Artificial Intelligence and the Future of Recirculating Aquacultures Systems
Image by DALLE. Text still has some issues
Artificial Intelligence (AI) has been in zeitgeist for well over a decade but it has reached new highs of mind share with release of new generative AI tools such as LLM powered chatbots. All industries and not just software companies are now taking AI more seriously and thinking how it will change the future of their work.
Aquaculture is also not new to AI. A number of technology start ups have been utilizing machine learning (ML) technology to improve feeding efficiencies, biomass estimation, and disease/sea lice detection. All of these technologies have been primarily aimed at salmon farmers by using computer vision in tandem with an underwater camera. A few other start ups have also focused on the shrimp industry and feeding optimization. In the recirculating aquaculture industry their have also been a few small startups but nothing has broken through yet to be widely adopted by the industry.
The one thing you need to train an AI system is lots of data. In RAS we primarily have data related to environmental conditions. This includes sensor measurements of temperature, oxygen, and pH. More advance sensors may also measure total dissolved solids or Redox potential. And you have data points from water quality samples processed in the lab for ammonia, nitrite, nitrate, alkalinity and potentially a long list of other parameters. You also have data for water flow rate, pump speeds, electricity usage, feeding weight and timing. You also have data on stocking density, weight samples, and harvest weight. The last potential data source could be underwater videos.
In a perfect world you could bring all of these data sources into a single platform and ask AI to mine it for insights and then ideally an AI agent could take over control of the system and make suggestions to the humans to optimize operations. For example the AI may identify that water temperatures should be held higher when young fish are growing from 10 grams to 100 grams but lower the temperature afterwards to improve growth rate. The truth is it is not so simple. You need a huge amount of labeled data, ideally hundreds of batches of fish already grown and all available data collected for each of those batches. You could then expect the AI to begin identifying patterns. No producer though has this data at the moment since most RAS operations are young and even older ones have not been focused on rigorous data collection.
No doubt AI will have an affect on the RAS industry, at first there will be small wins perhaps with water flow management by AI. Eventually we will see AI agents able to take over the control of systems and push alerts to human operators when anomalies arise.
If you know of anyone exploring AI and RAS please reach out, I would to love to learn more.