Hacker News Reader

Comment by CivBase | Hacker News Reader

> A malfunctioning sensor is believed to have tricked the Lion Air plane’s computers into thinking it needed to automatically bring the nose down to avoid a stall.

That is ridiculous logic to implement in a "safety" system. An automated system should never cause a plane to dive unless it also knows that it has enough altitude to safely do so - much less in a way that makes it difficult for pilots to override.

View on HN

Given the system assumed it was in stall, which means loss of altitude anyway, surely it's safer in general to go nose-down to avoid the stall? At least then you have a chance of recovery, which you don't in a stall. (Except of course, going nose-down.)

If you don't have enough altitude to afford going nose-down, let the pilot handle it. If your system can't come up with a safe solution, do not override the pilot's controls in favor of a dangerous solution.

A computer should never assume it knows better than a pilot. A computer is only as good as the data it gets and the software it runs. Sensors fail. Data gets corrupted. In the current state of the industry, software bugs are inevitable.

Airplane software is supposed to help pilots, not hinder them. In light of these events, I'm thinking twice about wanting a self-driving car in the near future.

Elsewhere I've read MCAS does take altitude into account, as well as flaps, i.e. it's only active above a certain altitude, and only when flaps are retracted. So... yeah, we don't have the full story. And also in another thread, it's reported from the flight prior to Lion Air 610 (same plane) there were airspeed and altitude disagreements. I'm not at all clear from available reporting whether airspeed, altitude, and angle of attack were inconsistent, if that was a source of either autopilot confusion, and then pilot confusion, whether pilots did set stabilizer trim to cutoff and when and whether it was too late.

I'm a pilot (former CFII) and the whole automation fail danger strikes me as terrible. John Q Public says "I want the automation to override the pilot's mistakes!" What? OK fine. What about Asiana Airlines Flight 214 where the pilot advanced throttles, an explicit intent input, and yet autothrottles were set so the automation said nope. And then John Q Public are all, well the pilot should have KNOWN!

It's like it's a game where the pilot is only there as the last resort to be blamed if they too fail, even after a sequence of automation failures. Automation betraying pilots at low altitude is in my view functionally equivalent to an in-flight breakup. And automation in the cockpit mentality in the face of failures has been, for 20+years, "add another button, add another feature, add another routine" to tack on all the others.

And yes this absolutely makes me think of autonomous driving as total b.s. Airplanes are in a standardized system, with far bigger budgets for automation and yet we still have to fall back to human pilots for routine procedures like parking, taxiing, VFR approaches and landings, and telling the automation literally every detail it needs to do, ATC communication. It's ripe for end to end automation and yet we still don't do that. Driving cars is wildly more complicated for automation: non-standard streets, paint, signage, laws, pedestrian behavior, bicycles, cars still driven by humans, weather - haha. Sounds nice, great idea, keep trying, but it's complete bullshit.

Reply on news.ycombinator.com