Upon receiving a corrupted message in 5G or 6G, a receiver generally rejects the message or ignores it entirely, because determining which message elements are faulted is difficult and complex. AI-based procedures are provided for localizing faults in specific message elements, and for determining the corrected values when possible. AI inputs may include the amplitude or phase modulation quality of each message element, the measured SNR of each message element, the modulation quality of a preceding demodulation reference, and current backgrounds, among other factors. After training (adjusting according to measured network data), the AI model may then determine the most likely faulted message elements, and may also direct the search for the most likely corrected values. By recovering the original corrected message without an unnecessary retransmission, the system can save time, reduce transmission energy, and avoid generating backgrounds. Many additional aspects are disclosed.