The sheep, cows, calves, bulls and pigs will wander in pavilion 1, the main one. Horses and ponies, in 6. Dogs and cats will bicker together, in 7. And the coders? You will find them in hall 4. It is one of the little attractions of the year at the Agricultural Show, scheduled for the end of February in Paris. The rural high mass is organizing its very first hackathon dedicated to generative artificial intelligence (AI), a technology that is currently booming. La Ferme Digitale, the community of AgriTech start-ups and businesses, is overseeing this IT competition with a prestigious partner: the French nugget Mistral, valued at more than two billion dollars in the AI sector. The latter makes its open source Mixtral 8x7B model available, one of the most powerful in the world. Eight teams will have 48 hours, on February 26 and 27, to use it and build, thanks to it, an application that could be useful to farmers.
Several ideas should logically find root in the anger that runs through the agricultural world. For example, those making it possible to reduce the administrative burden on operators. We thus find, among the declared objectives of the hackathon, “the deciphering of French and European standards and regulations”, at the heart of the tensions of the moment. “We established the program before the discontent and anticipated the news,” breathes David Joulin, one of the organizers on the side of La Ferme Digitale. “But these problems have been recurring for many years…” This son of a farmer and founder of the start-up Ekylibre, specializing in the digital management of farms, estimates the time spent by farmers on paperwork at around “500 hours per year”. A calculation in line with other studies, such as that of the agricultural barometer Terre-net/Bvawhich in 2015 estimated the administrative work of operators at 8.8 hours per week, or around 422 hours per year, the equivalent of one third-time employee every 35 hours.
“Farmers are particularly consumed with online forms and subsidy files to fill out, where they are asked for a summary of their activity, their project,” notes David Joulin. A mission largely within the ropes, today, of the large language models (LLM) of artificial intelligence, such as Mixtral or ChatGPT. Just like the response to simple legal requests, the elements of which are found in the Rural Code, but which sometimes cost “hundreds of euros in legal fees for operators”, deplores the boss of Ekylibre. Learning and helping to consult this type of text, again, seems within the reach of AI.
Extensive open access data
Of course, the potential of an agricultural “ChatGPT” goes well beyond the administrative question. Support for “crop management”, “herd management” or assistance with various broader agronomic problems are also among the objectives of the hackathon. A Dutch study published last summer in Nature Machine Intelligencejudges LLMs capable of guiding farmers in the future in the creation of personalized robots, for example for picking tomatoes.
Numerous data, most of which are freely accessible, also offer the possibility of creating virtual assistants useful to farmers in their daily lives. On aspects linked to climate change, for example. “Météo-France has put 100 years of French weather data online,” recalls David Joulin. Or on fertilizers and pesticides: the data.gouv site notably hosts the E-Phy catalog of plant protection products and other fertilizing materials. A first “kit” comprising a total of 150 datasets was put online on January 29 by Ferme Digitale, in order to inspire participants in the hackathon. Simple appetizer. It is very likely that tomorrow, farmers will be able to connect their own data to it, in order to make AI completely customizable.
Will they be tempted? More shared data could lead to ever more standards, some specialists fear. But others predict that acculturation to this new era of AI will be rather smooth. “Farmers’ history with technology is not new. They have been using GPS since the 1990s and data to manage their farms for around ten years. Artificial intelligence is already irrigating robots, sensors, which help decide on the date of sowing, harvesting, or irrigation”, explains Jérôme Le Roy, president of La Ferme Digitale and founder of the start-up Weenat, specializing in connected weather stations. France is also the third country in the world in terms of number of AgriTech start-ups created per capita, according to government data, and more than two billion euros have been invested in the sector, with that of food, in the framework of the France 2030 plan. In short, “Tech” has already conquered the agricultural world.
Diagnosing a sick plant with a photo
“But this technology must become financially affordable,” points out David Joulin. The automation of certain tasks requires equipment and training. Even if France is technologically ahead, we are today faced with a paradox: it There are already more than 1,500 digital solutions on the market, for a very average adoption, of around 10% to 15%, of these tools. Several obstacles remain: “The introduction to new technologies in agricultural schools is still insufficient, as is the collaboration between the different actors of innovation – large groups, research institutes and small start-ups -, without forgetting the deployment broadband, still uneven across the territory”, lists Jérôme Le Roy. “Could AI replace the agricultural technical advisor tomorrow? We are still very far from it,” tempers in a video consultant Arnaud Rey, innovation specialist at Crédit Agricole.
The incredible enthusiasm around AI in the world should, however, accelerate the movement. In the United States, the Farmers Business Network (FBN), a network of farmers, launched “Norm” last spring. This tool delivers a host of agronomic advice, boosted by the GPT 3.5 version of OpenAI. It currently constitutes the closest thing to a “Farmers’ ChatGPT”. Tractor king John Deere is aiming for fully autonomous corn and soybean production using AI by 2030.
Other big names in Tech, like Google and its start-up Mineral, have begun massive work to collect data on soils and plants, with the aim of feeding the virtual assistants of tomorrow. Finally, in India, the “Plantix” application provides diagnoses on diseased plants, using a simple photo of a crop. Several million people already use it. In a study, the American consulting firm Gartner estimates that 8 out of 10 companies in the world will have experimented, within two years, with generative AI applications or models in their production environment. It’s a safe bet that farmers will be there.
.