Google’s Douglas Eck opened up about the company’s Magenta AI in a recent interview with the MIT Technology Review. Eck is a research scientist on the Google Brain team, responsible for AI music creator Magenta. Magenta is being taught to create music, draw pictures and eventually tell jokes.
One of the biggest challenges Google faces is getting an AI to create coherent music. Magenta’s work passes for abstract improvisation but tends to omit any form of structure. In its current form, Magenta doesn’t have the capability to create an overarching structure for a piece.
Eck said the team is working on it though, exploring options like the use of dedicated sub-models to assist Magenta in the task. A separate model could generate chord changes or rhythmic effects, providing a baseline for the main AI to layer with a melody.
Eck also addressed criticism of Magenta’s music making. Because the model is trained using existing recordings, critics have suggested it’s not really producing an original work. Since current AI is reliant on training before it can attempt a task, computer-generated art could end up trapped in derivations of its source material. Eck dismissed the pessimism, noting that human works are also in some way derivative. He said the AI’s originality should grow over time as it comes to understand the world and how it works.
“I think it’s unlikely to me that a machine-learning algorithm is going to come along and generate some transformative new way of doing art. I think a person working with this technology might be able to do that. And I think we’re just so, so, so far from this AI having a sense of what the world is really like. Like it’s just so, so far away,” Eck said during the interview. “At the same time, I think that a lot of art is original in another sense. Like, I do one more cool EDM song with the drop at the right place, that’s fun to dance to and is new, but maybe is not, like, creating a completely new genre.”
READ NEXT: Will AI really prompt World War III?
Google’s next steps with Magenta involve tasking it with expressing a different form of creativity. The researchers are attempting to train the AI to craft words into jokes humans can appreciate.
The “very preliminary, very exploratory” project is aiming to discover the element that makes a perfectly timed joke so impactful. This in itself is a challenge but the AI then needs to appreciate the nuances that can make a joke appear flat, uninspired or even insulting.
Surprise is the critical component of the puns Google is working with. However, surprise is a difficult quality to represent. Besides the possibility of the surprise being revealed too early, it also carries the potential to create strongly negative emotions. The way it is perceived depends on its application.
Eck explained that “it’s surprising if a building collapses on your head” but this is a tangibly different surprise to that felt during a joke. To an AI, the differentiation is much less apparent though. Google’s trying to train Magenta to understand the semantics of a sentence, allowing it to then manipulate elements for comic effect. Magenta’s yet to tell any genuinely amusing jokes but Google’s persevering in its efforts to create the first comic chatbot.
