Article Preview
TopIntroduction
Art and technology have been interweaved in the course of human history. In the Renaissance, the advancement in anatomy led to realistic and perspective depiction of human body. During industrial revolution, the development of photography steered art history away from realism to impressionism and expressionism to capture the beauty of changing nature and inner expression. Nowadays, artificial intelligence (AI) demonstrates stronger potential for art creation. Many researches have been conducted to involve AI into poem generation (Zhang & Lapata, 2014; Cheng et al., 2018), creation of classical or pop music (Thakkar et al., 2018; Hadjeres et al., 2017) and automatic images generation (van den Oord et al., 2016; Yan et al., 2016; Xu et al., 2018).
Whereas there are fewer researches exploring the possibility of artificial imagination. In this paper, we want to shed light on if AI has imagination, and we tackle a more challenging task: the generation of Mind Map with a given topic or an idea. Firmly implanted in a scientific epistemology, a map is viewed as a representation of knowledge. Whereas a Mind Map is an artistic representation to visualize the abstract information in a figurative way (shown in Figure 1). Sharing the same form with conventional maps, Mind Map is a special diagram which connects and arranges related words, terms and ideas around a central keyword or idea (Hopper, 2012). Thus, the core issue for Mind Map creation is the information expansion, that is, given a keyword or idea, how to extend it with more ideas, concepts or ontologies. In that case, during this research, instead of discussing the drawing techniques for the visual representation of Mind Map, we focus on the AI-enabled imagination for concepts, ideas and all forms of information expansion in Mind Map.
Currently, research on word expansion, which mainly are based on word embedding techniques (Pennington et al., 2014), are a knowledge aggregation rather than imagination. To generate an imaginative and creative Mind Map, we are facing the following challenges. First of all, language is an informative and yet complex system for human mind expression. Words and their relations should be extended with a myriad level of information, such as the lexical, phonologically, psychological features. How to expand imagination beyond the semantic boundary is the first concern in this research. Secondly, same with other artwork, Mind Map needs to reflect individual artist’s mind, understanding and experience of the world. The concept of imitation also laid the historical foundations in both Chinese and Western art (Xie, n.d.; Coleridge, 1984). The second challenge is how to teach AI to learn from artist’s mind, knowledge, and experience during the Mind Map creation. Last but not least, unlike cartographers, artists pay less attention on neither following the convention nor defining accurate relation between concepts. Instead, they plant the seed of artistic free play of imaginations when defining connections between objects or concepts. The imaginative connections between literal representations and the metaphorical interpretations create possibility to semantic expansion which flourished artistically. Our third challenge is how we can transcend from simply following the convention rules to breaking the restrictions of domain, categorical information and regular convention.
To solve above challenges, we propose an AI injected Mind Map generator, Mappa Mundi. Specifically, to better imitate artist’s idea and knowledge in AI creation, we propose to establish a knowledge graph by extracting author’s ideas and thinking. Further, we break domain and restrictions for word expansion and try to explore more possibilities for words connection by creating rules influenced by the Dadaism.
The contribution of this research is in three-fold: