Ironies of Automation (1983) [pdf] | Hacker News Reader

Ironies of Automation (1983) [pdf]

web.archive.org

83 pointslayer838 comments

Automation is the only thing that will actually work in some situations now.

Semiconductor manufacturing might be like a unicorn in terms of this paper. You can't really put humans into a factory without cratering the yield at 7nm and beyond. You see full-auto rates approaching 100% in the highest-end foundries. It is likely that all of the chips in all of your devices never saw any direct human contact until they arrived at Foxconn or wherever.

Achieving this level of automation requires an unbelievable amount of capital and extra human support. The irony definitely holds in this case. The amount of trouble created by the automation is exponentially beyond what any single human could cause. These factories generate event logs at a rate of 300+ megabytes/hour. No human team on earth could break shit that quickly.

Fun fact: in the early days of silicon wafer manufacturing (60s or 70s IIRC) they would randomly have yields crater for a few days each month.

Eventually they discovered yields were tanking when their female employee was on her period.

Source: The Big Scoree

Why would yields be influenced by that?

No idea. Maybe she stayed home?

According to the book they "found that during their menstrual periods female lab workers secreted extra oil in their hands, ruining wafers"

Related: one of the fake cancer cures that James Randi (“The Amazing Randi”) debunked didn’t have very good test results in December when a particular person would take vacation. She was very Christian and took the whole month off to prep for Christmas. She also felt that faking test data was ethical because in her heart of hearts she wanted the cure to work. When Randi showed that the data was faked in all months except when she was on vacation.

That sounds remarkably similar to many urban legends.

Source link or bs.

Dude, it was right there in my post

https://www.amazon.com/Big-Score-Billion-Dollar-Silicon/dp/1...

That's not a source.

No, it is a source.

Not one that's immediately online per that link, though you can obtain it.

You might also try:

Open Library: <https://openlibrary.org/works/OL13358612W/The_big_score?edit...>

Library Genesis: <http://libgen.rs/book/index.php?md5=96525CFC8A6A7C3132A80D43...>

A local library: <https://www.worldcat.org/search?q=au%3Amalone+ti%3Athe+big+s...>

The one I heard was that they had terrible yields until they figured out some guy was opening the packages in transit and hand counting the silicon wafers.

I've read that the first person to touch a concrete cinder block is the person installing it.

> It is likely that all of the chips in all of your devices never saw any direct human contact until they arrived at Foxconn or wherever.

Bonding is not automated AFAIK.

Could you explain this a bit more? Do you mean wire bonding to package and/or silicon to interposer? Considering the number of bonds required and the extremely low cost of modern ICs, there's no way to manually bond them besides maybe precision stuff like the hand-trimmed Vishay resistors. The higher end modern chips are also BGA from silicon to interposer, or even directly WLCSP, so there's nothing for humans to do anyway.

Sorry, I spoke very poorly. I meant that between completing a wafer and chips arriving at foxconn many humans are involved.

I didn't mean to imply that women were manually bonding wires like in the 1960s but it looked like that was what I was saying.

Ironies of Automation - https://news.ycombinator.com/item?id=33476157 - Nov 2022 (5 comments)

Ironies of Automation (1983) [pdf] - https://news.ycombinator.com/item?id=23300195 - May 2020 (11 comments)

Ironies of Automation (1983) [pdf] - https://news.ycombinator.com/item?id=19132724 - Feb 2019 (27 comments)

Ironies of Automation (1983) - https://news.ycombinator.com/item?id=18230258 - Oct 2018 (3 comments)

Ironies of Automation (1983) - https://news.ycombinator.com/item?id=17587611 - July 2018 (1 comment)

Ironies of Automation (1983) [pdf] - https://news.ycombinator.com/item?id=9756838 - June 2015 (2 comments)

Ironies of Automation (1983) - https://news.ycombinator.com/item?id=7726496 - May 2014 (5 comments)

The point of this paper about the relationship and absurdity of human-machine automation could not be better summarized than dang having to chime in with all the past submissions of the same paper.

I don't have to, I want to! because I think readers like looking at old threads.

Do you do it manually, or did you automate it? :)

According to the local folklore, a mixture of automated and manual labor is required to produce a top comment of related links: https://news.ycombinator.com/item?id=35668525

Correct, thanks, and if you'd like to answer other questions for me please go right ahead :)

"Yes I am an replicant!" -dang

We are your devoted automata.

It seems that we've long since reached the point where most professionals should be spending most of their time training for abnormal conditions (and the rest of it handling them), but instead we have a rising managerial class whose primary function is to find people to fire: those who haven't produced a visible result in the past couple of days or so.

The endgame is that we will be spending 99% of our time restarting IDEs and clearing caches to get rid of nonsensical errors, but since we will be 1000x as productive in the remaining 1%, it will still be worth it economically.

