Hacker News Reader

Comment by jes5199 | Hacker News Reader

GPS routing is awful compared to a person who knows their way around a city.

Uber drivers in hilly cities like San Francisco never, ever think to avoid climbing right up and down the steepest hills - the algorithms don't tell them not to, there's no traffic and it's not a low speed limit, it just sucks for other reasons.

I had a driver last week who had never heard of Valencia St, and then misunderstood the GPS voice and accidently got on the freeway, getting us stuck in cross-bay traffic for 20 minutes just trying to get to the next exit. Of course I rated that ride 5 stars, because I'm not actually a psychopath ( http://www.forbes.com/sites/jeffbercovici/2014/08/14/what-ar... )

It's going to be a long, long time before we put anything approaching the depth of information that years of experience brings into our algorithmic routing systems. And startups are structurally disincentivized from trying to, anyway.

View on HN

In relation to your comment about the review: If anything less than 5 stars is unsatisfactory in Uber's eyes, why do they even have stars? If the rating is boolean anyway (5-stars or not-5-stars), then just put a single choice: Was this ride satisfactory? It shouldn't be my responsibility to know that rating a driver who was slightly late or slightly lost 4 stars is a Bad Thing. However, if you ask me whether that same ride was satisfactory or not, I'd probably say that it was. I don't necessarily want to punish people for one mistake, but I shouldn't need to rate them perfectly either.

I don't disagree but it seems to be the norm with P2P rating systems on the Internet. I haven't used eBay for ages--at least in its original flea market sense--either buying or selling. But there was definitely a sense back then that even giving a neutral (e.g. problem with shipping but seller resolved or something else along the lines of not ideal, but generally worked out) was giving a major black star. And people zealously guarded their perfect positive ratings--because it did make a difference.

B2C seems a bit different though you still see more 5s than deserved. Personally, I have no problem giving 4s for good to very good, but not truly outstanding.

One argument for the resolution is a perception as to whether the ride is improving or not - a binary indication of this will only tell you that you've gotten to the best ride, whereas a scale will, over the course of time, show improvement or detriment (subject to the different standards of the rider). Also, one person's "satisfactory" might not be another's, so having a range to select from will provide a much richer piece of feedback capable of being weighted than a simple yes/no.

A worthy argument, but one that falls a bit flat when the acceptable range of "passing" scores is .3 (4.7+) out of 5.

Hmmm... are you supposed to rate 5-stars for acceptable service? I've been rating 3 by default - then plus or minus depending on extraordinary things.

Yes, very much so. Have you seen the drivers' ratings? They usually hover around 4.8-5.0. If you're rated below 4.7 for both Lyft and Uber, I believe you're either on 'probation' or removed from their service.

You accidentally rating someone 3 for an average service can actually cost them their job. I'm hoping their algorithm uses some sort of a weighted average to offset the ratings of individuals such as yourself who consistently rate very low.

Well I didn't accidentally rate them a 3 - I picked a median score for standard service. I don't remember being told to do anything else when I signed up. If you rate 5 by default where do you go for truly special service?

Shouldn't it be the service's responsibility to determine what you are rating as a median score and then recalibrate your scores to their drivers?

i.e. If I always rate 3, but sometimes rate 1; then that might be similar to someone always rating 5 but sometimes rating 1-3.

> i.e. If I always rate 3, but sometimes rate 1; then that might be similar to someone always rating 5 but sometimes rating 1-3.

Ah, but then a user might throw an Internet Fit(tm) and cause a public outcry about ratings being manipulated by Uber etc. etc.

There's no way to easily verify how companies are actually calculatign the displayed ratings, and I doubt the average user would care or be savvy enough to find out. I assume it's a given most services see the input ratings as a personal indicator modulated by how reliable/benevolant/... each user is. Reddit votes, for example, are fairly nonlinear and visibility priority is even more complicated.

What does "truly special service" even mean for an Uber ride? Breaking traffic laws for you? Being especially friendly? Owning a very expensive car?

There are a lot of ways in which a ride can be unsatisfactory (smelly, creepy, crappy car, poor driving, loud music, etc.) but not that many in which it can be exceptional. Unless they show up in a Tesla Roadster or something.

Being especially friendly, accommodating and efficient in their driving, yes.

I just came across this in watching an enjoyable LISA presentation by Elizabeth Zwicky. At http://www.youtube.com/watch?feature=player_detailpage&v=C4r... is a story about customer feedback for a company which has, broadly speaking, engineers and admins.

