As technology and artificial intelligence continue to advance, large language models, such as OpenAI’s GPT-4, have demonstrated remarkable capabilities in various applications.
I am curious to know your thoughts on the potential integration of these language models into functional-structural plant models (FSPMs).
Do you believe that large language models can contribute to enhancing the accuracy, efficiency, and predictive capabilities of FSPMs? If so, what specific applications do you foresee these models having a significant impact on, and what challenges might we face in implementing this integration?
I was thinking about this topic because the Lyndenmayer Grammar is indeed a grammar for a language that can describe plant topology, geometry, and development. I was fantasizing: could I ask a prompt to build a plant just by describing it in natural language?
I look forward to your insights and discussions on this fascinating topic.
1 Like
Hi Michele,
Would be great if you could give some references for that first statement, so our discussion could start from the same background.
First thought that comes to mind would be that it might be a way to get template models for plants, but describing details of growth rates and durations and the timing of processes controlling branching and flowering, and responses to environment, for instance, may be more difficult.
1 Like
The closest thing may be AIs for code generation (Code Whisperer or Github Copilot) but those AIs rely on being trained on large amounts of code and performance is much lower on less common languages or types of applications. By the way, such AIs can already help you with writing bits of the code in an FSP model and sometimes write an entire function for you (e.g. Github Copilot can write the code to compute solar angles from DOY, latitude and time of day).
If you would want it to generate L-systems, you would probably need to feed a lot of examples for such an AI to produce reasonable outcomes and even then, they tend to work better for “code completion”, meaning that they help with small steps within the program. We are still far from AI producing a whole program, unless it is something small and very routine like what I illustrate above.
Having said, GPT-4 is being connected with Wolfram Alpha which is a logical system (more along the lines of classic “symbolic AI”) that knows about L-systems. So maybe that is one possible route, but I would be skeptical for now and focus on “code completion” AIs to help you write the functionality part of the model.
1 Like
do you mean this statement?
As technology and artificial intelligence continue to advance, large language models, such as OpenAI’s GPT-4, have demonstrated remarkable capabilities in various applications.
thank you! I just joined the chatGPT plugin waiting list; as soon as I get access to the plugin, I will try and post the result here.
What I am thinking of is definitely support for coding, but also support for generating the basic structure of the plant and its development.
An example of what I imagine is a prompt where I insert the following requests:
“Generate an organ module.”
“Generate leaf, stem, fruit, flower, and meristem as organ modules.”
“At the start, there are just 10 meristems representing individual plants. They are disposed in a 2x5 grid at a 50 cm distance.”
“At germination, the meristem generates a stem and two leaves.”
etc
Yep… just want an idea of the applications you are thinking are relevant, so I can see how its applied.
1 Like
Few examples are the following:
https://chat.openai.com/
Something like Github Copilot to help build FSPM would be great
Is this what you were asking?
Yes thanks, I’ll have a look…
1 Like
