Creativity in AI

The idea of a creative AI can be inherently terrifying. I recently sat down with a singer/songwriter to discuss just building a dataset to train a machine. Question one: ‘So you’re trying to replace musicians?’ Not quite. 

It’s hard to imagine that, in a world where machines understand and can generate music from scratch, people as musicians will still be required. As a writer, the idea of an artificial intelligence being able to generate articles is intimidating. That said, I input ‘Creativity in AI’ as the headline with an aim for ~500 word articles and this came out.

Not that scary. Though it does raise some good points, solely by virtue of scraping the Internet for answers. ‘Algorithms are great to extrapolate from past information, but they are still lagging behind human creativity in terms of radical, interesting jumps,’ it argues, quoting this Aeon article.

But it’s not the only writing AI. Famously, short film Sunspring was written by an AI after it was fed a series of guidelines. If you’ve watched it, it’s sort of nonsensical. So the question then becomes – who did more work to generate that script into consumable entertainment? The producers, who gathered everything required to make the film? The director, who turned a largely nonsense script into something that actually can hold attention? The actors, who took nonsense words and delivered them like they meant it?

For the curious, a sequel, It’s No Game – with far fewer hits – was released early last year. I’ll leave you to form your own opinions about it.

Another filmmaking example is IBM Watson’s trailer for sci-fi thriller Morgan. Its methodology seems relatively standard and presents what I would argue is a time-saving exercise instead of an actually engrossing trailer. The most interesting piece of that breakdown video is likely the salient sentiment graph of the film’s structure (~2.14). (Hint: nothing happens in the third act, which is basically in line with the film’s reviews).

In music, Jukedeck and Amper are AI services designed to produce songs for the budget- and time-poor producer. Their outputs are serviceable but not amazing. To be honest, I’m considering using Amper for an upcoming project to help keep scope down but I’m not planning to replace my collaborations with musicians. They understand the feel of a scene better than a machine could, even with an understanding of genre conventions. As a co-worker put it to me: ‘The AI has no concept of life.’

Machines can play by the rules but, so far, they can’t really push boundaries. Any breakthroughs in this space are made by humans extrapolating from their assistant AIs making associations we might not have. I’d suggest this is where the future of creative AI lies. At Eyeo 2017, writer Robin Sloan outlines his approach to using AI in his work with largely this framework in mind. Human work has to be edited, he says, and so must the work of computers.

I understand the trepidation around the effectiveness of AI that can be creative. I don’t want to be replaced either.

But instead of aiming for cynicism, instead of trying to dominate creatives into helping machines be better, we’re trying to use machines to help humans to be better. To waste less time being stuck. To build, as per IBM, a smart, efficient and inspirational assistant.