What’s All the Fuss About? Shimon
The 21st century’s answer to Mozart could be mechanical
What is it?
A robot that has taught itself how to write musical scores. After looking for common patterns in over 5,000 pieces of music, across all genres, it can now compose and play its own music on a marimba – a percussion instrument similar to a xylophone. Previously, it could perform with humans in real time, jamming and improvising, but not generate its own compositions.
Is this another example of machine learning?
Yes. Shimon’s creator Mason Bretan, from the Georgia Institute of Technology, created the algorithms that let it analyse music, then gave it the first four bars of a new composition and waited to see what would happen. Shimon didn’t disappoint, producing two pieces of music each around 30 seconds long.
Where can I hear them?
On Youtube. These videos ( www.snipca. com/24733 and www.snipca.com/24734) show the robot, with four arms and eight sticks, plinking and plonking on the marimba (as seen from above in our main image).age) It can see what bars to hit through a camera on its head. The tunes, a cross between jazz and classical, are best described as ‘interesting’ - a bit more avant-garde than what Patrick Moore might have played on his xylophone. Your postman would struggle to whistle them. But then the aim is not to top the charts. It’s to show that a robot can think like a human being when writing music by considering how chords and harmonies – not just individual notes – fit into the overall structure of the composition.
Could it play more conventional music?
Yes, it could be programmed to use traditional verse-chorus structures. But that would require more human involvement, which defeats the object of machine learning. It will always produce a different tune as long as it is given different opening bars.
Is it the first robot to write music?
No. Last year an artificially intelligent program called Flow Machines wrote the song Daddy’s Car based on Beatles tracks (hear it at www.flow-machines.com). Google has also been busy with its Project Magenta, which is aiming to make “compelling art and music” using machine learning. Its first attempt was this basic (and not particularly compelling) 90-second piano melody: www.snipca.com/24736. But nothing has been more prolific than Bot Dylan.
What’s that?
Named after the music legend – give or take a letter – it’s a program that has written 100,000 tunes after scanning 23,000 pieces of Irish folk music. It was built by researchers at Kingston University, who chose folk music because it has a well-defined structure and plenty of material. Dr Oded Ben-tal, a senior lecturer in music technology at the university, said he was “really surprised” by the quality of Bot Dylan’s music. Judge for yourself at www.snipca.com/24739.
Will these robots replace musicians?
Not according to Dr Ben-tal. He said that Bot Dylan is meant to aid musicians, not take their jobs. Beginners could work with it to compose their first tunes, he added, while more experienced musicians could use it to come up with new ideas. However, all scientists involved in these projects agree that it’s wrong to think of creativity, musical or otherwise, as being exclusive to humans. Might we soon see a West End musical written by a computer? Andrew Lloyd Webber, you have been warned.