Hacker News Reader

> Why don't we trust the machines? Is it because they have bugs and are sometimes actually wrong? Or because we trust our feelings more?

I can't recommend enough to read the book "Command and Control: Nuclear Weapons, the Damascus Accident, and the Illusion of Safety" by Eric Schlosser. You will be surprised how often we barely escaped nuclear war because wise people chose to not trust the machines.

What's amusing is that I don't think anyone trusts complex systems less than software developers.

You'll never find a developer on casino slot machines.

Because of knowledge of probability or software quality? Casino software has a lot of review and regulation because the customers have money at stake.

I'd stop it at "because of knowledge"

You should have stopped it earlier. There is such a thing as advantage slots, and many software developers are drawn to advantage play. Because of knowledge.

News to me. Developers are universally considered fish.

https://xkcd.com/2030/

Okay, but is there something particularly wise about getting an alert that a boarding pass has already been used and presuming it must be a glitch? I would only call that wise if the operator in fact had knowledge about how the system works and could explain why that alert was likely to be a glitch.

> Okay, but is there something particularly wise about getting an alert that a boarding pass has already been used and presuming it must be a glitch?

Did the gate agent do any follow up verification? Like checking the flagged passenger's ID?

In retrospect, modern computer systems are inscrutable and unreliable enough (as a category) that it's rational to disregard them, especially if you've done some independent verification. Rational doesn't always mean correct in a specific case.

The ID checking the second time probably would have worked since that was presumably the real person. They don’t normally collect ID at boarding according to the article so they wouldn’t be able to see the previous person’s ID.

Exactly. I don't see how strictly trusting the computer would have lead to a better outcome:

1. Computer says no, leading to the true passenger being rejected. Probably would have resulted in the guy managing to steal her seat and maybe not getting caught.

2. Computer says no, overreact, and reconcile everything until the computer is satisfied (e.g. deplane everyone and start over, maybe checking IDs). I'd guess this attitude would result in hundreds or thousands of delayed flights when everything was fine.

Sure, but it says he made sure to board before the correct passenger… it didn’t alarm for him because he was the first. When the second passenger came through, they would ask them for ID and it would be correct because it was actually them.

It's a great book and very terrifying!

> It's interesting that we use machines, and when they flag the type of error they were designed to identify, we simply assume they don't work.

The question is what's the incidence rate of real glitches. If the system errs on 1% of boarding passes, staff will be used to overriding it.

>If the system errs on 1% of boarding passes

On a large plane that would be 2-4 people per flight.

Honestly an error rate of even .01% would be enough errors per day that your average gate agent would order ride pretty commonly.

In this specific case, it's quite possible that the gate agent, being a costumer service worker, didn't want a fuss. If someone comes to a gate agent with a boarding pass, they're most likely not double-spending the pass, and suggesting otherwise to them would be a quick way to a tweet saying "today I bought a plane ticket, and when I tried boarding I was told I already boarded and they refused to reimburse me, do not fly [airline]". Of course it would have been the "correct" course of action here but as someone who has interacted with The Public, the gate agent did their best.

Beyond that, the UX of this is always subverted by incentives.

The system's incentive: (1) make sure no pass is used to board more than once, (2) indicate whether all ticketed passengers have boarded, and if not which are missing

The gate agent's incentive: (1) board everyone as pleasantly and quickly as possible

So the errant boarder played on the misaligned incentives and time pressure, to bet that the gate agent would ignore any errors.

I'm incredibly distrustful of any system that requires:

   - Someone deal with a rare situation
   - In realtime
   - Under pressure
   - While not being incentivized against an adverse outcome

"Well, the person didn't do what they were supposed to in the 1-in-1,000,000 situation" isn't a valid design strategy. (See self-driving-car-interventions and TSA agents)

Another interesting question is: What were they supposed to do? Does Delta even have procedures for what to do in this 1-in-1,000,000 situation?

If the adverse outcome of a 1-in-1000000 situation was somebody possibly sneaking onto a plane and gets a free ride, I'd say just leave it as-is.

Incentive implies a choice can be made.

No one tips a lawnmower for a really tight trim. The lawnmower manufacturer may have worried about that but the lawnmower itself doesn't even worry.

Likewise, the dumb software the agents are using is not making decisions or incentivized.

Now if it were AI, we might worry . . .

They are being incentivized against an adverse outcome: the outcome of a paying passenger (or a whole planeload of paying passengers) missing their flight, connections, etc. That's why they make the decisions they do.

Accidentally carrying a non-paying passenger on an underbooked flight is also adverse I guess but it's not the only way the system can err and it's not the only thing they are trying to avoid.

With this comment you've demonstrated why voting machines will never be trusted fully; lack of understanding.

You understand paper and pen. It's right there. A child fully understands the dynamic.

As soon as the process moves into a machine, you cannot even tell if your checkmark is really counted.

This is doubly worked into distrust, by the amount of bugs, weird isses, and strange behaviour software exhibits to end users. People are trained from youth, via personal experience, how incredibly buggy and untrustworthy computers are.