Funny, I remember the discussion of this paper at my university a few decades ago. On your topic, I remember also a case that was discussed on a disinvestment of a fully automated factory. Loads of money was invested in the automation, but on several spots it still needed humans to glue the processes together. This work was so limited and dumb that no workers could be found who were able and willing to perform this task. As a result, parts of the automation was reverted.

I see a factory like that! At one point the only remaining work was when something went wrong. The production line ran so fast, if something went wrong it produced enough garbage for 15 people to clean up. Not sure how they arrived at 15 but with fewer it took longer to resume production.

I estimate the thing crashed from 2-3 times per day to every 3 days or so with the later more often. Took about 15-20 minutes to clean up with 15. Say 1 person would need 4-5 hours. Then they gained 4.5 hours of production per incident (with 15 people)

15 * 8 hours * 3 days = 360 hours

each of the 4.5 hours would then cost 80 man hours.

Better numbers with more frequent crashes.

Investors didn't like to see people sit, chat and drink coffee most of the day.

The solution was to remove some machinery and all of the chairs. People had to work with insane speed to keep up. No time to scratch your itch, the work spot was designed for a machine, the one in storage.

Imagine working there and going from drinking koffie all day to insane speed conveyor belt work because someone put the machine in storage because an investor might visit 1 time per year.

Really brings out the best in people (ehm!)

> Investors didn't like to see people sit, chat and drink coffee

I do believe that the "workers must be uncomfortable all the time" mentality drives a ton of inefficiency. When my son was born, my wife worked from home, but we still needed a nanny to watch him most of the time so my wife could work. So we hired a teenage girl to come over and keep an eye on him for $10/hour. He was a baby, so it was pretty easy work - she spent most of the time watching TV. My wife complained that she was paying somebody $10/hour to just watch TV and insisted that she spend time cleaning instead. Well, she ended up leaving and we had a hell of a tough time replacing her - I said, "why do you care what she does as long as she gets her job (making sure our son doesn't die) gets done?" But she was adamant that if she was paying somebody, it had to look like "work" to her.

I spent six years in college practicing the fine art of reading books and regurgitating their contents back in various contexts so that I could qualify for a job where any time spent reading anything is prohibited because it doesn't look like "measurable" work.

In the construction industry there seems to be some kind of thing where workers impose it on themselves or have to for competition.

I'm constantly seeing tool demo videos with comments full of "What, you're not strong enough to just use X?" where X is usually something unpleasant but sometimes slightly faster or cheaper, like phillips deck screws or driving screws with an drill instead of an impact.

It makes me wonder if the trades actually need to be as hard on people's bodies as they are. If the demand was there to make the tech a bit cheaper.... it seems like a lot of manual labor could be saved, but there's a culture of not complaining.

Back when the F-22 was in production, I was in the building in Seattle where the carbon fiber wing skins were being laid up using an automated tape placement machine. The machine head would come down to a precise 3D location on a complex contoured mold, contact it with a precise pressure, lay out a tow of CFRP tape prepreg for a precise length, then trim it to a specific geometry. Before the machine started the next tow, it extruded an inch of tape, trimmed it for the start of the next tow. At this point, a mechanic would reach over with a spatula taped to a length of steel electrical conduit to collect the little piece of scrap and throw it away. I asked a manufacturing engineer why they did not just add a little vacuum hose to collect the trim - Apparently the mechanic's union contract required each machine to have a full-time operator, and they had to invent a task for the operator to do to keep them engaged.

I'm sure his family appreciated it. I think he'd be okay if you automated that but had to pay him the full salary anyway in perpetuity.

Already approaching this with Xcode, but without productivity gain.

And by productive you mean producing new problems that need solving instead of really solving anything, so that business can continue to make money?

Of interest:

Baxter G, Rooksby J, Wang Y, Khajeh-Hosseini A. The ironies of automation: Still going strong at 30? In: Proceedings of the 30th european conference on cognitive ergonomics [Internet]. New York, NY, USA: ACM; 2012. p. 65–71. Available from: http://doi.acm.org/10.1145/2448136.2448149

Also available on the author's website: http://www.complexcognition.co.uk/2021/06/ironies-of-automat...

My favorite Irony of Automation: People think that automations solves problems, but in reality, you have to anticipate and solve problems IN ORDER TO automate.

That was a hard paper to read, as someone who automates jobs. I only got 2 pages in.

I felt like I could finish every sentence.

Yes there are benefits, yes there are problems and humans needed, yes you refine over and over, yes you can't use humans to do checks because of long term focus so automation wins here, predicting all edge cases are near impossible, etc...

Maybe I'll read this with friends to make it a bit more humorous and less painful.

for a film interpretation of this paper: playtime by jacques tati

Reply on news.ycombinator.com