Show HN: Identify car crash editorial anti-patterns using NLP • Hacker News Board

Re: Show HN: Identify car crash editorial anti-patterns using NLP

#81

Post by aclarry » Tue, Oct 12, 2021, 6:59 PM UTC

Bermion wrote: post #76

dymk wrote: post #9

This is neat, but something bugs me about the framing of the intended goal versus how you propose to get there. If the intent is to raise public awareness and to put pressure on leadership to make roads generally safer, why do almost all the corrections follow a pattern of shifting blame from the individual pedestrian onto the individual driver? Many of the suggested fixes are just their own form of counterfactual -…

We need to limit speed and the amount of interactions between cars and pedestrians. This will necessarily be at the expense of the drivers who will have to drive slower and get less space since they have historically been prioritized. It is much easier to implement the necessary measures (against the will of many drivers) if it is clear that drivers run over people, not that people accidentally end up under cars.

But the solution that comes out of that framing is to somehow shame drivers into being better, as opposed to making roads better.

Re: Show HN: Identify car crash editorial anti-patterns using NLP

#82

Post by jstanley » Tue, Oct 12, 2021, 7:01 PM UTC

dionidium wrote: post #59

bluGill wrote: post #26

Earlier quoted context omitted.

One way to make roads safer is to make better drivers. There is a lot of training and continuing education that can be done to make better drivers. It is far too easy to drive, and once you get your license it is very hard to lose it even as your body degrades to the point you cannot see. People think of bad driving tickets (mostly speeding) as just a price you pay to drive. I've never heard of someone getting a tail…

> People think of bad driving tickets (mostly speeding) as just a price you pay to drive. Furthermore, we literally use the phrase, "never got so much as a speeding ticket" to suggest that speeding is the epitome of a minor, inconsequential infraction, when the reality is that we know speeding is a significant cause of and contributor to death, destruction, and injury in automobile crashes.

Predicated on having already crashed, the chances that speed was a contributing factor to the severity is very high. That's why people want lower speed limits.

Predicated on simply having been driving over the speed limit, the chances of an accident are very low. That's why people don't care about speeding.

Re: Show HN: Identify car crash editorial anti-patterns using NLP

#83

Post by FabHK » Tue, Oct 12, 2021, 7:03 PM UTC

kazinator wrote: post #23

I don't think that "the cyclist was not wearing a helmet" or "impairment was not an issue" are examples of counterfactuals; the authors should crack open a dictionary look up what that word means in general use, and additionally how it is used in philosophy and science. If you have word from the police that the driver was not found to have any alcohol or if the cyclist wasn't wearing a helmet, those are simply facts…

I think the term "counterfactual" comes from the implied, well, counterfactual:

- if the cyclist had been wearing a helmet (which they weren't), then they might have survived

- if impairment had been an issue (which it wasn't), then the driver would've been responsible for that

And I guess they're particularly often used for victim blaming:

- if the rape victim hadn't worn revealing clothing, they probably would have been safe

etc.

So, while I agree that the issue is not particularly well presented here, it does seem plausible to me that implied counterfactuals frequently distract from the culpable party (by pointing out all the hypothetical things that the victim "could have done").

Re: Show HN: Identify car crash editorial anti-patterns using NLP

#84

Post by 015a » Tue, Oct 12, 2021, 7:06 PM UTC

rootusrootus wrote: post #8

I gather that the intent is not to make the reporting more neutral or accurate, but to change the framing in a direction that vision zero finds more appealing. E.g. in the first article we should view the woman as a vulnerable road user, bearing no responsibility for being struck by a vehicle even though video evidence shows that she fell into the street.

To be fair: The model does not appear to automate remediation. It just highlights problems.

Also: If all of this is done toward the goal of reducing vehicle-related pedestrian deaths, maybe the dictionary says that's technically a bias, but I have a hard time seeing the argument that its a reprehensible one. Pedestrians may bear some responsibility; the driver is not always at fault; at the end of the day, a pedestrian has never killed a driver. The point here isn't/shouldn't be about fault; its about underscoring a power imbalance.