They crash, can lose data, be wrong, and the worst of it? The layman cannot even figure out why. Computers are a black box, a mystery, they function by magic.

No expert, no claims of audits, or oversight, will ever, ever refute what end users know. Computers make mistakes all the time.

Taking these impressions and incorporating them into the votng system is madness for democracy.

Not just those impressions, but those realities.

The trust model for paper ballots and observed human counting is so superior to any proposed or implemented electronic voting that it's madness to downgrade to electronic.

I agree completely, but find that my argument is easier to explain to literally everyone, where as describing trust models is harder to explain to many, and invites debate from others.

I prefer "mixed" > "paper" > "electronic", but a few year ago we had elections with only "paper" in our national soccer association:

From https://en.wikipedia.org/wiki/Argentine_Football_Association

> With 75 presidents of different Argentine clubs voting, on election day something went wrong when the final count resulted in a draw of 38 to 38 (76 votes in total). The explanation given was that one of the electors put a double vote and that mistake was not reported. As a result, the executive committee decided to postpone the election.

As with anything, paper voting must be done correctly. There are literally hundreds of years of tweaking, bugfixing if you will, of process and method around paper voting.

Proper paper voting cannot be gamed by ballot stuffing. Voting paper has serial numbers, is guarded like a currency during creation, and like currency during transfer, has checks and bounds at both ends, is counted by representatives of all parties, the list is endless.

For example in Canada you vote, fold the form shut from prying eyes, take it to the lady with the ballot box (usually a retired person), and she detaches the two parts of the card, one she keeps, the other you deposit into the sealed ballot box under her watchful eye.

There is no chance to stuff anything extra, again members of all political parties are present, and any trickery is made doubly difficult by unknown serial number ranges at the voting station. A serial number you are assigned when your voting card is taken off of the randomized stack.

I assure you any "but... what about..." has been thought of, and any security hole you perceive is likely just me missing facts.

Here the ballots are printed in normal paper. Each party prints their own, and the goverment send them to the polling stations. But other parties may steal the ballots from the voting booth, so each party must print additional ballots to replace them, just in case.

There is a plan to make a single ballot printed by the goverment, but the local county chiefs don't like it becuase it avoids a lot of dirty tricks. (The elections are quite fair anyway. The petty cheating just changes very little globaly. I don't agree with cheating, but I can't fix it.)

Some parties distribute their ballot before the election to avoid the missing ballots problem. Some local county chief ever distribute their own local ballot with the national ballot of the opposing party if people want to vote a combination. Some people distribute old ballots to that are invalid to void the votes of confused persons of the other party. In some places, the rabbit hole is deeper than you think.

I don't know what to say, exceptn such rampant fraud and miscreantic activity is the problem.

And that has nothing to do with paper ballots.

Here, you'd end up in jail for those things, and trickery wouldn't get you out of it. And you'd get caught too, because members of your own party would turn you in.

Sadly, I have a higher threshold for "rampant".

No design is resistant to incompetent implementation.

I'm not sure what happened, they never released an official explanation. (It may have been just cheating, but my conspiracy theory is that it may have been a truce between the two factions to get a few more days to negociate.)

https://xkcd.com/2030/

> You understand paper and pen. It's right there.

You think you do, but then you look into the 2000 Bush-Gore election. We'll never know who actually won because of hanging chads.

https://en.wikipedia.org/wiki/2000_United_States_presidentia...

It could happen with any analog system. If it's fill-in-the-bubble, how do you count underfils? Or one candidate fully filled and the opponent slightly filled? When you get 10M pen and paper samples, you'll get sampling errors, and in a close election, there won't be a clear winner.

No need to hypothesize about how it would work. Australia uses pen and paper for elections. It works fine.

And with more complex ballots with preferential numbering.

(Aside: it’s actually pencil and paper since pencils are more reliable. https://www.aec.gov.au/faqs/polling-place.htm#pencils)

> (Aside: it’s actually pencil and paper since pencils are more reliable. https://www.aec.gov.au/faqs/polling-place.htm#pencils)

In some ways, but the ability to erase a pencil mark is a huge issue with ballots.

IMHO, pens with tamper resistant ink, plus a process for making an exchange for a new ballot paper in case of a marking error, is a much better choice.

Canada uses pencil too. Erasing ballots and remarking them is never an issue, why would it be?

The security is in how ballots are treated, how they are monitored and handled, not how they are marked. It's never a problem that they are in pencil. Ever.

If someone has an opportunity to monkey with the ballots? The system has already failed.

As I said in another post, if you think you found a hole, it's because you didn't research what's actuallly happening.

Yes, hanging chads created by a machine, punching answers through perforated hole cutouts. Perforated holes, which started to break and fall out more and more as the ballots were counted.

The ballots were fragile, and this is an example of not using pen and paper.

You're actually proving my point here. Keep it simple! Pen and paper is perfectly fine, and all problems with how to handle the marking has been resolved for centuries.

It's change that is the problem here, and change for no sensible reason.

There's nothing to fix, to improve, to resolve with paper ballots. Nothing, except the pocket books of companies pushing ridiculous solutions.

>> You understand paper and pen. It's right there.

