Theme 4: Harnessing Exponential Knowledge Growth

Scientific knowledge has grown exponentially over the last 200 years, doubling every nine years, and 1.9 million articles were published in 2012. Even single topics can be daunting: there were about 66,500 articles on graphene since 2005. Assuming 2 hours to understand each article, it would take more than 40 years, working 8 hours a day, to read the current graphene literature. This raises the question of how our Materials Design Engine should interact with the global knowledge base.

We will tackle this by embracing recent developments in cognitive computing and AI. Our 10-year objective is for Centre researchers to work with an AI ‘research advisor’ that acts as a semantic reasoning interface with the knowledge base. This is ambitious, controversial, and unprecedented in materials research. Convincing examples of ‘AI-assisted research’ are scarce: the IBM Watson Discovery Advisor used semantic reasoning in the areas of oncology and protein research (Nagarajan et al., Proc. 21th ACM SIGKDD Int. Conf. Knowledge Discovery and Data Mining, pp. 2019-2028), but these are focused and highly codified subjects compared to materials research.

We will lead the way by developing an AI research advisor for crystalline solids, for which there is already a database, the Cambridge Structural Database, curated by our CCDC partners. We will expand this knowledge base with both computed and experimental crystal structures from this Centre (Themes 1–3). By coupling computational developments in Theme 1 with rapid developments in natural language processing, our disruptive target is an AI research advisor that can answer real language research questions, and which can be used by ‘Materials Engineers’ who are not experts in computer science. This AI advisor will unlock creativity for researchers in the Centre – for instance, by ‘seeing’ buried functional relationships between superficially unrelated materials in the knowledge base, which then inspires an entirely new synthetic design approach.