If drivers make deadly mistakes at X rate, and software makes deadly mistakes at Y rate where Y < X, should we recall the software? What if Y is substantially less than X but still not zero?
The complication with that is that software makes different mistakes than drivers do. And sometimes, software refuses to drive, in which case, their accident rates are subject to selection bias. For cars which can be operated by software or a driver, you’d want to compare for an exact equivalent scenario.
The problem is Y is more uniform but X is not. Good drivers make significantly fewer mistakes than the average. For these people, using software potentially increases risk by order of magnitudes.