June 2026 talk of the monthly meeting of the #animal-genomics special interest group.

We are happy to introduce Miguel Pérez-Enciso (Centre for Research in Agricultural Genomics (CRAG), UAB campus, Spain). Miguel Pérez-Enciso is a Biologist who obtained his PhD in 1990 in Genetics (Universidad Complutense, Madrid). After his PhD he moved to the USA and France for three years of postdoctoral studies, specialising in Bayesian Statistics applied to Animal Breeding and Quantitative Genetics. He worked at the Institut de Recerca i Tecnologia Agroalimentaria (IRTA) from 1993 to 1999 and at INRA (Toulouse, France) from 1999 to 2003, when he became an ICREA Research Professor. He is currently based at the Centre for Research in Agricultural Genomics (CRAG) on the UAB campus. From 2022 to 2024 he was on leave working as a Quantitative Geneticist for Corteva Agriscience, one of the largest plant breeding companies.

Reflections on AI for breeding (and beyond)

In this talk, I will review some of the impacts that, in my opinion, AI is having on the breeding industry and academy. Given that AI has been around for decades, the first issue is: Why all the fuss now? I argue that a main reason is that, while we can accept that AI generates images, numbers, even music, the fact that modern large language models (LLM) seem to ‘understand’ and produce text of such realism — a so far distinctly human feature — is disturbing. As a ‘consolation’, we should be aware that these models have a huge number of parameters, e.g., about 1M parameters per Wikipedia article. Focusing on breeding, genomic prediction is one of the early topics where AI has been applied. However, despite numerous efforts, its advantage over linear models has not been convincingly proved. Another area where AI is currently undergoing rapid development is generative AI. The purpose of generative AI is to produce samples of any arbitrary complexity by ‘mimicking’ training data. There are similarities but also important conceptual differences with standard simulation used in breeding. Generative AI can have important advantages relative to standard simulation, e.g., it is not limited by following a defined statistical distribution. AI can then be used to generate, e.g., shapes, plant architecture or animal conformation. Finally, given the speed of AI developments, we can speculate whether AI will supersede standard quantitative genetic principles or even Statistics. To conclude, AI offers endless opportunities for breeding companies to optimise their schemes and offers numerous opportunities for academia to understand AI results. There are important risks, though. In my opinion, a main one is reproducibility. A second one is learning itself and the advancement of science: will AI compromise our interest in learning and criticism? Will AI make it more difficult to go beyond established boundaries of knowledge?