This is an automated rejection. No LLM generated, assisted/co-written, or edited work.
Read full explanation
My example is AI sea-lice control in Norwegian salmon farming. Stingray has built over 3,000 machine-vision lasers that find lice on swimming salmon and shoot at them, and says it is on a quarter of Norwegian farm sites already. Around 30 companies also count lice with automated camera systems under a temporary exemption, though none of these counters is approved yet.
Norway produces over half the world’s farmed Atlantic salmon, about 400–450 million salmon and trout a year. About 55 million salmon died at sea in 2025, and fish-health staff rank delousing injuries as the leading cause of death and poor welfare. Cleaner fish (19 million used, 12.7 million died), wild salmon and the lice themselves are affected too.
Net impact per fish is probably a gain: lasers can replace delousing methods that handle the fish, which are linked to higher mortality. In total, I think it is more likely harmful, because lice are what Norway uses to cap growth. Every second year, each production area gets a traffic-light colour based on how much farm lice harm wild salmon: green areas may grow up to 6%, red must shrink 6%. The rule ignores how the farmed fish are doing. AI can loosen that brake in two ways: lasers may lower lice, and automated counts are becoming the evidence that decides who may grow. Whether lasers actually reduce lice is unproven, and counter accuracy is disputed. But if AI freed up as much growth as the traffic lights have since 2018, roughly 7–10% of capacity, some 30–45 million more fish a year would go into pens, and at today’s death rate 4–5 million more would die. Gentler delousing might save a similar number, but then tens of millions more fish would spend their whole lives in pens.
Norway’s parliament has ordered a scheme that lets farms with low lice emissions use extra capacity, and the minister says it depends on automated counting. It is stuck because “low-emission” has no definition yet, which gives advocates a rare chance to decide what the machines must measure. The same AI cameras that count lice can already see wounds: MSD’s Falcon, launched in Norway in 2025, reports both. Advocates should push for the definition to put a welfare test next to the lice counts: growth only for farms with mortality in the lowest quarter of their area, as two Norwegian veterinary associations have proposed, with the cameras’ wound data reported too. Whatever definition comes first will be hard to change. My prediction: by the end of 2027, the first definition will let automated lice counts unlock growth with no farmed-fish mortality criterion.
My example is AI sea-lice control in Norwegian salmon farming. Stingray has built over 3,000 machine-vision lasers that find lice on swimming salmon and shoot at them, and says it is on a quarter of Norwegian farm sites already. Around 30 companies also count lice with automated camera systems under a temporary exemption, though none of these counters is approved yet.
Norway produces over half the world’s farmed Atlantic salmon, about 400–450 million salmon and trout a year. About 55 million salmon died at sea in 2025, and fish-health staff rank delousing injuries as the leading cause of death and poor welfare. Cleaner fish (19 million used, 12.7 million died), wild salmon and the lice themselves are affected too.
Net impact per fish is probably a gain: lasers can replace delousing methods that handle the fish, which are linked to higher mortality. In total, I think it is more likely harmful, because lice are what Norway uses to cap growth. Every second year, each production area gets a traffic-light colour based on how much farm lice harm wild salmon: green areas may grow up to 6%, red must shrink 6%. The rule ignores how the farmed fish are doing. AI can loosen that brake in two ways: lasers may lower lice, and automated counts are becoming the evidence that decides who may grow. Whether lasers actually reduce lice is unproven, and counter accuracy is disputed. But if AI freed up as much growth as the traffic lights have since 2018, roughly 7–10% of capacity, some 30–45 million more fish a year would go into pens, and at today’s death rate 4–5 million more would die. Gentler delousing might save a similar number, but then tens of millions more fish would spend their whole lives in pens.
Norway’s parliament has ordered a scheme that lets farms with low lice emissions use extra capacity, and the minister says it depends on automated counting. It is stuck because “low-emission” has no definition yet, which gives advocates a rare chance to decide what the machines must measure. The same AI cameras that count lice can already see wounds: MSD’s Falcon, launched in Norway in 2025, reports both. Advocates should push for the definition to put a welfare test next to the lice counts: growth only for farms with mortality in the lowest quarter of their area, as two Norwegian veterinary associations have proposed, with the cameras’ wound data reported too. Whatever definition comes first will be hard to change. My prediction: by the end of 2027, the first definition will let automated lice counts unlock growth with no farmed-fish mortality criterion.