The AI revolution comes to crop and seed innovation
Artificial intelligence is beginning to reshape how new crops and treatments are invented. The implications go well beyond technology.
Why AI has arrived at the right time
For most of the past century, agricultural innovation has followed a predictable rhythm: observe, test, learn, repeat. Breeders coax new plant varieties into being; scientists concoct promising treatments; both are then subjected to years of trials in laboratories and muddy fields. The method has fed a growing world, more or less. But as the low-hanging fruit has been picked, new gains are proving harder to reach, and the rate at which genuinely new crop-protection products reach the market has been falling for decades.
Artificial intelligence arrives at an opportune moment. Our new whitepaper, How Artificial Intelligence is Transforming Agricultural Innovation, draws on voices from across the sector to show how AI is being applied across the innovation lifecycle – and its potential to deliver faster, more targeted and more sustainable product development.
From prediction to creation
One of the most talked-about changes is the "dry lab", in which more experimentation happens in silico, with AI applied to a company's own data to create designs of new ingredients or formulations – so that physical trials begin further along and run more efficiently. The promise is not new, but newer forms of generative AI are unlocking fresh value. Whereas ‘traditional’ AI learns patterns from historical data to make predictions such as whether a seed treatment will improve germination, a generative model can propose an entirely new formulation for scientists to evaluate.
DNA is basically a sequence of letters, which means it is easier to process with today's language models
AI and the future of crop genetics
Genetics is proving unexpectedly amenable, for a pleasingly simple reason. “DNA is basically a sequence of letters, which means it is easier to process with today's Language Models,” Tonny Otjens of Rijk Zwaan told us. His firm is exploring how such models can combine phenotypic observations with genetic data to predict which varieties will deliver desired traits – insights that could also drive the fast-growing market for biological coatings and treatments.
Quick wins
Not every gain is so cerebral. Some of the most popular applications simply strip away drudgery, automating the documentation and compliance work that devours researchers' time. Rijk Zwaan built a tool for the thankless task of naming products, screening candidates against trademark, linguistic and regulatory requirements. What once took weeks – and could end in rejection if a name echoed a Danish rival or a Spanish swear word – now takes minutes. Unglamorous, but obviously useful. And such quick wins often convince sceptics that the technology is worth the bother.
Advantage comes from combining frontier models with the best proprietary data, scientific expertise and market insights
The real competitive advantage
Yet a central insight is that none of this, on its own, confers lasting advantage. The frontier models everyone marvels at are, in the words of BASF's Stephan Köhler, ‘commodities’. Advantage comes from combining them with the best proprietary data, scientific expertise and market insight. Firms with significant proprietary data – created or acquired – will end up with the best models. The years Rijk Zwaan spent photographing plants for observational studies have become a treasure trove of R&D data.
Rethinking innovation
But AI also heralds change in how the work is done. Formulation scientists, Mr Köhler notes, typically use experience to pick promising starting points, then refine them through experimentation. In an AI world, the sequence partially inverts: generate large volumes of data first, train models on the results, then use those models to explore possible solutions. That shifts the duller work to the very start but expands the options and accelerates optimisation later.
Building an AI-ready organisation
Which is why the report insists the real obstacle is organisational, not technological. Agriculture will need better data, gathered and standardised with future models in mind. It needs new organisational thinking and structures; a rigorous, auditable approach to trust and explainability; and an AI-ready workforce comfortable directing models rather than merely using them. And it will need an ecosystem of technology and research partners beyond anything traditional R&D is used to.
Where to start
None of this will be easy, and the whitepaper ranges far wider than can be captured here – across knowledge graphs, foundation models, AI audits, human-AI teams and the first steps leaders might take. For anyone in the business of growing things – and wondering how the biggest shake-up of agriculture in a generation will affect them – it is a good place to start.
Download the whitepaper: How artificial intelligence is transforming agricultural innovation
White paper: How artificial intelligence is transforming agricultural innovation
