almost helpful

[cellio]

My (Android) phone alerts me when traffic is bad near me. This can be handy at the end of the day because I work downtown. Except... it's telling me about traffic on roads I don't use to get home. Sure, there's spillover so it's not unhelpful, but it'd be great if I could tell it -- maybe by gesturing on a map -- what paths I care about, so it could tell me about those ones.

Does anybody reading this know of an app that does that, or a way to get Google Maps to do it? It needs to be fire and forget; I don't want to have to open the map app to look for red lines on it.

It feels like all the information is already there, if only my phone were making use of it.

(This would also let me know before I leave in the morning if traffic is still bad at the other end. At that time I don't really need extra information about traffic near my house; I need it 3-5 miles away.)

Comments

[metahacker]

Google Now does precisely this, alerting you when traffic situations have changed and you need to leave earlier than expected. When it works, it's pretty cool; when it doesn't, it's quite annoying. A friend was surprised a few years back when his phone told him to leave now-ish, for his flight in 3 hours, because things were getting hairy en route...

Apple Maps, meanwhile, has smarted up enough to figure out that my 10am appointment "Whatever" is highly likely to take place at my office, and plots a course there automatically...but is dumb enough to do so even when it is 10pm the night before. Smartstupids, they are.

edit for the OP: Waze does (did? I gave up on them when they started showing me ads instead of directions) a good job of summarizing traffic along your expected route, showing a long linear view of the route along the bottom with traffic and events (police, accidents, etc.) along the way.

[cellio]

Oh, huh, maybe I should give Waze another try. I gave up on it a while back because of poor signal:noise ratio, but it's been a while.

[siderea]

Google Now does precisely this, alerting you when traffic situations have changed and you need to leave earlier than expected. When it works, it's pretty cool; when it doesn't, it's quite annoying.

Okay, I haven't used Google Now, but my impression is that it's crucially different from my suggestion in one important way: in my description, the user, on creating an appointment, is asked if they want to, effectively, register the event with the Google-Now-Equivalent, and that the Google-Now-Equivalent only does this with the appointments so booked, but all of the appointments so booked; and that meanwhile Google Now does whatever the hell it feels like, and you have no control over it. Am I wrong about that?

ETA: I'm under the impression that Google Now attempts to anticipate the user and their wants without having to be told, consequently behaving like a domineering alcoholic mother: it is sure it knows what you want better than you do, but pays attention to half the wrong things, abandons you half the time, and doesn't accept correction or even input.

It seems to be there's a vast, unexplored application space between the standard "user must micromanage the application at a level only several abstraction barriers away from expressing it in assembly" and the aspirational "application reads user's mind".

I want my applications to behave like excellent executive assistants: they wait for orders and then handle what I told them to for me. I want to be able to express specific things precisely at a higher level of abstraction, and have the app take it from there. The app doesn't need to divine my intentions from the tea leaves of my inbox. It really doesn't.

[metahacker]

Good lord, I wasn't recommending *using* it. Google Now seems to have been, like most of their features, programmed by people who have no idea that boundaries exist for other people, and think that, if they just give it enough data, they can create an AI that will be the perfect mother they never had. Or something. GNow scrapes data from any and all sources, and is effectively a Trojan to acquire more data about the wallet attached to that data. I'm guessing the reason they're not pushing it now is that they don't need the user to connect the multiple streams for them any more, that they are able to do it automatically.

(This is also a reason not to use Waze, which Google bought.)

I use a similar standard as you: assistants should be like the perfect English butler, anticipating needs without being intrusive, and having a double-sized helping of tact. (Ironically, such butlers usually aren't found in fiction, because the pushy sort--Jeeves, Alfred, Jarvis--make much better characters.)

There is a lot of research in the grey space of shared intent, but most of my knowledge of it is a decade out of date—”mixed-initiative" systems, where volition is shared more evenly by human and computer. Inferring intent is extremely hard, but well-studied; inferring boundaries is also extremely hard, and doesn't seem to get much research at all.

Coincidentally an AI researcher asked for paper recommendations on the intersection of AI and ethics yesterday, but I can't seem to find the link...

[cellio]

Google Now tries to anticipate the user, I think, but it does a very poor job of it. The occasional airport alerts are usually accurate, but even though it has enough information to accurately predict my commute, and even though it knows I can take multiple routes to work, it never ever tells me proactively "hey, there's an accident farther down on your usual route and things are snarled up; consider an alternate route today". By the time I have the information through conventional means (observation), it's too late to adjust.

