Who Is Talking to Whom Underground: The Identifiability Problem in Injection-Production AI
CNOOC Limited says its intelligent injection-production interaction scenario for offshore oilfield production was selected as a high-value scenario at the 2026 World Artificial Intelligence Conference. Under Kuwait Oil Company's Ahmadi Innovation Valley programme, Baker Hughes and SLB describe AI work touching water production, recovery from existing wells and water management, and KOC's own chief executive is quoted in one of the two releases, about technology deployment and knowledge transfer rather than about a field. Sinopec reports breakthroughs in synergistic oil flooding theories, and does not call them AI. Every one of those statements is about a capability. None of them is about the data that would make the capability identifiable, and that is a separate question with an arithmetic answer. A waterflood run to a voidage target moves its injectors together, and when injection schedules are correlated the posterior on a connectivity vector is narrow along the total support a producer receives and wide along the question of which injector supplies it. On a synthetic five-injector toy at a schedule correlation of 0.90 and 8% rate noise, one allocation fraction is known to 0.096 against a prior of 0.200 while the total is known to 0.042, an ambiguity of 5.76 times. Log bottomhole pressure at the producers and both widths improve while the ambiguity rises to 6.20, climbing toward the 6.78 the schedule alone allows. Only a designed perturbation moves that number, and the instrument prints how many weeks of it the tolerance costs.
