As the stack will collect various sources of data under a unified platform, data privacy and data sharing protocols will become necessary, where data can be anonymized when needed and strict standards of access will need to be developed to protect the most vulnerable. The stack-based approach might risk monetization by a few and consequentially increase the data gaps, thereby limiting the application. Data standardizing and operability Once data is integrated into the stack, it needs to be standardized into a consistent format that can be processed by AI algorithms. This includes cleaning the data, resolving inconsistencies, ensuring platform compatibility, and applying consistent metrics. A set of automated ETL (extract, transform, load) pipelines can synthesize data from various sources into a unified format (i.e. clean/normalize it) and load it into a central database for analysis.20 A cloud- based data lake – e.g. AWS,21 Azure22 or Google Cloud23 – can act as a scalable storage solution for all collected data, making it easier to aggregate and access. Here, open data standards – e.g. JavaScript Notation (JSON), comma-separated values (CSV), extensible markup language (XML), representation state transfer (REST) application programming interfaces (APIs) – are implemented to ensure that data from different systems can be easily integrated and exchanged across platforms.24 Data interpretation, analysis and forecasting Once data is available in a standardized form, the next layer of the stack applies AI and machine learning models to interpret the data, uncover patterns and forecast trends (e.g. crop yields, drought predictions, soil fertility decline and/or water use). AI models require large amounts of high-quality, diverse data to generate accurate insights. However, in many cases, particularly in underdeveloped or remote regions, this data may be incomplete or low quality.25 To solve this, the layer can use the following techniques: –Synthetic data generation uses simulated datasets (e.g. from climate models or remote sensing) to fill gaps in real-world data, thus improving model accuracy.26 –The layer can use models that are pre-trained on similar problems (such as crop yield prediction or weather forecasting) and have been adapted to specific local conditions, reducing the need for large volumes of local data. –The stack should also incorporate crowdsourced data from farmers and local communities via mobile applications to supplement sensor and satellite data, increasing both the quantity and diversity of available data.27 This also acknowledges the unique role of agronomists and advisers in checking for the accuracy of such output as part of data quality assessment. –It is necessary to conduct regular bias audits of the AI models, using fairness metrics to test whether the predictions are disproportionately skewed towards or against certain populations (e.g. smallholder farmers in low-income regions). It is similarly important to engage with local stakeholders to source input for the model development process.28 The ethics of the AI should be continuously monitored to ensure that the food-water stack remains a public good. AI as a companion BOX 5 To assess the complex interactions in food and water systems, AI models that integrate data from various sources (e.g. climate data, market data, soil sensors, socioeconomic data) should be considered to generate more holistic predictions. Furthermore, it is also essential to implement agent-based models in conjunction with AI to simulate interactions between different agents (e.g. farmers, markets, policy-makers, water systems) and environmental factors.29 Finally, it is crucial to use AI to generate scenario-based forecasts rather than singular predictions and incorporate feedback loops into the AI model so that the system continually learns from real-time data and adjusts its predictions accordingly.30 This can help the model keep up with the ever-changing nature of food and water systems. Ultimately, AI models are meant to complement, not replace, existing and future human intelligence. Food and Water Systems in the Intelligent Age 15

Food and Water Systems in the Intelligent Age 2024