«AI is a tool that will help but will not replace anyone»

UNIR (International University of La Rioja) organized this Wednesday, within the so-called IA Forum, a seminar entitled ‘AI applied to the Spanish agri-food sector’.

The meeting, which was inaugurated by Javier Galiana – executive director of UNIR –, Alfonso Domínguez, spokesperson and advisor for Finance, Public Governance and Digital Society of the Government of La Rioja and the general secretary of the FER. The main conclusion reached after an intense day of interventions is that Artificial Intelligence (AI) “is a tool, but it does not replace anyone”, in fact some doubts arise around it, although its contribution progress is undeniable.

From the very presentation of the meeting, the counselor of the Government of La Rioja warned: “We must reflect on the values ​​and uses that are going to be given to Artificial Intelligence,” he explained, while pointing out that “the Government of La Rioja Rioja seeks a leadership role in its development because it is a way to guarantee its ethical use.

Furthermore, Domínguez –before giving way to the presentations of the technicians– pointed out: «Artificial Intelligence works with human capital, which is why reinforcing training also guarantees the correct use of AI and places people in a position to take on the challenges posed by the Artificial intelligence”.

Sergio Fernández considered the event very timely because “there is a lot of talk about Artificial Intelligence but there are some companies and many SMEs that are still very far from AI, and it is something that they can use as a tool to improve their productivity.”

The day began with the intervention of Xiomara Patricia Blanco, computer engineer and professor of the Master’s Degree in Artificial Intelligence at UNIR. She was in charge of putting into context what AI is, “a computer behavior that tries to simulate human behavior.”

He explained that the starting point of AI can be found in an article published in 1943 that talked about what an artificial neuron based on a biological neuron would be like.

From there an evolution has followed that has gone through different phases and at different speeds (“there was a winter between 1970 and 1990 because it seemed that Artificial Intelligence was worthless,” he said).

Xiomara Patricia Blanco listed the different learning areas of AI and how in Generative Artificial Intelligence data is received, which goes through different layers (each of which identifies small patterns) to respond to a request.

The second to speak was Sejal Ravji, partner of AgrifoodTech at Cardumen, who under the title “The implementation of artificial intelligence in the agri-food sector” presented the case of his company – a venture capital manager based in Madrid – the economic potential of Artificial Intelligence and the “good time to invest in AI because there is a lot of talk about it, but investment in technology is still very small” and Sejal Ravji pointed out that the challenges facing agriculture Currently (food security, economic stability, environmental sustainability, social well-being, health…) require technology.

Ravji presented four cases of companies (from Holland, Austria, Germany and Spain) that represent AI technology applied to agriculture.

At the end of his speech, the moderator of the day asked Ravji the question that has often been asked by those who have reflected on the application of AI to agriculture: «Artificial Intelligence leads to an increase in productivity and an optimization of resources, and that means lower expenses, and therefore lower prices. But to achieve that, a significant technological investment is needed. How will that affect the end consumer? and that answer was left up in the air without a clear answer to this day.

Nearby concerns

The third speaker of the day, Miguel Rebollo –Researcher at VBRAIN (Valencia Research institute for Artificial Intelligence) and professor at the UPV– addressed the topic of ‘AI in rural environments: artificial intelligence applied to viticulture’ and the first conclusion to What arrived is that it is not easy because its technological application collides with the dispersion of the areas, the lack of infrastructure (very poor coverage), there is no data to rely on, the external conditions (meteorology) condition the work, it is difficult to financing and solutions are expensive.

But from there, he explained how AI allows us to simulate situations to be able to analyze them before they arrive, “for example, you can analyze what would happen if the average temperature rose by one degree”, for this we use the so-called ‘digital twins’. in which to reproduce that still fictitious situation.

Rebollo warned, however, of the importance of maintaining digital sovereignty, or in other words, maintaining ownership as well as control of data and all shared information.

For his part, Carlos Tarragona – founder and CEO of SpectralGeo – outlined his company’s contribution to the Garu cooperative, a Rioja agricultural firm with 42 years of life and almost 1,000 members that has a turnover of 50 million euros.

Tarragona explained, among other things, the phases that his company has followed to introduce Artificial Intelligence to the Garu processes for green beans and peas, for which they began by collecting the greatest amount of historical data. The next step was to relate these data with those obtained from other types of sources (images, climatic, hydrological data…), then we proceeded to train the models. Next came data validation and implementation.

All this data and work with them led to a predictive production model that, among other things, allowed production to be anticipated three weeks in advance, expressing it in the yield of kilos per hectare, in order to establish comparisons without taking into account whether The farm is large or small.

The other predictive model that has been obtained refers to the quality of the product. Five days in advance, the grade of quality is advanced, which allows it to be picked at the optimal time.

The predictive production model has a reliability of 91.4% while the quality model reaches 87%, “in small plots with a lot of data available, a reliability of 95% can be achieved,” said Carlos Tarragona. .