The Technology


In the last few years we have seen some captivating demonstrations of the power of deep neural networks to generate artistic artefacts including imagery, music and poetry. However the domain of conceptual creativity, namely the ability to model and synthesise abstract ideas and concepts, remains a hard challenge.

Our mission at Move 37 is to create radical tools to augment human conceptual creativity. We are developing technology to power intelligent agents that will assist us with tasks involving creative thinking, exploration and problem solving.

The roadmap to realising this vision is through increasingly sophisticated language understanding. Throughout history, language has been the primary instrument of human progress and accomplishment, enabling us to exchange ideas, to capture and share  knowledge, to teach, to tell stories, to express opinions and emotions, and to cooperate in increasingly sophisticated social groups.

By teaching machines to understand language and applying this at internet scale we can start to capture the meaning represented in human discourse. Open AI's GPT-2 language model has been shown to learn common sense knowledge of concepts and their relationships through training on vast amounts of human written text. But, the current knowledge representation within a deep neural network is opaque and so it suffers from poor explainability and limited interactivity in a collaborative human+machine setting. 

Powerful language models such as GPT-2 and BERT should be seen as foundational building blocks of more sophisticated downstream applications in language and semantic understanding, rather than an end goal.

At Move 37, we are building on these foundations to develop core capabilities in creative reasoning, including the ability to:

  1. Read and comprehend natural language at scale from a diverse collection of sources such as news media, online encyclopaedias, literature and social media discourse

  2. Distill language into a rich representation of concepts and the relationships between them

  3. Discover patterns in concepts, such as analogy and causality

  4. Synthesise ideas by forming new connections between concepts and evaluating factors such as novelty, plausibility and appropriateness

Talk to us

We're always interested in discussing the above themes and related technology, and actively participate in industry and academic collaborations. In particular, if you are a researcher or developer interested in the following areas please do get in touch:

  • Information extraction from open domain and noisy text

  • Concept representation in knowledge graphs and vector space models

  • Deep learning for reasoning over knowledge with an emphasis on explainability

  • Generative models for conceptual creativity