The Google Brain team recently released a stunning new creation by the machine intelligence that it has been working to develop. While the tune isn’t performed with any soul, the melody in itself isn’t horrible, especially when paired with a catchy little back beat that the computer added.
The composition is part of what Google Brain has titled “Project Magenta,” an initiative to expand the capabilities of machine intelligence in terms of creating art and music. The Magenta team is currently developing algorithms that enable artificial intelligence systems that will attempt to teach AI systems how to create compelling art and music on their own. Magenta also seeks to create a new community where artists, coders and machine learning researchers can share ideas and work towards a common experiment.
According to company spokesperson Jason Friedenfelds, Google software engineer Elliot Waite was responsible for creating Magenta’s first tune. To construct it, Waite used an LSTM, or long short-term memory neural network trained to use some newly developed techniques in attention. LSTM networks are especially helpful in circumstances when AI systems need to learn from experience how to classify, process and predict time series in situations where there are long lags of unknown duration between events.
“The important parts there are memory and attention,” explained Friedenfelds. “The neural net has to be able to look over a longer range, and to get a sense of what’s important to focus on, to either repeat it or change it. That’s why it seems to have some structure and some repeating elements.”
According to Friedenfelds, the Magenta tune “was completely self-learned using just a large collection of MIDI pop tunes.” It actually turns out that the team “Added some drums just to hold it together,” but Friedenfels clarified that “the melody is machine-generated.”
“We didn’t give it any rules about music, or any little rules of thumb to help it generate anything nice-sounding, as most previous machine-generated music has done.”
As to what the purpose of all this is, Jim McGregor, founder and principal analyst at Tirias Research, explains:
“If you have the processing power to analyze color and note patterns, you’ll come up with stuff that’s unique and will be of interest to a wide range of people… Then you can take the art or music produced and have the system learn- from hits on the Web or comments by the people- or see what appeals to the most people. It’s beats to music or color patterns that catch the user’s eye.”
I’m still in the dark as to the purpose of having machines create art, except that perhaps their creations might inspire other human creations. If that’s possible, it’s exciting, but the right role of machines in our world is worth considering when we decide what kinds of software to spend millions of dollars producing. Does human society need machines to help us complete all the art we need to make in the future? Not really, we’re supposed to make art for fun. But perhaps this is tinkering with what art is in a way that is art in our digital world.