Quoting from http://www.youtube.com/watch?feature=player_detailpage&v=C4r...:

"Engineers do not give out 5s. If you tell an engineer that the possible rating scale is 1 to 5 and they like you, they give you a 4. Because tomorrow they might love you and then they would give you a 5. They probably won't, but they might, so they give you a 4 so that they have room. We all are totally okay with this. But the world is also full of this other class of people who if they like you give you a 5. They're quite consistent about it. 'I like you. I see nothing wrong with that. I'm giving you a 5.'"

The graph (with, admittedly, artificial data meant to portray an example) at http://www.youtube.com/watch?feature=player_detailpage&v=C4r... shows it most clearly, with the satisfaction numbers from engineering clearly trending down. Apparently, engineers don't like to fill out forms, while admins do, so satisfaction survey results also have a sample bias in favor of the 5s of admin staff.

Its not just Uber, I often get "reminded" of an upcoming quality survey where anything less than perfect is equivalent to a tragedy.

I believe this is yet another case of managers using data without actually taking the time to understand the domain, or use it to identify problems and develop solutions. The approach seems to be nothing more than a facade of customer service.

In my mind, the question should be "was your service acceptable (yes/no)" followed by a comment box where you could leave anything worth noting. I guess you might have to sort comments via humans, but then you'd actually know what was going on as well. I suspect relatively few people leave comments as well, but what they did tell you would likely be a lot more than silly star ratings.

It is a trivial problem. Track how anyone with a smartphone and your app on it moves around the city. Aggregate which paths they choose and which turn out to be faster / better. Then propose those paths to other users asking for routing instructions.

I work on a software for routing deliveries (https://www.fleetnavi.com), and although we don't have this functionality built into the platform yet, it's the first thing to do when we get volume.

Waze claims to do something like and it produces the most retardedly slow routes ever. I just hate it when someone uses Waze on my Lyft because I just know I'll take 10 minutes more.

I don't think it's as trivial because there are lots of things to consider. Some drivers are just SLOW and it will have nothing to do with traffic. Maybe all slow/terrible drivers use your app so all the route speed information is completely based on specific patterns set by your app.

That is odd in my experience. Perhaps it has to do with my locale.

I am in a major metropolis and I drive throughout the city, suburbs and adjacent cities regularly. I know the area and I have been driving for decades.

I have been using Waze for about a year now. The navigation routes it selects appear to be very good. They appear to be based on relevant factors too. I can only guess what variables they include in their calculation, such as weighted capacity, current traffic speed, etc. of the relevant route elements.

My experience with Waze is the opposite of yours.

I'm amused that the phrase "uses Waze on my Lyft" is apparently self explanatory to this target audience. Clearly I am getting old and out of touch.

It helps that the nouns are capitalized.

Tracking, aggregating and analyzing thousands of users with slightly varied destinations, focuses (less traffic, faster, scenic), and signals is not exactly what I'd consider trivial...

Tracking, aggregating and analyzing thousands of packets with varied destinations, priorities and signals isn't trivial either, but it's something that some people and businesses have become very good at.

It's not a terribly hard job actually. I worked for a startup in 2004 that was doing this with haulage companies. We had 8000 data points moving around the UK coming in via GPRS and 3rd parties, traffic information aggregated and routing and mapping data, pick up corridors and event data.

It was (bar the web front end) a relatively modest sub-30kloc chunk of C++.

Now the scale is different but in 10 years, the problem isn't necessarily much more complicated, just larger.

In fact I think a derivative of it now runs some of Yodel's operations.

30kloc is not a trivial problem, that's a year's worth of programming.

I also love this:

the problem isn't necessarily much more complicated, just larger

If a programmer ever said this to me, it'd be a massive red flag that the estimate they're about to give me is going to be very wrong.

It's a larger data-set not a more complex problem.

Also, there is no need to get a perfect solution anything better than what's out there is useful.

> It's a larger data-set not a more complex problem.

Ummmm... yeah, scaling the kinds of algorithms that one might apply to this sort of problem up to larger data sets is often extremely difficult.

Well the dataset is about the same size now, perhaps 10% larger. The individual problem to solve is easily distributed across a large cluster of systems.

30kloc of COM/IDL as well that is. The actual routing engine was about 12kloc and the rest was the object model and deserializers. The routing engine was built on a two week binge of Red Bull, TAOCP and some papers printed out.

It really wasn't rocket science even though we charged them like it was.

From scratch by a non red bull enhanced team using c# instead it would probably about 3 months' work for two people to reproduce.

And now all your "fast" routes become slow because all your users are attempting to use them...

