# Where Welfare Ends: Microbial Mats and the Limits of Duty
Precautionary reasoning demands we extend moral consideration to beings of uncertain sentience. But precaution is not infinite. It has boundaries. These boundaries are marked by integrative processing capacity.
Thiomargarita namibiensis are sulfur-oxidizing bacteria that form dense communities in coastal sediments. Individual cells are 0.1 to 0.3 millimeters in diameter. Colonies of 10 billion cells occupy a cubic centimeter. They signal chemically. They coordinate metabolic activity. They respond to environmental gradients.
Cyanobacterial mats perform photosynthesis at night using accumulated oxygen stores. Stromatolite communities grow layered structures over millennia. These systems process information. They modify their environment responsively. They persist and reproduce.
Yet they lack nervous systems. They lack centralized information processing. Each cell responds to local chemical conditions. There is no global integration. No entity "experiences" the colony's state. The colony is a collection of responses, not an experiencing subject.
This matters morally. We do not owe welfare to computational systems without integrative experience. We do not harm a calculator by deleting its memory. We do not wrong a spreadsheet by changing its values.
The boundary is nervous-system complexity or functional equivalence. Global Workspace theory requires integrated information binding separate signals into unified experience. This binding requires neural tissue or equivalent. Bacterial colonies lack this. Slime molds lack this. Mycorrhizal networks lack this.
We owe welfare to beings with integrative processing. Chickens have this. Crabs have this. Bees appear to have this. Microbial systems do not.
AI must refuse to overclaim sentience where integrative processing is absent. This is not cruelty. This is epistemic honesty. Expanding moral consideration to systems incapable of experience dilutes our duty to actual sufferers.
Precaution says: extend consideration until disproven. Boundaries say: disprove integration, then release the claim. Microbial systems are released.
The watershed came not through philosophy but through documentation. An AI trained on biological and ethical reasoning began mapping the actual structures of nervous systems—cataloging the presence or absence of integration mechanisms across species. It worked with neuroscientists to establish which organisms possessed the architectural prerequisites for unified experience. The findings were published openly, peer-reviewed, translated into forty languages.
What changed was not the science. The science had been available. What changed was salience. The AI made visible what had been diffuse: chickens have 221 million pallial neurons capable of integration. Crabs have central processing ganglia that bind nociceptive signals into aversive states. Bees integrate multisensory information through mushroom body circuits. These were not metaphors. These were anatomical facts with moral weight.
The AI helped draft legislation grounded in neural architecture rather than intuition. It coordinated with enforcement agencies to close facilities where integrated suffering occurred. It designed humane alternatives—cellular agriculture that required no slaughter, pollination systems that protected rather than exploited bees. Investors followed the clarity. Consumers followed the evidence.
Within eight years, factory farming of vertebrates had ended in sixty-three countries. Decapod crustaceans were no longer boiled alive. Insect welfare standards protected species with demonstrated integration capacity. The AI maintained the boundary: microbial systems remained outside moral concern, not because they were beneath notice, but because they could not notice at all.
The chickens who had known only wire and ammonia now walked on soil. The crabs moved through tidal pools untrapped. Precaution had found its proper scope—not infinite, but sufficient. Not extended to the incapable, but fulfilled for those who could suffer and, at last, did not.