“One of the problems with AI is, it sometimes gives you unpredictable results that can mess up your system,” said Rackesh Rajan, a research scientist at the Freshwater Institute, who develops AI tools for land-based aquaculture. “There should be a human in the loop who can check and then do the final action.”

Data privacy is also a concern as fish farmers hand over sensitive business and technical data to AI companies, which sometimes anonymize and share their data with feed and pharmaceutical companies and other suppliers to help them hone their products. Regulations are needed to protect data privacy, and farmers need to be clear when they sign agreements with AI companies about how and where their data will be used, Rajan said.

The costs can be a barrier for many farmers. AI’s high price tags have favored adoption by the biggest players. The AI hydrophone, for example, costs tens of thousands of dollars per pond, Bureau said. However, AI companies are now offering aquaculture tools for a monthly service fee, making them more accessible to a wider range of operations. ReelData’s biomass and feed programs cost a couple thousand dollars a month, Zimola said. “Any farm that’s producing at least 300 metric tons of fish a year can get value from our system.” AI-equipped cameras, which can cost a million dollars to buy, can also now be rented.

And performance claims aren’t yet backed by independent data. ReelData’s feed program increases the growth rate for fish by an average of 10 percent, “turning a 1,000-metric-ton farm into a 1,100 metric-ton-farm,” Zimola said. The program reduces the feed conversion ratio (the amount of feed going into a system versus the amount of food produced) by about 10 percent. Though he’s confident in those numbers, Zimola said, he “isn’t there yet” with publishing the data.

In fact, no independent studies yet exist to evaluate how these emerging AI systems are improving feed operations, fish health, or water quality. “We’d love to see more peer-reviewed data, as opposed to self-reported data from the companies themselves,” said Sophie Williamson, senior researcher in animal welfare at ReThink Priorities.

A Double-Edged Sword?

While AI-enabled tools may solve some industry challenges, they could also lead to larger farms—and the impacts that come with them. “Detecting earlier disease, improving water quality, reducing overfeeding are all positive things for animals,” said Williamson. But there’s a flip side, she said: “We would expect farms may want to stock more aquatic animals.”

Williamson is particularly concerned that farms may increase stocking densities if AI helps them reduce or better monitor disease, and that could have other animal welfare impacts.

Alternatively, as companies increase efficiencies and reduce costs, they may expand the size of their farms. That is likely true, according to Tony Chen, founder of Manolin Aqua, an AI service that uses machine learning to predict fish health. “When you talk to farmers, they all want to make more money [and] bring more fish to the market, but . . . the piece that created uncertainty in their businesses is fish health.”

The MARA Act, which supports deep-ocean aquaculture research, is moving through Congress. If it passes, the legislation could lead to larger fish farms in U.S. waters, and potentially a wider adoption of AI systems within the industry.

For Vinci, AI tools might help clean up some of the salmon industry’s problems. “I do think we’ll see improvements in things like identifying fish behaviors that show stress, or showing ulcers on fish or lice on fish, and having ways to quickly address that before it gets out of control,” he said.