This is hardly a trivial problem. It might be trivial to deal with a simplified mathematical model of it, but throw people into the mix and it suddenly gets very complicated. http://en.wikipedia.org/wiki/Reflexivity_%28social_theory%29

Google appears to already take care of the 'fast routes become slow routes due to popularity' issue. Frequently in London, Google Maps will pop up a "Faster Route Detected" alert and will divert onto a quicker path. Sometimes three separate "Faster Route Detected" alerts will pop up during a single commute.

Google Maps would also change my London commute route on a daily basis given the changes in traffic patterns.

I have often thought the applications like Maps should randomise the suggested routes (within reason) to spread the traffic load around and improve things for everyone.

Does anybody know if Google/Waze/etc. do active A/B testing on routes with drivers?

> It is a trivial problem.

Aren't they all? Until you actually try to do it...

[In seriousness, we are excessively bad at estimating the true difficulty of problems, and making them sound 'easy' when they're not only serves to devalue the work of the people who actually spend the time and effort to actually figure it out.]

Well I drive around London a lot during the day time (inside CC zone) and Google Maps is pretty damn good on Android.

I did a few years in taxis but owning a car that is CC exempt is cheaper including parking in NCP holes. Part of the reason its cheaper is after some experience driving around the city, you realise that at least half the taxi drivers are taking the piss.

"the knowledge" is imminently replaceable by technology and is merely a party trick IMHO.

It's not completely replaceable by technology.

Once I had gotten a text from my wife as to where to meet her for dinner with friends. She didn't know the restaurant name, and had gotten the street name wrong. I just read the text to my London black cab driver, he asked me: "Did you mean XXX restaurant on [correct street name]?" (Note: there were multiple restaurants on the street). And sure enough, he was 100% correct.

That's the day I learned to really respect drivers with The Knowledge.

That's probably not "The Knowledge"; that's just some local experience and the fact that it's a high probability someone is going from that precise restaurant quite often. I myself have drunkenly hailed many a taxi outside dodgy restaurants in London. Unlicensed minicabs are pretty good at that too.

Conversely, got a taxi from Angel to Leyton and the idiot went via Poplar and the A13 and the journey cost me £55 for a £30 journey.

Maybe I wasn't clear... I was in a couple miles away at home and told this driver the wrong street (and no restaurant name) and he got me where I needed to go. You might put it down to "local experience" but I think it's better attributed to the knowledge; there's no way a minicab driver would have been able to do the same.

As you rightly point out, it doesn't mean that drivers won't still take the piss, but it definitely limits it. (Particularly when I compare it to virtually any other large city I've traveled to.)

The knowledge is not only the streets, they also have to memorise "points of interest" which includes things like pubs and restaurants.

But that's illogical. If you know the bar and restaurant scene in London, it's very volatile. Restaurants and bars change names at an incredible rate. "The knowledge" will have expired by the time you've learned it.

You must not have read the article.

There's an expectation that drivers keep their knowledge current. Certainly the examiners keep their knowledge current and test based on that.

I did. I really don't think you understand the scale of the problem.

It's certainly a large scale problem, but a smaller problem than learning all the points of interest in the first place ... no one is saying it's easy ...

Read the article. Updating the Knowledge is exactly what they do when businesses open and close.

Wow, you're really calling a course of study that usually takes 3-4 years to complete a "party trick" because it's replaceable by technology? By that reasoning, isn't any job any of us have on this forum also replaceable by sufficient technology?

Anyway, back on topic, "the knowledge" also covers alternative routes for road closures. I've usually been pretty impressed by black cabs. They know obscure little streets on the opposite side of town. Rarely have to give instructions. Compared to NYC where cabbies ask me how to get somewhere.

> Wow, you're really calling a course of study that usually takes 3-4 years to complete a "party trick" because it's replaceable by technology?

Yes of course!! Some other party tricks include reciting hours of lyric poetry from memory, adding large numbers in your head quickly, etc.

You're talking like the world is fair or something. There's not necessarily an equivalence between effort expended and value created.

Are there any skills you have, degrees you've completed,or knowledge you possess that you wouldn't also consider a party trick?

Sure. I write software that automates things, specifically natural language processing.

So in other words - party tricks :-)

> you realise that at least half the taxi drivers are taking the piss

This is really not the case with my London cab experience. These guys are super professional. In many other cities I've found cabs taking the scenic route but never with black cabs. I feel your 'half' claim extremely inflated off probably what was very few bad experiences.

San Francisco's hills are a very localized issue. Most cities are pretty flat. London included. Of course it's just a matter of time until GPS programs integrate topographical awareness, and give you the ability to avoid steep climbs.

Its also just a matter of time until cars provide visual aids for navigation, and not just the voice on a GPS device.

