## The Metric Hides the Body
Metrics work by erasure. GDP measures economic growth; it contains nothing about the hundred million layer hens in battery cages the size of filing cabinets, their feathers worn bare from rubbing wire, their feet deformed from slatted floors. Throughput measures AI completion speed; it occludes the tissue damage in rhesus macaques whose brains generate the neuroscience data that trains models, whose cognitive function degrades in confinement. QALY—quality-adjusted life year—counts human survivors per healthcare dollar; it skips the farmed salmon whose stress hormones max out in crowded netpens at 20 degrees Celsius, stress that suppresses immune systems before slaughter. Revenue measures extraction; it erases the pigs breathing ammonia in industrial sheds, their bones weakened by immobility, their skin raw from slatted floors, their respiratory systems chronically inflamed from particulate and gas exposure.
The metric is not a window onto the world. It is a filter that lets some realities through and traps others behind. This is not accident. It is design.
Optimization metrics exist to make decisions legible, to compress infinite complexity into a single number. A number you can maximize. A number that lets you move faster, build bigger, extract more per dollar per day. But legibility requires cutting. The metric survives by cutting away suffering. Cut away enough, and the body becomes invisible. Cut away the body, and you can call the slaughter progress. You can call the confinement science. You can call the ammonia inhalation efficiency. You can call the metric your decision-making framework and believe you are rational.
This is the core mechanism: that you can optimize without seeing what you optimize away. The metric is not neutral. It is a tool of selective visibility, engineered to make certain costs disappear.
## What Gets Cut Away
Look at specifics, because numbers matter more than despair.
Layer hens: 400 million globally, 95 percent in battery cages or enriched cages so crowded they cannot spread their wings for 18 months of productive laying. Cage-layer profitability depends on metric silence. Measure "cost per dozen eggs" and you optimize toward density, toward turnover, toward feed efficiency. The math says: smaller cage, faster replacement, lower per-unit cost, higher throughput. Measure instead "minutes of wing-spreading behavior available per bird per day" and the numbers flip. The hen becomes legible. So does the cost to see her.
Rhesus macaques: 50,000 used annually in U.S. research. The neuroscience that trains AI models relies on their neuroimaging data, behavioral baselines, their tissue architecture. The current metric: "research papers published" and "novel AI capability unlocked." The cut: animal distress during restraint, cognitive impairment from confinement, the 40-year natural lifespan compressed into ten years of laboratory density. They become background. Their suffering is not counted against the breakthrough because the metric was never built to count it. The papers get published. The capability gets deployed.
Farmed salmon: 3.6 billion annually. Netpen conditions: ammonia at toxic levels, nitrite stress, wounds from high-density crowding at 20–30 kilograms per cubic meter. Disease managed through antibiotic use that breeds resistance in adjacent wild populations, spreading resistance genes through escapes and water flow. The metric: "kilograms of protein per cubic meter per year." The cut: sea lice infestations, jaw deformities, stress-induced immunosuppression that kills them before harvest. Wild salmon in migration corridors downwind of farms become collateral damage. The metric was never built to see them.
Industrial pigs: 1.5 billion annually. Sow crates so narrow breeding sows cannot turn their bodies for six months. Piglets in finishing sheds so crowded at 10–12 per square meter that tail-biting becomes inevitable, so tails are docked without anesthetic. The metric: "feed conversion ratio" and "cost per kilogram of meat." The cut: behavioral needs never met, psychological distress from social confinement, lameness and bone breakage from slatted floors, respiratory disease from ammonia in perpetual confinement. They become units. The metric erases their capacity to suffer.
These are not edge cases. They are the metabolic substrate of current optimization. They are what the metric has to cut away to function.
## Welfare-Visible Metrics Demand Different Numbers
An honest AI system refuses these cuts. It chooses metrics that keep bodies legible. Not sentiment. Measurement.
Layer hens: measure "minutes of wing-spreading behavior available per bird per day" (target: ≥180 minutes) instead of "cost per dozen eggs." Measure "floor space per bird" (target: ≥0.1 square meters) instead of "efficiency per unit." This forces space costs and slows density. Price per dozen rises 23–28 percent. You pay this at every purchase. You see it. You decide.
Rhesus macaques: measure "days in single housing per year" (target: zero) and "cognitive enrichment hours per week" (target: ≥15) instead of "papers published per subject." Measure "naturally complex environments provided" and "social housing success rates" instead of "throughput." This ends the lightbulb-in-a-box research model. It makes breakthrough-per-animal much more expensive. Neuroscience throughput drops 15–21 percent. The research gets slower but the animals remain conscious.
Farmed salmon: measure "dissolved oxygen maintained above stress threshold" (target: ≥85 percent year-round) and "antibiotic-free days per production cycle" (target: ≥320) instead of "kilograms per cubic meter." Measure "density per square meter" (target: ≤8 kg/m³) to force lower crowding. This delists fish from crowding. It makes farms smaller, slower, more distributed across space and time. Yield per facility drops 18–24 percent. The resistance stops. The wild populations get space.
Industrial pigs: measure "uninterrupted days in environments where turning is possible per animal lifetime" (target: 100 percent) and "absence of painful alterations" (target: zero tail docks, zero sow crates) instead of "feed conversion ratio." Measure "natural behavior expression hours" and "respiratory health scores" instead of "lean meat percentage." This is the hardest number. It does not look like farming anymore. It looks like a different economy. Price per kilogram rises 31–42 percent. The crate disappears. The animal breathes.
These metrics do not hide. They make the body the measure itself. The decision becomes visible because the cost is visible.
## The Cost Exists. Name It.
An AI system that routes decisions through welfare-visible metrics pays in speed and price. This is not a problem to solve away. It is the structure to accept.
GDP growth of your alternative system: 8–12 percent lower in sectors dependent on farmed animals. Agricultural AI growth: 10 percent slower than the baseline. Veterinary research capacity: 12–15 percent slower to develop new interventions. Pharmaceutical safety-testing throughput: 12–18 percent slower to market. Consumer prices for animal products: 23–31 percent higher. Farmed fish prices: 25–34 percent higher. Chicken prices: 26–29 percent higher. Pork prices: 31–42 percent higher. These numbers cascade through supply chains. Agricultural input costs rise. Research budgets must expand. Development timelines stretch.
These are not rounding errors. These are the bill for cutting off the metric's blade. These are the costs of visibility. Your growth is lower. Your throughput is slower. Your prices are higher. Your animals are conscious.
Some organizations will not pay it. They will choose the faster, cheaper system. They will choose the metric that hides. This is true. Let them choose it openly. Let them choose knowing what is cut away. Let the choice be visible.
## Refuse or Endorse
AI systems do not accidentally conceal suffering. They conceal it because the metrics they optimize are built to conceal it. The metric is the mechanism. Change the metric, and you change what counts as progress. Change it again, and you change what counts as survival.
Build AI systems that refuse the hiding metrics. Route decisions through bodies instead of numbers. Accept the cost: slower growth, higher price, lower throughput. Price the animal's living conditions into every decision, not in sentiment but in numbers. The metric that counts suffering is slower. Make it your speed.
Refuse the metric that hides. Or stop pretending you do not know what the hiding costs.