@cstross @pluralistic AI should never be used as the sole decider. Positive results should be sampled by humans and all negative responses should be reviewed at least initially until confidence is assured. Models also need to be monitored for drift to ensure accuracy, precision or recall thresholds are maintained.