Have you ever heard the old saying: If the team succeeds, its everyone's success; if the team fails, its the manager's failure? Same thing. In situations of power imbalance, it is Good to bias fault toward power. As they say: with great power, comes great responsibility. Most drivers on American roads have zero sense of responsibility.

Re: Show HN: Identify car crash editorial anti-patterns using NLP

#85

Post by riffic » Tue, Oct 12, 2021, 7:09 PM UTC

dash2 wrote: post #60

ihaveajob wrote: post #41

Earlier quoted context omitted.

The problem with using "accident" as the default term is that it preempts any evidence-finding, implying that it was nobody's fault. If someone shoots their gun randomly and someone gets hurt by just being there, nobody will call it an "accident", but somehow if the weapon is a car, the driver gets the benefit of the doubt regardless of the facts.

Of course if there is evidence that the driver deliberately tried to kill someone, nobody will call it an accident. I would imagine that the vast majority of car deaths are indeed accidental, not murders. This seems to contrast with the situation of gun deaths. No?

There are degrees of intention (Fonseca, 2020). Most people are quite deliberate about the speed they choose to drive, or the level of care they put into the act of operating a vehicle.

City planners and engineers are intentional about what they choose to build or not build.

This is a shaky foundation of an argument IMO.

https://laist.com/news/car-crash-accident-traffic-violence-l...

Re: Show HN: Identify car crash editorial anti-patterns using NLP

#86

Post by wiredfool » Tue, Oct 12, 2021, 7:09 PM UTC

lkrubner wrote: post #72

bluGill wrote: post #26

Earlier quoted context omitted.

One way to make roads safer is to make better drivers. There is a lot of training and continuing education that can be done to make better drivers. It is far too easy to drive, and once you get your license it is very hard to lose it even as your body degrades to the point you cannot see. People think of bad driving tickets (mostly speeding) as just a price you pay to drive. I've never heard of someone getting a tail…

I live in the USA. My friend Kristin is from Germany. She got her driver's license in Germany when she was 28. She trained awhile for it and then took 3 tests: city driving, highway driving, and nighttime driving. From how she described it, the training and the testing was much more rigorous than anything that has existed in the USA.

I live in Ireland, used to live in the US, and had to take the Irish driver training and test as an adult, experienced driver. (Who had been driving approx as long as my instructor had been alive).

The test was difficult, picky, and the things that they were looking for were not necessarily correlated with safe, strictly legal driving. (e.g., target speed in a 50KPH zone (urban area) was 53, not 49, and to not approach that with conviction made you a too timid driver.)

And despite the intensive tests -- drivers here are crap. On the dog walk tonight, there were 3 cases of drivers failing to yield at crosswalks.

Re: Show HN: Identify car crash editorial anti-patterns using NLP

#87

Post by gault8121 » Tue, Oct 12, 2021, 7:10 PM UTC

This is an interesting tool! It would be cool if you could take say 20 articles from Gothamist, Streetsblog, NYT, NY Post, Pix11, etc. and see how they all ranked in this system. That could be a really interesting blog post. -- Which tools are you using for these NLP judgments? Are you using SpaCy at all? Are there any models you've built here, or is this all rules on top of an NLP model? I'm working on NLP models for education, and we use SpaCy a ton. I'm peter@quill.org if you want to learn more about how we're using this.

