Why does Siri seem so dumb?(recode.net) |
Why does Siri seem so dumb?(recode.net) |
>At the Chinese drive in & Then.. (ProTip: Cooking recipes line by line for a win)
Absolutely this. I mean, you try a few of this these "long tail" queries and you eventually say "Fuck it!" I attempt these "long tail" queries (which to me, that term sounds like some shitty play-it-down excuse from Apple) weekly just to see if Apple is finally getting their act together in regards to Siri usefulness. I am consistently disappointed and never surprised and delighted.
I try to screenshot "Siri fails". Here are the last few in my screenshot album:
-------------------
"Open last screenshot" → "You don't seem to have an app named 'last screenshot' We could see if the App Store has it [App Store]"
"Share this with my wife" (Photo was open in Photos.app) → "I'm sorry Joe, I'm afraid I can't do that"
"When is the last time I exercised" → "Interesting Question, Joe"
"What year was this song recorded" (In Music app with song playing) → "Interesting question, Joe"
-------------------
We're living in a world where we can do more and more without using the screen as an interface. It's happening. I'm worried that Siri is a low priority product for Apple right now and they will soon be scrambling to play catch up if they aren't already.
EDIT: I'm hoping that Apple's recent buddying up with IBM allows for some of Watson's intellect to seep into Siri. Who knows, maybe in a future keynote we'll hear Tim say, "Ladies and gentlemen, Watson is coming to your Mac and iPhone!"
I believe there is a difference, but that's purely speculation on my park. I'll be able to test it more in ~2 weeks when my Pixel XL arrives.
Tech reporter humblebrag of the year.
Odds are, if you have them in your contact database, you already know them; you're not going to want Siri to give you their Wikipedia bio.
Siri has myriad faults, and thankfully someone of Mossberg's stature might push Apple to address them, but this is not one of them.
Most of the narrative about Apple's maps products has simply been "Apple bad, Google good" and hasn't looked much deeper than this (with the exception of Justin O'Beirne's cartographic commentary, which is remarkably detailed though much of it is just a matter of taste). But Mossberg really hits on something here:
Apple's geocoding is way below par. The cartography is superb IMO. The routing is very good[1]. The source data isn't bad at all. But the geocoding is really error-prone. I've asked it for directions to Milton Keynes, a town with a population of 250,000 just 40 miles from here, and it won't give me anything other than the nearest branch office (in a completely different town) of a company whose HQ happens to be in MK. I've asked it for directions to the village of Brize Norton and it flat-out refuses, sending me instead to the RAF base of the same name, even if I ask for "Brize Norton village", "Brize Norton School" and so on.
Yes, geocoding is famously hard. But as an OpenStreetMap volunteer I see that for anything other than granular street-level addresses (where we don't have the data), our grassroots geocoder is more reliable than that built by the most valuable company on earth. And as feedback to OSM shows, bad geocoding is the easiest way to make people think your entire map product is no good.
[1] except there's no bike routing... but hey Apple, if you want top-notch bike routing, give me a call ;)
Public transport info is only available for very small set of predefined cities (e.g SF, Paris). Apple Maps doesn't even show the stops on the map, when every OSM app has them.
I stay signed out and opted out of all Google Now features and tracking so this is just it does by default. Voice is my primary ui for my phone. It is very useful and I find it parses my input correctly nearly everytime, it's very rare for it to error.
Meanwhile, my wifes iphone 6 gets basic commands wrong frequently and she spends minutes typing all the time. Siri isn't losing, it has lost.
I don't think this actually matters to the iOS population however. It's well known they don't buy iphones for features.
edit: A sample question from the article "when is the presidential debate" is handled fine as an example
It reminds me in many ways of the handwriting recognition on the Newton--which is to say it appears to be only a hairsbreadth from absolutely useless.
I always select English as my default language in my phones, some years ago if I wanted to call someone with a Spanish name saying something like "Call Ramon Hernandez" or asking for directions "Take me to Periférico de la Juventud" I used to had to fake an American accent. Today I don't have to do that anymore, I can speak natural Spanish and then it can understand my Mexican accent when I speak English. So it's not just improving in bilingual queries, they have focused in accents and mispronunciations.
"here is what I found on the web for 'what is Yen's email address'"
O. M. G. ...
Siri is great for setting alarms and timers -- much faster than using the UI.
I asked her for the nearest gas station the other day while driving and I think she responded with a list of google search results? Which is almost comical. I'm not going to grab my phone and sift through Google search results while behind the wheel.
