[Paper] Is AI Learning to Understand Emotions through Visual Art?

analytics-insightThis post was originally published by Meenu EG at Analytics Insight

AI Learning

Emotional AI is not far away considering the nascent developments in the field.

Artificial intelligence has already made its mark in our lives. The adoption of disruptive technologies redefined industries and their operations. However, the fear looming over AI, rooted in it being capable of taking over the human race has existed since the start. And most of us would have been influenced by those Sci-fi movies and books, which portray AI as an evil entity, talking and behaving like humans.

Well, studies show that we are yet to reach the point where AI can fully augment human intelligence and emotions. Emotional AI is a developing field of study and researchers are trying to incorporate emotional intelligence into AI algorithms so that they can augment human behavior better.

There have been instances where AI created visual art. For example, back in 2019, a gallery in Chelsea organized an exhibition of computer prints that were created by an AI named AICAN. But what about interpreting art? AI in interpreting art is a developing area since it has certain complexities.

To interpret art, AI needs to understand the type and the meaning of a particular visual art piece. Researchers from Zhejiang University of Technology, Hangzhou, China recently published an article on art classification. They tested 7 different models on 3 datasets to compare their art classification performances. The study was to understand the ability of these neural network models in identifying styles, artists, and genres in particular artwork. According to the article, the convolutional neural network models and computer vision techniques used delivered state-of-the-art results, especially in smaller datasets.

Computer vision has been a revolutionary innovation that currently impacts many industries including automotive and construction. AI technologies have developed to an extent where it almost augments the human brain, though not fully replicate it. A group of researchers from Stanford University, Ecole Polytechnique, and King Abdullah University of Science and Technology, published a study titled ‘ArtEmis: Affective Language for Visual Arts.’

This team has trained an AI algorithm to interpret emotions in great pieces of art. They have used the WikiArt dataset consisting of 81,446 artworks from 1,119 artists and collected more than 4 Lakhs of emotional explanations and responses from annotators from Amazon Mechanical Turk (AMT) services.

This experiment aims to bring AI and emotional intelligence closer by enhancing the capabilities of machine learning algorithms to analyze data based on emotions, metaphors, descriptions, etc. The algorithm they developed could dissect the artworks based on eight emotional categories. As an example, we can consider Vincent Van Gogh’s Starry Night and its explanation given by ArtEmis. According to the study, ArtEmis identifies emotion in the painting as ‘awe’ and explains it as “The blue and white colours of these paintings make me feel like I am looking at a dream”. ArtEmis is a great development since it can also provide written explanations of visual stimuli ed into the algorithms apart from just labelling it with emotion.

These nascent developments in the field of AI will enhance its capabilities. Adding emotional intelligence in AI will benefit many business operations, especially customer interaction and engagement. But there will always be ethical concerns around these human replicating abilities of artificial intelligence. Will we be able to make it through these controversies and concerns? Will ethical AI and emotional AI replace humans completely? These are some questions that we need to seek answers for in the near future.

The ability to deal with the images’ emotional attributes opens an exciting new direction in human-computer communication and interaction.

ArtEmis: Affective Language for Visual Art (pdf.)

Youtube: View 80,000 paintings in 8 minutes!

Github | Dataset

Spread the word

This post was originally published by Meenu EG at Analytics Insight

Related posts