AI and the democratization of knowledge

Abstract

The solution of the longstanding “protein folding problem” in 2021 showcased the transformative capabilities of AI in advancing the biomedical sciences. AI was characterized as successfully learning from protein structure data, which then spurred a more general call for AI-ready datasets to drive forward medical research. Here, we argue that it is the broad availability of knowledge, not just data, that is required to fuel further advances in AI in the scientific domain. This represents a quantum leap in a trend toward knowledge democratization that had already been developing in the biomedical sciences: knowledge is no longer primarily applied by specialists in a sub-field of biomedicine, but rather multidisciplinary teams, diverse biomedical research programs, and now machine learning. The development and application of explicit knowledge representations underpinning democratization is becoming a core scientific activity, and more investment in this activity is required if we are to achieve the promise of AI.

Fig. 1

The need for democratization of data and knowledge. For data to become broadly valuable, it must be transformed into a form that can be correctly interpreted and used by a broad community comprising both non-experts and AI.

Fig. 2

The data-information-knowledge hierarchy in empirical sciences and the path to democratization. Increasing democratization requires additional effort to transform toward consistent, explicit knowledge models, which for complex models requires extensive curation of training sets sufficient for AI.