Generative AI is pretty good at speaking the language of humans, but one European startup is now building a generative model to help us speak “the language of plants”.
Hortiya — founded in 2020 in the west German city of Duisberg — is now applying the technology that powers large language models (LLMs) like GPT-4 to the food production industry.
And while many headlines highlight how great generative AI is for helping schoolkids cheat at homework, or making high-fashion deepfakes of the Pope, Hortiya's CEO Marc Weimer-Hablitzel says his company will leverage the technology to dramatically reduce the amount of energy that goes into making our food.
Hortiya is currently building a dataset for training an AI model that can understand how different inputs, or climatic conditions, affect the internal systems and growth of plants. If it can do this, the company could increase the energy efficiency of food production by allowing growers to plan how they use expensive resources like electric light and fertiliser.
Weimer-Hablitzel says that building Hortiya’s foundation model will likely require a different approach to existing LLMs, but it'll be based on the same principles. Some data will be gathered from greenhouse sensors that measure how plants respond to changes in light levels, heat and soil conditions, but this will be augmented with more cutting-edge information.
He explains that this might include data on how plants’ roots send electrical messages to their leaves, or new research on how stressed plants emit high-frequency sounds inaudible to the human ear.
Stir in a little market and weather data, and the model will be able to predict how events like droughts or spikes in energy prices will affect growers and how they allocate resources.
Bigger is better
Hortiya’s AI model represents an evolutionary use of the sensors and data being increasingly deployed in agriculture. Hortiya has already developed a product that combines greenhouse sensors and AI to help food producers to use lighting more efficiently, optimising for when plants are most receptive to absorbing energy.
“The plant has to be in a certain state: plants have rest time, they can be in a stress state. If a plant’s neighbour is sick it will perform less well,” says Weimer-Hablitzel. “We only want to give energy to the plants when they are ready to metabolise it.”
This technology is already being used in two research pilots and by one commercial client in a 54k sq ft greenhouse. Weimer-Hablitzel says that Hortiya has proved in the lab that its existing technology can reduce energy consumption in plants while maintaining the same yield by as much as 30%.
He warns that the savings are likely to be lower in a real-world setting, but stresses that even by saving 10% of energy consumption, his startup would be making a big impact if deployed on a large scale.
“Saving 5-10% is already a huge impact, considering how many megawatt hours are actually burned in greenhouses,” says Weimer-Hablitzel. “You can't imagine how much our tomatoes cost in terms of energy so that we get them in winter… The product that we currently do is already creating value for our clients.”
Hortiya’s large general-purpose model, it says, will take these potential energy savings to the next level, by allowing growers to make longer-term planning choices around how to deploy resources like light and nutrients.
“You’ll be able to ask this model things like: 'This is how the month is going to look like, what would be the best strategy if we had, say, two weeks of drought, or if electricity and we could add additional light?’” Weimer-Hablitzel says.
The perfect VC storm
Like most research-heavy startups, Hortiya says it’s always in fundraising mode and has currently raised $1m in pre-seed funding — a mix of grants and convertible loan notes.
And, while the funding environment is tough out there, Hortiya has a couple of tailwinds blowing in its favour. The meteoric rise of ChatGPT has made investors more aware of the concept of AI foundation models, and at the same time more and more VCs understand the urgency of solving the climate crisis.
“You see that in Britain, France and Spain there’s no water available, even in winter. So even investors start understanding, ‘Oh, my God, what do we do if open-field [agriculture] doesn't work anymore if there's no water available,” Weimer-Hablitzel says. “There's a huge need for a more efficient and more sustainable way of growing, and investors have started understanding that as well.”
Hortiya still has a long way to go in building a whole new class of AI foundation models that can help us interpret plant signals, and training these systems on large amounts of data is expensive.
Weimer-Hablitzel believes that, by the time his startup is ready to train a model, it’s likely that the process will be more efficient than it is today. Whether or not it's able to find enough data, capital and talent to build its model successfully, it’s refreshing to see a generative use case that goes beyond using ChatGPT to write viral Twitter posts about use cases for ChatGPT.