Develop a human-readable formal language for complex scientific models
Much modern research is based on complex models for complex systems. Examples are biological macromolecules, climate models, or systems biology. These models are too complex for publishing in journal articles. They currently exist only in the heads of a few experts and embedded in simulation software, inaccessible for inspection by peers. Most users of simulation software have no precise knowledge of the models they are applying. See here for an in-depth discussion.
Such models should be encoded in formal languages designed for this purpose. In contrast to programming languages, these languages are designed for simplicity and readability by humans. Formalized complex models can then be used for testing simulation software, and read as input by future simulation software. They can also be worked on by software tools similar to today's computer algebra systems, and published as datasets.
The idea is similar to the use of specification languages for software. A specification for scientific software would consist of several models plus a task to be performed based on these models. A formal language for scientific models would be a hybrid of a software specification language and a notation like OpenMath.