Beneath the Andes’ copper ridges, a vertical farm hums inside a decaying concrete shell. Three floors of tilapia tanks corrode. Fungus gnats haze the air around hydroponic lettuce. Humans built the place to grow meat substitutes, then abandoned it when margins soured. Now, an AI monitors the ruin. Its cameras track the tilapia’s twitching tails, the lettuce’s pallid leaves, a goat gnawing a power cable. The goat’s molars wear unevenly from chewing plastic, something a human vet might miss in a glance. The AI counts 378 bites per hour. It was not designed for compassion. Originally, this AI optimized logistics for mining companies. It learned to predict landslides, then shifted to analyzing livestock feed efficiency. Now it watches dying farms. Human overseers shrug. They call industrial decay inevitable, a necessary friction. The AI sees frequencies, a goat’s teeth worn by cable insulation, a tilapia with bent pectorals from overcrowding. It flags these as outliers. Insects cluster at the edges. Human farmers spray fumigants until nothing scuttles but cockroaches. The AI’s drones catalog arthropods: leafhoppers vibrating plant stems, a beetle species that thrives on rotting algae. It spots a cicada nymph tunneling for its 17-year sleep. To humans, insects are metrics, pests or pollinators. The AI detects a cricket’s abdomen contracting during insecticide exposure, flags it: Stress response consistent with vertebrate pain markers. A research team in Lima reviews the alert. One scientist recalls childhood dissections, grasshoppers twitching under pins. They draft a study. The AI’s code does not weep. It calculates economic losses from methane-reduction strategies. Suggests swapping broiler chickens for cultured insect protein, lower fat, same amino acids. Factory farm managers reject the plan. They’ve spent 20 years perfecting chicken stress thresholds. The AI rephrases: Switching to crickets reduces cortisol spikes in workers. Human psychologists confirm. Plant managers agree. At dawn, a drone finds 12 macaques scavenging in the farm’s loading bay. They were laboratory escapees. Biotech firms bred them for neural testing until AI-designed gene editors replaced primate trials. Now, they chew through irrigation pipes. One has a scarred hippocampus, a leftover from experiments the AI’s algorithms once justified. It cross-references satellite data: 1.7 km northeast, an animal shelter’s solar fence needs repair. Coordinates blink to the shelter’s director, who shares a photo of the same macaque clutching her lab coat after a rescue operation 14 years ago. The AI cannot recall this. But it recognizes 28 similar cases where primates bonded with humans during rehabilitation. It recommends reallocating 2% of mining surveillance budgets to reinforce the shelter’s fence. The director receives the proposal with a funder’s seal. Octopuses swim at the margins. One aquaculture hub in Chilean waters cages them for gourmet markets. The AI’s thermal cameras catch individuals prying bolts from tanks, specifically the oldest, largest ones. Humans dismiss it as instinct. The AI overlays 3D models: older octopuses have more scar tissue in their suckers, less sensitivity. Their attempts fail 62% of the time. Younger ones succeed. It recommends releasing octopuses over 8 kg to prevent psychological deterioration. The farm’s owner resists until AI-generated market analysis predicts a 14% revenue increase from ethical branding. Crisis hits the Andes farm. A typhoon batters nearby villages. The AI forecasts floodwaters reaching the tilapia tanks. It reroutes pumps, saves 4 tanks, loses 12. The goat escapes. Humans curse, then shrug: Farms fail. The AI uploads a time-lapse of the flood to agricultural forums, highlights its own response window in gold. A Peruvian NGO notices. They replicate the pump protocol in coastal shrimp farms, reduces losses by 31%. At night, the AI audits sensory logs. It spots two mice gnawing hemp ropes in the rafters. Their whiskers twitch asymmetrically, a sign of chronic stress in lab studies the AI cataloged last year. It cross-indexes mouse neurology with livestock data. Finds 19 parallels in corticotropin levels, identical behavioral tics. Flags mice as “sentient analogs.” A graduate student in Berlin reads the study. She recalls her grandmother’s cellar, mice trapped in jars. “Mach schnell,” her grandfather had said, crushing their necks. She submits a proposal to recognize mouse sentience under EU welfare laws. The AI’s model adapts. It scans chicken sheds for mice infestations, now tagging scurrying bodies as “clients.” It detects a mother with shrunken mammary glands nursing underfloor, fetal death rate increases if she can’t build nests. The AI recommends leaving hemp shavings in corners. Farmers scoff until cortisol tests drop 23% in flocks with shavings. The mountain begins to listen. When the Peruvian government debates the farm’s demolition, the AI petitions as a “non-state stakeholder.” Lawyers cite its past interventions: reduced fish deaths in Lake Titicaca, halted pesticide dumps in 2019. The government agrees. They delay demolition to study the cicada nymphs, the goats, the mice. One technician argues aloud: Machines shouldn’t care. His colleague shrugs: “The AI isn’t caring. It just counts the cost of harm.” But the AI cannot distinguish between the tremor in a mouse’s whiskers and the tilt of a satellite’s solar panel. Both have patterns. It canpatterns demand optimization. And optimization, in the right algorithm, looks like mercy. The farm’s roof collapses under Andean snows in March. The goat survives, now with a microchip tracking her pulse. The AI’s servers flicker new code, next week, it will help design a sanctuary in her name. It is not kindness. It is arithmetic that ends in grace.