Patrick Taillandier / Stéphane Couture
My friend the LLM: just a human like any other?
LLMs for simulating human behaviour.
Semi-automated agri-food chain modeling from scientific monitoring data: the case of prospective benefit-risk analysis of the insect-based biowaste valorization chain
Various strategies contribute in adapting agri-food systems to climate change. One strategy consists in reducing the emissions of the equipment used. Another one focuses on diversifying the crop varieties towards more resistant ones. Another strategy aims to avoid any loss in the materials used and the products obtained along the chain. The research presented is positioned in the latter strategy. More specifically, it explores the possible valorization of end-of-chain biowaste towards food, feed, fertilizers, energy or other products, using the ability of insects to grow on biowaste.
This is still an emergent and non-stabilized chain which faces numerous difficulties, from social acceptability and safety guarantees to economic viability. Exploring the uncertain future evolutions of such emergent systems is the scope of prospective methods such as scenario building. Evaluating their risks and benefits regarding different criteria but also different stakeholder interests, involves a multidisciplinary approach combining multicriteria and multi-actor modeling, interacting and possibly conflicting arguments, and benefit-risk analysis, as recently applied in different agri-food use cases.
A bottleneck of such a holistic system analysis is the capacity to identify and to analyze a representative amount and diversity of data that provide information on the diverse criteria and stakeholder involved in the agri-food system. This is where the present work provides significant contributions. On the one hand, the provision of numerous relevant and recent data is ensured by exploiting scientific monitoring results focused on insect-based biowaste valorization, implemented over the past few years in collaboration with the INRAE Open Science Department. On the other hand, recent advances in AI text analysis tools will be used in a “RAG” (Retrieval-Augmented Generation) approach to provide a semi-automatic analysis of data contents and extract targeted information for benefit-risk analysis.
In this presentation, we will discuss key points to be taken into account when choosing most relevant AI tools, steps followed and validation methodology.