Artificial Intelligence (AI) has already pervaded many industries. In recent years, a few AI use cases also caught the attention of the creative industry: Microsoft’s Chinese chatbot XiaoIce became capable of generating decent image-inspired poems, Sony started dishing out pop songs using its AI Lab’s FlowComposer, and 20th Century Fox asked IBM Watson to create a trailer for the horror movie “Morgan”. Now one has to wonder: Is the creative human mind still needed?
To answer this question, we need to look behind the scenes. For AIs to become “creative”, they have to be trained on relevant datasets first to understand what to read out of future inputs. In our examples, thousands of existing image-poem pairs were used for XiaoIce to learn how to find poetic clues in images; hundreds of songs of the same genre were fed into FlowComposer to make it adapt to different music styles; and Watson was forced to “watch” tons of horror movies to understand which scenes from “Morgan” may be useful for the trailer. Except for the poems, both the songs and the trailer actually also required extensive manual arrangement before release.
The results in all three categories are definitely remarkable for their technological achievement, but not as satisfying when compared to pure human creations. The poems, despite having passed the Turing test, tend to be more descriptive and bland rather than emotional or meaningful. The songs do show some typical characteristics of the respective genres, but are not very catchy due to the lack of recognizable motives. As for the trailer, it summarized the movie in an almost chaotic way, because the AI was trained to focus on salient emotions instead of the plot.
As we can see, AIs nowadays still miss a human touch when it comes to creating original content. And it is questionable if they will ever obtain real creativity since their outputs heavily depend on the datasets they were trained on. Yet, their ability to extract patterns from vast amounts of materials may help human creators see and break artistic boundaries to set new standards – something the creative industry urgently needs, and always should strive for.
Sources:
https://thenextweb.com/artificial-intelligence/2018/08/10/microsofts-ai-can-convert-images-into-chinese-poetry/
Click to access 1804.08473.pdf
https://www.scientificamerican.com/article/a-compendium-of-ai-composed-pop-songs/
https://www.ibm.com/watson/advantage-reports/future-of-artificial-intelligence/ai-creativity.html
