Pew Charitable Trusts hosted the first global summit on the topic of artificial intelligence (AI) in fisheries monitoring on 17 and 18 January, 2023.
The conference hosted data scientists, analysts, researchers, and policymakers to start the conversation on how AI can improve fisheries management.
The Pew Charitable Trusts International Fisheries Senior Associate Raiana McKinney summarized the three main topics discussed at the summit. The first, she said, is managing expectations between stakeholders so groups understand what is possible, what is required, and what can run into regulatory issues. The second is defining what success for an AI means.
"There's a fine balance between the minimum precision that's needed for accurate data versus how we define success for these applications,” McKinney said.
The last topic, she said, is actually moving AI from a "proof of concept" to a "fully integrated and approved tool."
"This is currently a slow process, and we heard from a lot of attendees that it's important for governments to clearly identify AI as an essential tool for existing management problems," McKinney said.
Multiple companies pioneering AI solutions for fisheries monitoring attended the summit, including CVision AI. CVision AI was founded approximately seven years ago to build AI and software for video-based analytics. The focus has been on activity recognition to reduce the amount of data that either must be collected or transferred to bring down overall costs.
“It can be extremely important when it comes to sustainable fisheries because the costs of things like at-sea monitoring, to get good fisheries management practices, is just not sustainable," CVision AI CEO and Co-Founder Benjamin Woodward said. "Therefore, we need to move towards electronic monitoring and reduce the costs associated with that, specifically video review, and that's where I see AI helping the most. Helping human analysts become more efficient at churning through large volumes of video.”
Woodward said there's a "hierarchy" of potential uses for AI, from easier to harder.
"The first is activity recognition. The broad idea of something interesting is happening that gets flagged by the AI. Then, a video analyst will look at this section of flagged video for further review," Woodward said. "This can cut close to 90 percent of the recorded video down to only the interesting fishing events, especially on multi-day or multi-week cruises."
From there on down, the steps get more complicated, with each one further reducing the workload on human observers.
"The next would-be basic object identification. Recognizing a fish in the video, so the AI knows that there was a fish on deck or something like that," Woodward said. "After that, would be trying to determine the species identification of that fish. And then finally, it would recognize there are this many fish within a video. Each of those steps down the hierarchy, reduces the amount of burden on human review but also increases the complexity and cost associated with building those algorithms.”
Another company, Teem.fish, was founded by Ecotrust Canada, but in was acquired by SnapIT in 2021. The company is an electronic monitoring service provider with the goal to produce cost-effective electronic monitoring with custom features for individual fisheries.
According to Teem Fish Monitoring CEO Amanda Barney, the Covid-19 pandemic shut down many observer programs and created demand for the uptake of electronic monitoring technology. This left companies with hours and days of video to analyze and rising costs. AI offers a solution to cut down the cost and work of video analysts, by cutting out large chunks of electronic monitoring video recordings that have nothing happening.
Barney explained additional benefits including, an increase in data confidence with 100 percent coverage, especially in fisheries where observers cannot be present all the time. As well as, rare species encounters caught on camera to aid in science and building a training set for these species."
However getting access to training data, Woodward added, is one of the many challenges AI implementation is facing.
“Of the biggest barriers in implementing AI in EM is the lack of good quality training data. And one of the main reasons for that is the legality and sensitivity around the data itself. As an example, the optics and data around the protected species interactions which are really something that is just part of fishing, needs to be thoughtfully accommodated in access to that for AI training purposes," Woodward said. "If we want to be able to reduce interactions, then we want to be able to really focus algorithms in to see those things. But they can only see those things if you can train them on it. And you can only train them if you have access to data on them. And if you don't want anybody to see that data then you can't train an algorithm on it. There is a siloed aspect of who owns the data between governments and electronic monitoring providers, as well as legal contracts that some vendors have preventing data sharing, that are kind of large-scale barriers."
Beyond barriers to the need for large datasets for machine learning, Barney said there are still questions regarding some of the benefits that have been listed for AI use in electronic monitoring - such as the costs associated with an AI program.
"We have very clear ideas about how automation could bring down the costs of that electronic monitoring program. But, until we know all the costs of using and maintaining that automation, we don't know the overall savings to the overall program,” Barney said. “What we really dove into at the summit with Pew is this ability for AI developers, data scientists, electronic monitoring service providers, and regulators to have the opportunity to all start using the same language and for different folks to understand and for regulators to understand, that you actually have to maintain machine learning and that potentially if your fleet changes, or new vessels come in, that you might have to add to or change your training set over time incurring maintenance costs.”
Photo courtesy of Dogora Sun/Shutterstock
