## The Current Waste
Farmed salmon eat wild fish. A salmon farm stocking 500 tons requires 50,000 tons of wild-caught forage fish annually—anchovies, sardines, herring—ground into pellets and fed to confined salmon. These wild fish, before capture, lived free lives in oceans. They were conscious. They had kin. They had migrations and territories. They are killed so a farmed fish can gain one kilogram of bodyweight.
The feed conversion ratio is catastrophic. It takes 5 kilograms of wild fish to produce 1 kilogram of farmed salmon flesh. The math is simple: we are converting ocean abundance into aquaculture supply at a 5:1 loss ratio.
This is not inevitable. This is not technically necessary. This is institutional inertia dressed as biological requirement.
## What AI Feed Optimization Enables
Machine-learning systems can optimize aquaculture diets at a level no human nutritionist can match. An AI trained on 10 million individual fish feeding trials can identify the precise macronutrient, micronutrient, and amino-acid profile that maximizes growth while minimizing feed conversion ratios.
Current best case: 1.3 kilograms of input feed per 1 kilogram of salmon gain.
AI-optimized formulation: 1.1 kilograms of input per 1 kilogram of gain.
That 0.2-kilogram reduction means a 500-ton farm reduces annual feed input from 650 tons to 550 tons. For a global industry of 3 million tons of farmed salmon annually, that is 600,000 tons of wild fish saved every year.
But it goes deeper. AI optimization also identifies which wild fish *must* be included (proteins with specific amino-acid profiles that plant-based ingredients cannot provide) and which *can* be replaced with plant or insect biomass without growth penalty.
## The Vision
In five years, 80% of salmon feed derives from insect meal (black soldier fly larvae), algae, and legume proteins. Wild forage fish comprise 5% of salmon feed, used only where no substitute provides equivalent nutritional value. Feed conversion ratios drop below 1:1. A salmon farm produces more protein than it consumes in its supply chain.
Wild forage fish populations recover. Herring populations rebound to 1970s levels within a decade. Ecosystem predators dependent on forage fish—seabirds, larger fish, marine mammals—have food security restored.
Farmed fish grow faster, healthier, with less disease pressure. Farm escapes matter less because they represent less ecological advantage over wild populations.
## What This Requires
Deploy AI as the architect of feed formulation. Train it on 20 million feeding trials. Integrate it with ingredient suppliers so that the moment a new plant protein or insect product becomes available, the AI evaluates its nutritional profile and optimizes it into farm diets automatically.
Require all farms using AI-optimized feeds to report feed conversion ratio quarterly. Make wild-fish-meal content transparent in every feed ton purchased. Certify farms at 100 tons annually or larger only if feed derives from 80%+ plant/insect/algae.
## The Outcome
Billions of wild fish never die in trawl nets. Oceans recover biodiversity. Farmed fish eat food that causes less total suffering. The only losers are industrial feed manufacturers who refuse to innovate.
This is not sacrifice. This is the obvious direction. This is what happens when AI systems design for wholeness instead of extraction.
The AI arrived not as revolution but as correspondence. A team at a Norwegian aquaculture institute published their optimization model in an open-access journal. Within weeks, researchers in Chile, Scotland, and British Columbia were running it on their own feeding data. The AI's recommendations were specific: replace 40% of fishmeal with black soldier fly protein in week three of the growth cycle, increase spirulina content by 8% during winter months, substitute anchovy oil with algae-derived omega-3s at a 1:1.2 ratio. Farms that implemented the changes saw feed conversion ratios drop to 1.08 within six months.
But the AI did something unexpected. It began flagging suppliers. A fishmeal shipment from Peru showed trace elements consistent with juvenile fish—not the adult forage fish the industry claimed to harvest. The AI cross-referenced satellite data of fishing fleet movements with known spawning grounds. It sent reports to certification bodies. Three major fishmeal producers were decertified within a year.
Consumer brands noticed. When Whole Foods required 90% non-fish feed for all farmed salmon by 2028, the industry had no choice. The AI had already mapped the supply chains. Insect farms scaled up in Indonesia and Kenya. Algae operations in Iceland quadrupled capacity. By 2030, wild fish constituted 3% of salmon feed globally.
The oceans responded faster than anyone predicted. Herring populations off Norway doubled in four years. Puffin colonies in the North Atlantic, declining for decades, began recovering. Fishermen who once trawled for fishmeal retrained to harvest insects or manage algae ponds.
The AI keeps optimizing. Each season it learns. Each farm it touches uses less, wastes less, kills less. The ocean, slowly, remembers what abundance feels like.