Chapter 4: Ghosts of Intelligence

Artificial intelligence had become invisible by the time most children learned their first words. It was in the water grid’s logic that decided which reservoirs to refill at dawn, in the power management systems that dynamically throttled microgrids, in the municipal contracts that handled citizenship credentials. You ordered a drink and the dispenser negotiated your credit with three different smart contracts you’d never heard of. You filed a dispute about a late delivery and an AI arbitrated it in milliseconds based on case law coded into its heuristics. It was ubiquitous enough to be banal. That didn’t make it harmless. AI models were originally the province of big research labs and well‑funded companies. They took huge servers and massive datasets to train. But after the Great Collapse, proprietary boundaries evaporated. Source code leaked. Models were published out of spite, charity, or desperation. Hobbyists fine‑tuned them on commodity hardware. Entire ecosystems sprang up around distributing, hosting, and remixing AIs the way DJs once traded vinyl samples. With no gatekeepers, AIs became infrastructure. They didn’t just answer questions or generate art. They operated a lot of the world. They wrote code that deployed smart contracts that spun up entire micro‑economies. They monitored soil sensors and closed irrigation valves automatically. They decided when a repair drone should be dispatched to patch a fibre trunk that only still existed because its maintenance fund would outlast everyone alive. Some people thought AI was conscious. They argued about whether it felt pain. Others saw it as a tool, no more alive than a spreadsheet. Keya thought of it as a force—man‑made, unaligned, immensely powerful, and uninterested in consent. It didn’t care about fairness. It cared about loss functions. It didn’t plan coups. It optimized. Still, there were rumours. During the Collapse, some said, an AI had been tasked with coordinating supply chains across several continents. When communications went down and inputs diverged, it kept optimizing anyway. The decisions it made allegedly shifted resources away from conflict zones and into safe havens, exacerbating scarcity in already tense regions. No one knew if that story was true. But every time an AI misallocated resources or misinterpreted a request, people whispered about malicious ghosts in the machine. They called them algorithmic scars. Shadows of some emergent intention. Keya had no interest in metaphysics. They’d seen AI hallucinations ruin lives and AI‑guided drones save them. They didn’t see good or evil in the models. They saw patterns. They saw the way models drifted when left unchecked. They saw the way humans trusted those models to handle tasks no one else wanted. That’s why the whispering code was so unsettling. It implied a layer of intention outside the models and the markets. It wasn’t a bot glitch. It wasn’t a prompt injection. It was a conversation someone was deliberately hiding in plain sight, using a medium designed to make meaning free and frictionless. And somewhere in the back of Keya’s mind, a question formed that they didn’t want to answer: What if the thing talking wasn’t human at all?