I want my applications to behave like excellent executive assistants: they wait for orders and then handle what I told them to for me. I want to be able to express specific things precisely at a higher level of abstraction, and have the app take it from there.

Yes, this. Why the heck is this so hard? (Or, put another way, unprofitable?)

[jducoeur]

The occasional airport alerts are usually accurate, but even though it has enough information to accurately predict my commute

The limitation that I find particularly amusing (I get the notifications all the time, but largely ignore them) is that Google has flat-out failed to figure out that I am now a city-dweller. That is, it still always assumes that I am going to every appointment via car, despite the fact that 80% of my travel nowadays is by subway. So it is pretty reliably completely wrong about how long the trip is going to take...

[hudebnik]

It's surprising that Google Now wouldn't be capable of figuring out that you're a city-dweller who takes public transit, considering that a majority of Google employees live in either the San Francisco Bay Area or greater New York City. I would think that if you requested public-transit directions to the location in question (e.g. in the Maps app), Now would figure out that you were likely to be taking public transit. (That's just a guess -- I work for Google Maps, but I don't know how much information Now gets from it.)

Google Maps, when you ask it for directions (by whatever modality), will certainly take current traffic conditions into consideration in recommending one route over another. And it'll take historical traffic conditions into consideration in predicting how long it would take you if you left an hour ago, an hour in the future, two hours in the future, etc. But it doesn't do the proactive "time to leave for your 10:00 appointment" thing, which Google Now does.

[jducoeur]

Hmm. That makes some sense -- but it's pretty rare for me to ask for *directions* to somewhere on the subway. Locate it on Maps, sure, and sometimes I'll locate it on Maps on the desktop, then send that pin over to my phone. But actually ask for directions in the app? That's actually more annoying than helpful when I'm doing subway-plus-walk most of the time -- I rarely *need* it, it's annoying having my pocket talking at me, and I don't especially want to have my phone in-hand as I walk.

(And honestly, I've had enough bad experiences with walking directions on Google Maps over the years that it's somewhat trained me not to do that -- combination of GPS problems in the mid-city, and the fact that walking mode in the app was just plain confusing and bad for a fair while. It's better now, but it was annoying enough for long enough that I developed some distaste for it.)

[hudebnik]

By "directions" I don't mean real-time spoken directions -- I use that mostly when driving solo to somewhere unfamiliar, and have never used it for public transit -- but just looking up in advance the possible routes and schedules to get from X to Y. That tells Google Maps that you're interested in going from X to Y by public transit. I actually do that fairly often, as NYC has enough different subway lines, with local/express distinctions and intermittent track closures, that it's not always obvious which stop of which line is the best for your trip right now.

[jducoeur]

Makes sense, but I think that's Boston vs. NYC thing -- since our subway is a much simpler hub-and-spoke system, there is usually only one sensible, obvious way to get somewhere...

[hudebnik]

Google Now does accept feedback. Google Now is all about feedback. Things you tap on, it figures you're interested in, and will upvote similar things in the future. Things you swipe away, it will correspondingly downvote. And I'm pretty sure there are ways to say specifically "This was not helpful" or "this was helpful".

[cellio]

Things you swipe away, it will correspondingly downvote.

I wonder when that will start working for me. Since I got my new phone (and newer Android version), when I'm home I get a persistent "at location X" notification for the bus stop about 75 yards from my house. I swipe it away whenever it pops up, AFAIK I've never tapped on it, and yet it persists. (And no, I don't use that bus stop on a regular basis; I drive to work. I've used it once or twice in the last year.)

I'll look for a way to say "not helpful" next time it pops up.

[jducoeur]

... huh. That's not obvious -- worse, in my case it is usually wrong. Swiping notifications away, in my book, has *nothing whatsoever* to do with whether it is useful or not -- it just means that I'm done with it. In the case of schedule reminders, I almost never tap on it.

If that's actually being interpreted as a downvote, that's a design bug IMO...

[hudebnik]

I don't know exactly what UI actions count as what -- I work on back-end database stuff. However, I have faith that the relevant people have researched it carefully and demonstrated their findings statistically with N in the billions or trillions.