How can NEON data be made AI ready?

[sydnerecord]

NEON has a treasure trove of different data types that are ripe for being ingested into a multimodal model. Are those data AI ready? If not, how can NEON serve up data to the community in an AI/ML ready format?

[JCGiron]

My question would be, ready for what?
What is the kind of information that the research community would be able to extract from images using AI?

Or perhaps I'm not aware of what "AI ready" means?

[Choco31415]

@JCGiron Some data formats are easier or harder for an AI model to work with.

For example, a LLM might not have been trained to understand Markdown code, or a time series model might not be able to read a pdf file for data.

I looked at a few of the NEON datasets and they look pretty AI ready. csv files are easy to read.

[JCGiron]

@Choco31415, got it.
I'm a beetle taxonomist and have only very general notions of what AI models do.
When I asked ready for what, I was referring to what kinds of questions in biology can be answered using AI models. My guess is that depending on the question (ecology, variation through time, taxonomic identification), different kinds of data would be required.
As a taxonomist and from the ideas posted in #8 and #10, I would be interested to see if AI models could, for example, recognize specific parts of a beetle (in an image) that are used for differentiating species, so that it can effectively be used for accurate species identifications.
When working with images, how would that information be "AI ready" to run models for species identifications or for marking up parts of an image that contains diagnostic features? besides a jpg or a tiff file, what would the metadata be, and would those simple image files need some sort of accompanying information that is recorded in a csv file?