|Follow on Twitter as @Ayes_Sal|
New developments in Artificial Intelligence (AI) are now challenging the age old assumption that computers cannot be creative. Computers are composing symphonies and IBM’s famous AI machine Watson, has recently created a cookbook. In the art world, now a machine algorithm can make connections between paintings and ‘study’ art. We catch up with Ahmed Elgammal at Rutgers University in New Jersey over email to discuss his work with partner Babak Saleh on this machine algorithm to find out more.
|Photo: IDG Connect|
How have you taught the machine to study art?
The interesting thing about our recent study on assessing creativity is that the machine did not know anything about art. The input to the algorithm are images of paintings, where we use off-the-shelf computer vision techniques to encode them. The algorithm then reasons about creativity based on similarities between paintings and their date of creation, using the definition of creativity that we used to design the algorithm. So we did not really teach the machine anything that is art-specific.
For earlier studies we did use annotations of paintings, such as style, genre, and their artists to teach the machine about art. However this is not used in this project.
What sort of tests have you done to see how the machine identifies “creativity”?
We proposed a validation methodology, which we call “time machine experiments”, where we change the date of an artwork to some point in the past or in the future, relative to its correct time of creation and recomputed their creativity scores.
We found that paintings from Impressionist, Post-Impressionist, Expressionist, and Cubism movements have significant gain in their creativity scores when moved back to around 1600 AD. In contrast, Neoclassicism paintings did not gain much when moved back to 1600. This makes sense, because Neoclassicism can be considered as revival to Renaissance. On the other hand, paintings from Renaissance and Baroque styles had losses in their creativity scores when moved forward to 1900 AD.
How does the algorithm decide between paintings that are creative and those that are not?
First, let’s go over the definition of creativity that we used. Historically there is an ongoing debate on how to define creativity. We can describe a person (e.g. artist, poet), a product (painting, poem), or the mental process as being creative. In our work we focused on the creativity of products, and we used the most common definition of creativity of products, which emphasises the originality of the product and its influential value.
The algorithm is based on constructing a network between paintings and using it to infer the originality and influence of them. Think of it as a game: each painting has the same amount of creativity tokens. Then these creativity tokens are passed between paintings based on their similarity and their dates until equilibrium is reached.
Source: IDG Connect