer Michel Legrand and the second one by Amaury Delort in the style of a songset of
songwriter. Every musician can use AI as a tool to compose in his/her style, and not as a
In this chapter you also had a glimpse of a “second level” where a discussion about the scenario
itself takes place. The idea here is to literally illustrate the process of metacreation, which
what happens when machines exhibiting creative behavior are able to reflect on their own process
Thanks to Artificial Intelligence, it
is possible to create content imitating
a specific style: the style of an
artist (for example, Picasso) or the
style of a specific work (for example,
In the Flow Machines project, we apply
this concept to music. For example, the
last track you heard is an automatic
composition in the style of Mile’s
Davis, rendered with guitar chunks from
But the question is, how can you
imitate a style without going too far,
that is to say falling into plain
plagiarism? The answer is… Max Order!
Papadopoulos A., Pachet F. and Roy P.,
Generating non-plagiaristic Markov
sequences with max order Sampling, in
“Creativity and Universality in
Language”, in Degli Esposti, M.,
Altmann, E. & Pachet, F. (eds),
Springer, Morphogenesis series, 2016.
All the music in this chapter is produced by a collaboration between our Flow Machines and French
and singer Benoît Carré.
Benoît has been using different machines learning systems: the FlowComposer,
artists composing leadsheets in specific styles, and and ReChord which renders harmonies and
melodies produced by machine learning with chunks of pre-recorded audio accompaniments.
You will notice that artificial intelligence has learned to play Ode to Joy in heavy metal
For all the other styles, Hear
AI play Beethoven like The Beatles on TechCrunch.
Pachet, F. and Roy, P. Imitative
Leadsheet Generation with User Constraints. 21st European Conference on Artificial
Intelligence (ECAI 2014), pages 1077-1078, Prague (Czech Republic), August 2014
Ramona, M., Cabral, G. and Pachet, F. Capturing
a Musician's Groove: Generation of Realistic Accompaniments from Single Song Recordings. 24th
International Joint Conference on Artificial Intelligence (IJCAI 2015), pages 4140-4142, Buenos
Aires (Argentina), July 2015
Flow Machines can not only recognize style, but also and above all understand it and turn into a
computational object to apply to any target – for instance, music.
In the first two chapters of Max Order, you encounter several examples of how different styles
applied to the same musical piece: Ode to Joy. You listened to Ode to Joy in Bach’s style, pop
style, Bossa nova style…
To understand how this magic works and look at the big picture, you can check this video:
The first thing to know is that machines can recognize style – have you found the first hint in
Some researchers claim that thanks to statistics and machine learning we can predict the
of artistic works. In music, this field has even a name: Hit Song Science. But, is it really a
Curious for more? Check this graphic facilitation video about Flow Machines!