Within a few short years, AI technologies have permeated nearly every aspect of our lives, and are re-defining many industries. Accordingly, AI has made its way into horticultural research and production, with the most common application being the rapid processing of imagery data to extract plant traits or other relevant information. There are many data-related, technical, and ethical challenges that must be addressed in order to facilitate widespread adoption of AI technologies in horticulture. The success of AI models hinges on the availability of high-quality and high-volume data with necessary contextual information that the model can use to learn. While general large language models (LLMs) have successfully leveraged the internet as a comprehensive data source, horticultural applications often lack the required data. This talk will introduce techniques our group has been developing to overcome the problem of data availability for AI model training by linking 3D biophysical models with AI models via synthetic data, whereby simulated data can be used to supplement available datasets or enable AI model prediction of traits and processes that cannot be readily measured with high throughput.
These ideas will be presented as one of many potential approaches for harnessing the power of AI tools in horticultural applications.