If you're worried that machine learning will lead us all into a future of robot overlords, take comfort in the fact that it now poses a different (and far less dangerous) set of possibilities.
Magenta, a Google Brain project, is designed to test whether machines can create compelling art. When it opened last year, it allowed people to explore music creation through a neural network. This project led to the creation of AI Duet, a programme that lets people play a piano duet with a computer. You don't need to know how to play the piano - just type in some notes on the keyboard and let the computer respond.
The result is a promising, if somewhat stilted, attempt at a long performance piece - the AI responds reasonably well, but perhaps without the flair that you might be expecting. The performance of the AI does match the notation and tone of your playing, so that it sounds remarkably like you. However, it's responses are only in short intervals and, if you don't stop playing after a certain period of time, it begins to play over you, creating a complicated jumble of notes.
Despite these setbacks, the technology still shows promise for how artists could use AI to enhance their creativity.
The Sony CSL Research Laboratory released pop music composed entirely by an AI. It's surprisingly - and somewhat maddeningly - really catchy. Fans of the Beatles or Electric Light Orchestra will appreciate the harmonies and bright tones of Daddy's Home. It's the kind of song you listen to twice and hum for the rest of the day.
The AI system, FlowMachines, analyses a database of songs and then follows a particular musical style to create similar music. The French composer Benoît Carré arranged the songs and writes the lyrics used in the AI's composition process.
Artificial intelligence still relies on human input to mimmic music. However, as the tech advances it becomes more likely that the most popular songs in coming years could be productions of machine learning - or at least be structured with a healthy dose of AI analytics.
This article was originally published by WIRED UK