And every one of those "hundreds per day" is another person one step closer to disabling Siri. Which is what I did last month. Siri got things wrong so often, it was worse than useless.
In the real world nothing goes like this. We start with context and work outwards from there. And I don't mean know-everything-about-your-personal-life context, there's no need for that either.
I'd say more but that is what I am solving at Optik. We're a bit over a year into development and things are really starting to pick up. Using Cortana in the Hololens is making it painfully obvious just how close but how far off the mark remains most voice command software.
What I'm thinking of is basically SHRDLU [1] on steroids. Parse a ton of web pages. There are great natural language parsers that can parse most well-formed English sentences. Start with simple sentences like "Golden delicious is an apple" or "Barack Obama is the President". Then you store this in a Subject-Verb-Object database (I just learned that this is called a Triplestore [2]).
Every statement gets a plausibility value. Deal with ambiguity by adding multiple interpretations of a sentence (with different plausibilites if available). Assign an origin (e.g. website, author, quoted person ...) to each statement. Then, you could query this by asking "What do mice like?"... and it would make "Subject: Mice, Verb: like (enjoy), Object: ???" and return a list of solutions, ordered by plausibility.
Does anybody have any insight into why this isn't done or wouldn't work? It seems wierd that I can't ask my phone simple facts about the world, other than those whose form have been hardcoded.
(Now that I think about it, the opposite would also work. Hardcode a ton more commands. Hire 100 people, let them sift through the most common queries. Watch a few dozen testers, add add all queries they try to use. Instead of throwing computing resources at the problem, this would throw cheap labor at it.)
Anyway, it boggles the mind that I can't shout "OK Google, play 'itsy bity spider' on youtube" when my toddler demands it but won't release my phone :-). It opens a search and shows what I want as the first result (probably customized from my history), but I have to go the last mile myself.
[1] https://de.wikipedia.org/wiki/SHRDLU [2] https://en.wikipedia.org/wiki/Triplestore
Amazon is doing an amazing job of allowing the community to extend the voice functionality of their Amazon Echo. You can build an app that does just about anything you want and release it on their Amazon Echo store. There are already hundreds on there, and it's only a matter of time before someone releases a truly conversational A.I. integration for it.
My family uses our Amazon Echoes (all 6 of them, scattered strategically around the house) for hundreds of little things throughout the week. I control all of my lights, my thermostat, my entire home theater (via custom voice activated scripts). I even use it to make my tempurpedic bed vibrate ("Alexa, turn on bed vibration"). I use Echo for timers, and my morning alarm to wake up (which also triggers my bed vibration). I use it to query wikipedia subjects, perform quick math calculations, order new paper towels...
If you gave me a couple of hours, I could even whip up an Alexa integration that would let me open my garage door and remote start my car in the morning ("Alexa, turn on my car's air conditioner")! It would just be an Alexa command that triggered a custom script which would log into the web interface for my car link and would click the remote start button. Easy. I love building this sort of thing.
The possibilities are endless when you have an open, extensible platform. I don't understand why Apple and Google are being so dumb and close-minded, and still don't understand this!
When it does, it replies 'Here's what I found on the web...'
I try tricking Google's app with unlikely stuff but it gets it right most of the time, and gives a reasonable answer too.
It does come in handy when I'm driving and for reminders. It seems to do pretty well with directions, playing songs, playing podcasts, and messages.
I much rather tell Siri "remind me not to forget my lunch when I get out of the car" than try to do that with the reminders app.
I quickly gave up on all the features but using it to control music.
Then, I figured out that it's useless at that too unless I set specific playlists.
"Play Underoath" > "Finding results for 'under oath'"
... It's even worst for artists with strange names.
How am I to say "Play 30h!3".
It is even worst when you try to have it figure out music in other languages.
...
I turned off Siri.
And this doesn't even include the automatic actions taken by Google Now. Alerting me about travel times, flight delays, sports scores etc. etc. without even being asked. Ultimately, in Google's vision, the phones are rapidly becoming just a tiny window (no Microsoft.. not you), into all the knowledge and power that actually resides on the cloud. Not so much for Apple, whose core competency is, and has always been, in hardware design and supply chain management.
I've been an Android user for about 10 years and just switched to iOS within the last week.
Siri is really the only feature on iPhone (so far) that I have found I prefer to Android.
While it's far from perfect, so far I have been quite pleased with Siri. It is able to let me do my most common tasks via voice commands. 9 times out of 10 Google Now couldn't even tell what I was saying. I attributed it to my somewhat strange accent, but Siri has been nearly perfect on this front. Good thing though as I loved the Swype keyboard on Android and hate typing on the iPhone; I fat finger everything on that tiny virtual keyboard.
