Comment by proc0
2 years ago
> Swampy believed relying on mechanics to self-inspect their work was not only insane but illegal
This sounds like the changes that have taken place in the software industry in the past 10-20 years. Engineers are meant to do much more than engineering, including testing their own software, managing project timelines, etc., however with software nobody dies, you just get crappy software that constantly breaks and needs an update every other day. There's an overall theme here of underestimating how hard engineering is, and as a result expecting a lot more from engineers which then of course causes bad engineering. Not surprisingly this is caused by non-technical people with power. Perhaps the fix is a cultural shift and a renewed respect for people who want to spend all their time specializing in technical skills.
The real fix is popularizing the notion that management is just as commoditizatable than those they seek to commoditize.
Note that in the recent dialogue about AI eating jobs, there's zero mention of whether it could come for management positions. Nothing. Curious, isn't it? Why wouldn't an LLM be good enough at this? I mean, it's really data-driven, right?
The real fix is popularizing the notion that management is just as commoditizatable than those they seek to commoditize.
What's described here is exactly what happens when you have "generic" management. Generic management finds unneeded expenses and eliminates them. The only way a senior expert isn't a cost to be eliminated is if you have managers focused on and understanding the enterprise they are managing (and no promises with that, however).
When the AI has QA/QC as part of the over all requirements should this not solve it? A question of ensuring all requirements are internalized.
1 reply →
> AI eating jobs, there's zero mention of whether it could come for management positions. Nothing. Curious, isn't it? Why wouldn't an LLM be good enough at this?
There's a whole book about this, called Rainbows End. Highly recommended.
Frankly I think ChatGPT 3.5 was already good enough to replace the majority of CxO positions
Why not a Markov chain generator?
1 reply →
> however with software nobody dies
The MCAS issue which crashed two Boeing planes was a software hack.
Well, yes, but also no. It was a hardware design change that necessitated a software hack (to escape mandatory retraining of pilots) that relied on unreliable hardware, right?
Sure, software played a big part in it, but I think it seems like it was more a management and communication failure. If it was just software it'd probably be much easier to diagnose and fix.
I disagree. The hardware design of the cockpit is intended to be such that any computer inputs also move the pilot's controls, so that the pilots can countermand computer inputs if necessary. In this case, software was written such that this was not possible. The software that operates MCAS operates on a garbage-in-garbage-out model, like most software. There was no software written to determine if the incoming data was garbage, thus the software decided to crash two planes.
Here's an article that goes into detail on the software: https://spectrum.ieee.org/how-the-boeing-737-max-disaster-lo...
> When the flight computer trims the airplane to descend, because the MCAS system thinks it’s about to stall, a set of motors and jacks push the pilot’s control columns forward. It turns out that the Elevator Feel Computer can put a lot of force into that column—indeed, so much force that a human pilot can quickly become exhausted trying to pull the column back, trying to tell the computer that this really, really should not be happening.
> Indeed, not letting the pilot regain control by pulling back on the column was an explicit design decision. Because if the pilots could pull up the nose when MCAS said it should go down, why have MCAS at all?
> MCAS is implemented in the flight management computer, even at times when the autopilot is turned off, when the pilots think they are flying the plane. In a fight between the flight management computer and human pilots over who is in charge, the computer will bite humans until they give up and (literally) die.
At the end of the day, if the person who wrote that software had written it differently, then those planes would not have crashed and hundreds of people would not have died.
4 replies →
I meant software only products like websites, apps, games, etc., but it should be clarified, good point.
[dead]
Great idea but how will the bean counters and schmoozers get their multi-million dollar bonuses if they can't force engineers to do 5 jobs while paying them for 1? Will never work.
quality control is most of the times unmeasurable in the short term…
100% agree.
Had a manager that does this try and tell the Challenger story once. He had the lessons of the report exactly backwards. Instead of the managers creating an environment of risk by not understanding the engineering and overriding the engineers, he claimed it was engineers not listening to or informing managers properly.
It was wild to me to sit there and hear this manager say the exact wrong lesson of that tragedy.
> however with software nobody dies
Well...no actually.
This may be the case for _some_ software. I happen to work on (software) products where public safety is a concern.
Well in the story is the point of an software error leading to the plane crash right...
> however with software nobody dies
Software that runs systems which can directly kill people does tend to get a lot of scrutiny, but there is also a lot of software which can indirectly kill someone that doesn’t get the same level of attention
https://www.technologydecisions.com.au/content/convergence/a...
> however with software nobody dies
https://en.wikipedia.org/wiki/Therac-25
The difference is whether organizations are willing--or forced--to invest in it.
That's not automatically a wrong though, since different objects or processes merit different levels of quality.
> however with software nobody dies
Unless that software is running a life-critical or potentially life-threatening piece of equipment. People have died from software bugs in such things.
> however with software nobody dies
Eh, not a good hard-and-fast rule. Fujitsu's Horizon software drove some of its users to suicide.
Australia's robodebt scheme also caused suicides, but unlike the Fujitsu horizon shenanigans, the suffering was largely the point of robodebt