Hacker News Reader

Comment by Someone1234 | Hacker News Reader

Great article, unfortunate choice of title and opening image.

A lot of people are going to skip reading it, and assume this is an article blaming the pilots for this accident, rather than pointing out that neither the airline nor pilots were informed about Boeing's new systems and so couldn't have been trained to counteract it.

Boeing's new system wasn't even in the updated manual. And therefore it wasn't in their QRH, which is likely what they were using while attempting to resolve the issue.

> With only fragmentary data available, Mr. Hansman said he suspects that a runaway of the M.C.A.S. system played a central role in the crash. “The system basically overrode the pilot in that situation,” Mr. Hansman said.

Every thread so far about this topic however have managed to find a way to blame Lion Air, or simply point out how much they suck while glossing over the actual facts.

View on HN

Replacing/overriding humans by shoddy AI will become bigger and bigger problem. From the Gmail auto-suggestions to the labyrinth of voice recognition options when you call your credit card or an airline this is all a part of an unfortunate trend.

Honestly if the aircraft just communicated what automated actions it was taking better, it may not have been so fatal.

"STALL AVOID - STABILIZER ADJUSTED - STALL AVOID - STABILIZER ADJUSTED" would be kinda hard to ignore. Even if they forgot to put it in the manual.

Then on most modern aircrafts, these alarms would be going off all the time. It's more important to know why the computer received incorrect data on the angle of the plane.

Usually there are three or more sensors reading it and the master computer uses the voting mechanism to arrive at the correct reading. So in this case probably the master computer had erred?

I don't understand. Do most modern aircraft fly right at the edge of their ability to stay in the air?

Not in climb, cruise, or descent. Only on maximum performance takeoffs and the last moments before touchdown are modern airliners near the edge of the envelope of controlled flight.

Engine "Stalling" is not a big issue at all. As a part of training we purposefully stall engines and recover from it. And as angle of attack increases due to increase/decrease in wind speed, engine stalls can be more frequent that usual.

You are conflating two separate things. Stalls in the aerodynamic sense (angle-of-attack above critical) have nothing to do with the engine "stalling" (engine-out) which is a mechanical failure.

I should have used better terminology. The article here describes aircraft stalls and I was talking about the same.

Aircraft stalls are not extra-ordinary events and with auto pilot on the computer is always adjusting the angle, speed , preventing stall etc.

Can we please stop calling this AI!? This is well within the established control theory. Nothing about flight controls is "AI".

While we're at it, can we stop using AI buzzword? Nothing about current state of machine learning is AI.

Voice recognition used to be AI, then we developed it.

A roomba used to be AI, then we built one.

Neural nets used to be AI, then we figured out how to make them practical.

Is a self-driving car AI? Probably not, since we have prototypes now.

Nothing about our current state of anything will ever be AI because AI just means "computers doing things computers can't do". As soon as we understand how to make a computer do something, it's no longer AI.

Please someone correct me if I'm wrong; but that is not what "we" mean with A.I. Neural net is actually one. It is the ability "to learn themselves through self-feedback". I don't think we have it covered down, as in we don't fully understand the outcome. It gives great results, like pretty generated pictures. And we do know the process, but the outcome is still unpredictable.

>"to learn themselves through self-feedback"

But now you're describing machine learning, which many people would describe as "applied statistics, not AI".

>And we do know the process, but the outcome is still unpredictable.

Unpredictable because you don't understand the statistics or unpredictable because of the size of the computation being done? Neither of those indicate "AI". Or maybe unpredictable because a relatively simple algorithm unexpectedly gives rise to good results? Then we're back to the roomba that just randomly moves until it bumps into something being AI.

There's no good definition for AI. It's like superfood. You can argue that it means high in antioxidants or high in fiber or probiotic, but at the end of the day, it's the marketing departments that determine the definition.

I think it's more "things we assume thought process is needed to manage". Because we use thought processes when we do them.

This is also where the disconnect is coming from. We have self-driving cars now, so (an alarmingly good chunk of) people assume the car has thought processes. While in reality the car is more like blind walking than having the experience you and I have while driving a car. Certainly it's not thinking any thoughts.

"Any sufficiently advanced technology is indistinguishable from AI."

I'm sure people thought computers were "AI" when they first did a complicated multiplication problem.

I don't think there is a qualitative difference between what's going on in a human brain and a self driving car.

Then how will we have unicorns or find employment for marketing mbas?

The article blames the new system in the very first paragraph, before it mentions anything else. And in the third paragraph it explains how Boeing mishandled its installation. Only then does the article mention pilots - and when they are mentioned, they are excused, rather than blamed.

So the evidence does not support your worries.

Reply on news.ycombinator.com