My favorite example was a couple of years ago when I was at a bar. There was a song on the jukebox ("Coffee Pot" by Cajmere) which, for reference, is a house music song with a repeated vocal sample that says "It's time for the perculator [sic]". I assume they mean "percolator" but that's how the guy pronounces it in the sample.
Anyway, we were joking around and even though the bar was quite loud with people talking loudly all around and music playing, I pulled out my phone and asked "Ok, Google: is it time for the perculator?" I pronounced it incorrectly like the guy in the song and within seconds, I had results. The top one was a link to the video for "Coffee Pot" by Cajmere on Youtube and the rest were links to other stuff regarding the song.
Now, this was at least 2 years ago and possibly more. The response to my silly joke of a question wasn't so witty as to give me a spoken "it's time for the perculator" or anything like that....but it could parse my words despite all the noise and competing speech around me, it understood a mispronounced word, and it knew that this lyric sample referred to a song with a different title (Coffee Pot) and linked me to the video.
That was seriously the first time I was really impressed by their voice assistant even if it was a ridiculous request. No idea what the result would be on today's Google Now or tomorrow's Assistant (guess I'll find out when my Pixel gets delivered). Still, even if just from an engineering and software angle, that made me grin a bigger geeky grin than I might have imagined.
Might be a lack of corpus? Idk. Seems like if they could perfectly solve translate, that wouldn't be an issue heh heh (strong if).
The wikipedia page for Knowledge graph cites Thomas Jefferson as an example, so I asked 'what years was Thomas Jefferson in office?', and Google deferred to the first paragraph of TJ's wiki.
Then I asked 'What year was Thomas Jefferson born?', and it gave me a direct answer.
So it sorta works, but it's not super awesome.
It's not that different with Siri, they always brag with features in English, localization takes a while and has much less resources allocated it seems.
So a good exercise is to rewatch that movie, and in each scene where the assistant does something useful or cool, take a moment to reflect on what sort of data they would need access to, and how that data might be used.
I think where things get murky is that the movie didn't really touch on the company behind it at all, and if anything it almost gave the impression of being more like Apple than Google. As such, it neatly skirted today's concerns where the clear leader in the space (Google), is an advertising company that monetizes it's users' data and is constantly raising privacy concerns.
It is a fine line to walk, but ultimately everyone will need to weigh the decision in their mind as to the right balance of privacy and convenience.
This is actually Google's bread and butter, and why Mossberg found so many cases where Google Now does better than Siri. It's not clear how Apple can catch up, either, given Google's massive advantage in training data.
What was cool about that was that I didn't have to know how to get their info nor what format it was in, provided that they followed some standard namespaces.
Look up JSON-LD while you are at it. This method can be extended to the IoT and then we will get some interesting outcomes.
I like that Google seems to just display information for may of the responses vs. trying to cram in some overly long witty reply or verbatim reading of the text that takes too long to speak back.
Having some sort of speed/succinctness setting would be helpful in that regard.
Didn't realize they might be different products.
In this case, it would have to know the various names of sky-blue colors, like "azure", "cerulean", "process cyan", or technical answers like "transparent with Rayleigh scattering" or "hue value between 190 and 200 degrees". Then it would have to cross reference that with things I have already seen or read, and then prioritize several potential responses before finally playing me an audio clip from X-Files 3:17 "Pusher", when the eponymous character is being transported in the back of a cop car.
The Venn overlap between the set of "contacts in your phone" and "people with Wikipedia bios" is likely rather small. Hence why I think it's a faulty premise to complain about Siri defaulting to contact card when these two sets do intersect.
If I'm asking the question - maybe a friend is nearby who doesn't know them and I'm too lazy to explain - then I want to know who they are, not what their contact info is.
Even simple questions like "Who is..." has many different interpretations. A human will understand the context. An AI won't, because you can't derive the context from the words themselves. It's a function of social setting, physical setting, relationship, previous conversations, and so on.
At the moment conversational interfaces are more like a Bash shell with a speech recogniser on the front. The shell needs a precisely formed command and has almost no concept of state or context at all. (I think Siri actually has some, but not much.)
So it's completely unrealistic to expect CIs to be able to do this today. It will only be possible when NLP gets a whole lot more sophisticated and starts tracking context and state - although even that will still be a hard problem, because social state is defined as much by location, physical surroundings, time of day, and custom as by the words being used.
A contact card could definitely be used for that. If one exists, Siri should give the info from it and then wait to see if the user also wants external information (from Wikipedia or elsewhere).