Communication between disciplines

Building AI models currently needs a good level of programming knowledge, and mathematical understanding of how aspects of the model work. But building a system for use in a specialist domain, plant phenotyping in our case, also requires understanding of that specialism – how and where a plant disease presents itself; the scenarios in which yield can be affected during growth; the phenotypes of pests at different stages of growth, etc. Being able to explain important aspects of a domain to a computational person, and likewise, being able to explain model choices and limits to a biologist, is an important hurdle to overcome.

What can be done to help with this?

• Involve the computer scientist early in the planning process. Let them see example images as soon as possible, to give them an idea of the scale of the challenge.
• Show the computational team how things are done manually. If you score plants, show them how you do this. If you collect images yourself, talk to them about the image capture process. Ideally, involve the people building the models before image capture even begins, so they can advise on any problems with that part of the process as early as possible.
• It is not the aim or expectation to make all biologists computer programmers! But, having a basic understanding of what coding or scripting is, and what example datasets in the area are comprised of (formats, metadata, etc.) could be helpful when having discussions with the computational team.