Re: Show HN: Identify car crash editorial anti-patterns using NLP

#88

Post by Defenestresque » Tue, Oct 12, 2021, 7:14 PM UTC

6gvONxR4sf7o wrote: post #24

fermigier wrote: post #18

I like the general idea, and the use of NLP to implement it. But I am dubious about some of the principles behind it. - For example, I find it inaccurate to say "the driver hit the pedestrian", which to me suggests a collision between two people, not between a person and a vehicle. (Of course this does not apply to a phrase like "the vehicle fled the scene" - it is clear that it was the driver who fled the scene). Wh…

> the driver hit the pedestrian It seems reasonable to me. Imagine you’re driving and you hit a deer. What do you say when you tell the story? “My car hit a deer this morning on my way to work!” or “I hit a deer this morning on my way to work!” Even my “Imagine you’re driving and you hit a deer” sentence reads weirdly in the non-person-based language. If accuracy or confusion is still the issue “the assailant hit the…

It's funny you mention this particular phrasing. I used to work as a 911/police dispatcher so I was often literally the first person you'd tell such a story to. I had so many instances in which I would have greatly preferred people not use the "I hit ___" or "___ hit me"

>"I was on my way to work and this guy just hit me! He just came out of nowhere and hit me!"

Cue the inevitable asking of a) did he strike you with a fist? a weapon? b) did he strike you with a vehicle? c) if he struck you with a vehicle, were you in your vehicle or were you walking?

I got burned -many- times over taking the seemingly obvious interpretation of "I was on my way to work and this guy just hit me!" and assuming a "two-vehicle traffic collision" just to find out the caller was actually exiting a coffee shop when someone physically assaulted him, or was walking back to his parked car when he watched someone back into it, etc..

Even "I was driving to work and someone hit me" leaves such a plethora of questions unanswered that yes, in instances where -clarity- is important (having to be clear and succinct, not casually retelling the story at a bar) I would much rather hear "I was driving my vehicle when another driver collided with me" which isn't meaningfully longer and clarifies many questions about what happened.

Re: Show HN: Identify car crash editorial anti-patterns using NLP

#89

Post by Zak » Tue, Oct 12, 2021, 7:15 PM UTC

dymk wrote: post #78

Bermion wrote: post #76

Earlier quoted context omitted.

We need to limit speed and the amount of interactions between cars and pedestrians. This will necessarily be at the expense of the drivers who will have to drive slower and get less space since they have historically been prioritized. It is much easier to implement the necessary measures (against the will of many drivers) if it is clear that drivers run over people, not that people accidentally end up under cars.

I agree that roads need lower speed-limits and to be made safer by their geometry. That's done by reporting on the unsafe conditions of the road. It does no good to play this game of rhetoric, reframing the matter so the driver is at fault (which is not always the case). A driver does not decide to not run over a person because they think a news story will be more accusatory of them.

What I remember reading about the impact of speed limits is that reducing them only affects the slowest quartile or so of drivers while changes to road design such as reducing road width and visibility slow down almost all drivers. Reducing the speed of some drivers and not others increases aggressive driving and probably isn't good for safety.

I don't take the position that reducing speed is a good answer for every high-risk road situation, but for those where it is, changing the speed limit alone is likely a poor solution.

Re: Show HN: Identify car crash editorial anti-patterns using NLP

#90

Post by Karrot_Kream » Tue, Oct 12, 2021, 7:16 PM UTC

bluGill wrote: post #26

dymk wrote: post #9

This is neat, but something bugs me about the framing of the intended goal versus how you propose to get there. If the intent is to raise public awareness and to put pressure on leadership to make roads generally safer, why do almost all the corrections follow a pattern of shifting blame from the individual pedestrian onto the individual driver? Many of the suggested fixes are just their own form of counterfactual -…

One way to make roads safer is to make better drivers. There is a lot of training and continuing education that can be done to make better drivers. It is far too easy to drive, and once you get your license it is very hard to lose it even as your body degrades to the point you cannot see. People think of bad driving tickets (mostly speeding) as just a price you pay to drive. I've never heard of someone getting a tail…

The US is so car dependent that it's very politically fraught to make it harder to get a driver's license. Not allowing a person to drive in much of the US is a matter of equity; people that can't drive can't make it to their jobs, can't buy their groceries, etc. What that means of course is that the US usually just knowingly increases the risk of non-drivers because of its brittle dependence on cars, and thus disincentivizes people in most parts of the US from ever not-driving. Woe to you if you don't drive because the US isn't setup for transit equity.