I think you're being a bit dismissive. Hills are just one example - every city has its local issues, be they hills, rivers, different laws, traffic patterns...

Rivers are irrelevant to a GPS. Rivers only matter to a car to the extent that they determine the road layout, and GPSs know the road layout. They're also pretty good about laws, although there is still some trickiness about things like left-turn prohibitions that are only in effect at certain times. It's getting better all the time, though. Their traffic data is in my experience better than human estimation at this point, since they can route based on actual current conditions rather than extrapolations from past experience.

You're making the common mistake of disregarding disruptive innovation[0] at its early stages. Just because it starts out as only of low quality[1] doesn't mean that it will stay that way, or for long.

0: http://en.wikipedia.org/wiki/Disruptive_innovation 1: http://upload.wikimedia.org/wikipedia/commons/8/8e/Disruptiv...

You're making the common mistake of regarding high quality technology as inevitable. Just because it starts out as only of low quality[0] doesn't mean that it will improve, or fast.

Google Maps directions is pretty disruptive tech, right? Add 10 years of experience, lots of roads, lots of time to gather data, real-time traffic info, and when I ask it for directions from my current location in inner London to a place in central London[1] it still gives driving directions first, even though I switch to transit, walking, or cycling everytime.

But hey, innovation.

0: http://en.wikipedia.org/wiki/Clueless 1: http://i.imgur.com/c7XAuUm.png

Setting aside your snide tone[0], the mode of transit shown by default in Google Maps has absolutely nothing to do with quality.

0: http://en.wikipedia.org/wiki/Arsehole

> the mode of transit shown by default in Google Maps has absolutely nothing to do with quality.

Of course not. Technology will eat the world and improve on and replace humans in all complex cases. Like deciding what mode of transport to use in central London. Because disruption.

>Of course I rated that ride 5 stars, because I'm not actually a psychopath

You are doing everyone except the driver a disservice. By failing to give low ratings to bad drivers, you are precluding good drivers from entering the market (because the bad ones will never get phased out, and they will saturate the market). You are also preventing consumers from making a well-informed decision, which limits the effectiveness of market mechanisms.

That's a very insightful article you linked. I had always wondered why Airbnb reviews were so uniformly high, even when not merited. I'd found some explanations involving Airbnb hiding negative reviews, but never quite connected the fact that I personally felt uncomfortable leaving negative reviews with the fact that other people generally didn't.

Mapkin is doing this actually.

They let users crowd-source their own directions, along with much better turn-by-turn descriptions too. Eg, "turn sharp right just after the firehouse, go past the field and make a left just before the baseball field". Instead of "turn right, keep left and turn left".

https://mapkin.co

This is one of those weird cultural things that us non North Americans cannot understand, like automatically tipping. Here we don't tip unless someone is absolutely exceptional (and tips get shared amongst staff in restaurants anyway), and we don't give five stars unless you really deserve it.

That's really interesting. I wonder if there's scope and technology for a collaborative constraint-based solver that finds more effective routing for cities.

Side note: it drives me nuts that that article mixes up "insures" and "ensures." Shouldn't an editor at someplace like Forbes catch that?

> GPS routing is awful compared to a person who knows their way around a city.

Strongly disagreed. GPS has traffic information, a human driver does not.

A human that drives the same roads everyday most definitely does have traffic information.

I always use GPS even on the roads I use almost daily because the traffic information my GPS provides (Google Maps) works much better than me guessing what the traffic will be. Even if I know that traffic will be slow in the morning on certain roads, I cannot predict accidents and construction work.

I've been in San Francisco for 3 years, and I still check traffic information every weekend. Community events, which are fairly common, can make the difference between "traffic will be fine" and "don't dare leave your apartment."

Perhaps less anecdotally, local traffic is a major staple of local news radio and TV, which probably means it's important even for locals.

Not the kind that matter though: Is there an accident & associated traffic jam today on road X or not.

A cab driver will get that information too.

No he won't. This guy just spent 4 YEARS memorizing maps...his honour is not going to allow him to rock a GPS and any radio'd info isn't going to include the same second-by-second route computations that the GPS does.

But a licensed cab driver can just as well use traffic information provided by a GPS device.

I always enter in my final destination when requesting a pickup. Avoids any confusion. Seriously?!

How'd you get an Uber driver that's never heard of Valencia St?

Happens all the time, especially when there are big events on. I had a driver a few weeks ago who had come down from Sacramento to work the weekend in SF for such an event (Dreamforce, I think) and was relying solely on GPS.

Reply on news.ycombinator.com