It retains no memory of being overridden by pilot input but will keep pitching down until cut-out of the loop or the aircraft exits what it thinks is the dangerous corner of the envelope. Unfortunately it often lacks the information to make that determination correctly, due to the poorly-designed sensor integration.
That sounds like something right out of the 'Catastrophic Failures' books we studied in uni.
This ultimately is not a training issue. Why was a system that can potentially plow the aircraft's nose into the ground utilizing only a single sensor without any redundancy in place? Boeing screwed up. Now both sensors will be used. Great, but sometimes 2 sensors fail. If pilots that operate these aircrafts are trusted with the lives of 150+ passengers, they should have an "easy button" to take complete manual control of the aircraft. My human sensors should trump any mechanical sensor(s) when my life is on the line.
Activating those switches cuts off electric trim and would completely deactivate MCAS. It can be done in seconds.
Are there other feedback mechanisms present that can interrupt that determination?
That appears to be the case in these scenarios. The MCAS doesn't perform the incorrect action continuously, it looks like initial corrective measures fix the problem until they suddenly don't.
And then what? How much longer will it take to manually move the stabilizer to a suitable position? How much time do the pilots have at under 7,000 ft AGL? Will the elevator have to be moved nose down to unload the stab before the stab can be retrimmed?
That's great...if they'd understood that the system was malfunctioning...but they had no such indication because their airline didn't pay for the optional "feature."
Place a random passenger into the cockpit of a modern commercial airliner without training and ask them to fly the thing safely and you'd probably have a 100% loss rate.