> You think you do, but then you look into the 2000 Bush-Gore election. We'll never know who actually won because of hanging chads.

Punch cards are not "paper and pen."

> It could happen with any analog system. If it's fill-in-the-bubble, how do you count underfils? Or one candidate fully filled and the opponent slightly filled? When you get 10M pen and paper samples, you'll get sampling errors, and in a close election, there won't be a clear winner.

A paper and pen system may not be perfect, but regular people can understand it fully including the failure modes and exceptions.

No so with computer systems, they can be so totally inscrutable. What if instead of hanging chads, it was a buffer overflow that corrupted 500 votes before they could be tabulated? CNN could have experts talk about it nonstop for a week, and all most people would get out of it is that computers are inscrutable and unreliable.

Washington State uses fill-in-the-bubble and that seems to work fine. We vote by mail so voters even provide their own pens!

It works fine until there's a high-profile, close election. Then you'll discover the data source was noisier than you thought. We're only talking about one in a thousand people goofing up a vote and moving on.

We do all elections in the mail. We frequently have votes come down to a few hundred ballots without issue. They are all counted and tracked.

Just heard about this:

https://www.mercurynews.com/2024/04/25/evan-low-up-by-one-vo...

> Some of those challenged ballots in San Mateo County had illegible or no postmark dates and Assistant Chief Elections Officer Jim Irizarry said they were waiting on information from the U.S. Postal Service

With small margins, an election could come down to how people read a smudged postmark. It might not be an issue, but you won't know who "really" won, either. It depends how you count it.

The other reason I don't want to use machines for voting is that it's not that much work to count them manually. We've been doing it well before computers existed. Government has never been interested in efficiency before, so why suddenly care about it with the most important component of a democracy?

My guess it had happened a few times in the past and had always been "glitches" or passenger doing something silly which also gets blamed as a "glitch".

... or the stowaway has found an empty seat (not likely nowadays).

1) Give the ticket to the first person in the check point, get scaned, talk to the their children that is makin fuzz, give the ticket to the second person in the check point, alarm, nobody remember that they gave the ticket to the first person because everyone was in autopilot.

2) Give the ticket to the first person in the check point, get scaned, notice that the partner went to the bathroom, wait, give the ticket again to the first person in the check point, alarm, nobody remember that they gave the ticket to the first person because everyone was in autopilot.

I bet passengers do this often enough.

You're traveling as a family. You print boarding passes and distribute them to everybody. Somebody loses theirs. So you print new copies, but you don't necessarily give people the same seat number as the first time.

When boarding the plane, one of you reaches into their bag and grabs the one they thought they had lost. So two of your family members are using the same pass, and one is going unused.

No harm is caused because the number of passengers on the plane is still as expected, and the family is able to figure out for themselves who takes which seat.

I was just recently traveling with someone and they had a copy of both of our boarding passes (I just had mine). I boarded first and heard behind me some issue happening: they'd accidentally given my pass so it popped up as already boarded. They just switched to the right pass and there was no issue. It would have been a little more complicated if we had just happened to board in the opposite order, though.

So, I'm sure the general issue of "this pass was used already" happens a lot. How often it gets resolved without the would-be passenger presenting a "good" one, who knows.

Well the passenger actually _hadn’t_ boarded. Someone else had. Their checks found the extra passenger afterwards.

The mistake was letting the original passenger on and that was fixed later.

I would disagree with that. The boarding ID had been used once and can't (shouldn't) be used twice. In that case the first use was in error but the system can't know that. So the correct course of action would be to re-check everyone's boarding pass. That may have been impractical / undesirable, but simply ignoring the error as a bug seems careless.

Maybe if flight attendants were actually getting paid while passengers are boarding they might be more attentive

I don't think their checks caught him though? He only got caught because the plane was full, and he stood out like a sore thumb when he couldn't sit down somewhere.

If boarded still means "on the vehicle," both people boarded.

One person was a passenger and the other was a stowaway.

This is because we don't have a plausible scenario for that case. "Sir, you already boarded the plane, please leave the airport" is not going to work. Calling the police, making a fuss with getting all the people off the plane and then back on is a lot of damage for the company in late flights and missing connections.

> Why don't we trust the machines?

We don't trust the machines because some of us aren't outsourcing their critical thinking faculties. If this person assumed it was a computer glitch that probably means that it happened before and they guessed, probably correctly, that glitches happen more often than people trying to reuse boarding passes

This is an individual in a specific situation. I can think of a few reasons off the top of my head why they might have done what they did.

Most importantly, there might not even be a standard process to follow here - remember this passenger was properly ticketed. There previously was a glitch. The agent probably didn't want to deal with angry passengers (there was a fight between passengers and a Delta agent in the past couple years).

It's true the system may have glitched in the past, but IMO the bigger issue is that the system provides no troubleshooting steps. What if it snapped a photo of every passenger as it scanned their ticket. Then on a duplicate scan it could display the photo of the passenger who already scanned the ticket. That could help the gate agent to actually troubleshoot the warning and figure out what happened.

Making the system more obviously complicated is no way to convince people it's more trustworthy.