Alexa is a glorified lighting management system for me. Given it’s current capabilities, that’s all it’s going to be for the foreseeable future. Oh, I’ll ask it the weather sometimes, but come on - what can Alexa really do that’s all that compelling, that already isn’t a capability of Siri/Google?
I don't have an Alexa myself, I'm a Google Home user. One thing that's always impressed me is the ability of Alexa to order things from Amazon just by issuing voice commands. Although, I'm sure that comes with its own host of problems, especially if you're not just ordering a generic product.
You can order things from Amazon using Alexa but you can’t delete items from your cart using Alexa. For me this makes it useless. Not being able to delete items makes it user unfriendly. Unless Amazon changes this I’m not going to use Alexa.
Two timers at once. Great in the kitchen. But you’re right... Why can’t Siri do that?
A device with multiple microphones is more useful from the couch, especially when I need to yell into the kitchen. I don’t want to feel the need to keep my phone within earshot at all times.
I wish Apple had such a $30 device for Siri but it’s unlikely.
edit Also, I completely agree. I wish there was a HomePod Dot type offering. I've been sprinkling HomePods throughout the home, but it's expensive and also obtrusive in places where you don't want something that bulky.
"Alexa set a timer for 15 minutes." (She's a cooking timer)
"Alexa what time is it" (She's a clock)
"Alexa switch the tv off" (She can't switch it on)
"Alexa, meow meow/Pika/fart" (Novelty nonsense)
"Alexa, turn the heating up in the living room" (Nest integration)
Yeah that's about it. The only really useful thing she does for us is music control on our sonos system - "Alexa, play the album X by Y in the kitchen". That's quite neat.
Calling between rooms has been the main reason I got a second device (first one for tinkering). Got a young baby so being able to wake up the partner across the other side of the house without needing hands when there's been an ... explosion is useful.
Adding things to a shopping list is the other one that works nicely.
That music integration with sonos sounds really nice.
I bought an Echo speaker when they first came out. The sound quality is impressive (to my untrained ear), but the development experience is not.
The first thing I wanted to do was to add a feature so I could add a task to my to-do list software which is not supported by Alexa. It turns out that you cannot construct a sentence along the lines of "Tell Asana to add a task: <task>". You can't actually have a 'slot' which contains a freeform piece of text, even if it is the last piece of text in the sentence.
The Alexa API differs between regions, so the North America version of the API supports this but the EU version doesn't. It was removed from the NA version for a while but placed back after a bit of an uproar.
I think you could develop some more useful stuff using Alexa if only this feature was consistently available. I cannot think of a good reason why it isn't.
I now just mainly have it as a Spotify speaker, and occasionally I use it as an expensive egg timer. I normally have it on mute because I find it activates and starts recording private conversations. I don't see it getting better.
For what it's worth, in two years our Google Home has only 'misfired' once, and that was when someone on TV said 'Hey Google'. It is in the living room and sees heavy use by a family of five.
> our Google Home has only 'misfired' once ... when someone on TV said 'Hey Google'
Is that a misfire? It reacted to the trigger phrase. Have you trained it to only react to your voice (I know that's an option for Google assistant, but I didn't do it for mine).
Amazon added a real-time sanity check, so if too many Echo devices are triggered with similar sounds within a few hundred milliseconds, it will ignore them. This was after some TV ads or shows had intentionally abused it.
One of my coworkers has a friend named Siri. When Apple's siri first came out somehow she kept getting texts from friends "remind me tuesday to call the doctor". She didn't realize what was happening so she kept track of those reminders and texted her friends back until everyone sorted their technology out.
My roommate and I have a Google Home Mini in the living room, and to make the accidental activations more noticeable, my roommate set the accessibility "ding" sound for when it (thinks it) heard the trigger phrase. It gets activated a lot by television (maybe once every few hours of TV), and we generally have no idea what phrase accidentally triggered it.
This is the correct solution. The Alexa Skills Kit contains the Amazon.SearchQuery slot which meets your needs, but Lex doesn't. This is one of the two missing slots from Alexa in Lex. If you read articles regarding Lex slots you might get inaccurate info as they are slightly differing products.
Thanks, I just did some searching and found that this was added in February 2018 - I think I gave up on my project in 2017 so it now looks like I could program something like what I wanted. I still have the source code so that's nice.
However I'm just thinking of getting rid of my Alexa devices anyway - I was discussing political events with my partner last night, spotted the Alexa device light up (I think I said "a letter") and noticed shortly after that I was moderating my speech and not referencing controversial topics. I later just set it on mute.
For research: have another device perform "audio fuzzing" of Alexa, trying random phrases. Use a camera to look for Alexa turning on. Bonus points, check the list of Alexa recordings (via web interface) to confirm 1:1 relationship between device activation and recorded file. Publish the inadvertent "watchlist" of phrases, monitor over time.
This issue has killed my interest in basically all smart-home products I can find on the market, even ignoring their massive privacy issues.
The most useful 'basic' behavior I could think of for a smart home device was to check the weather and trigger my alarm earlier in bad weather or heavy traffic. I knew IFTTT was capable of running scripts then a trigger happened, checking the weather and traffic, and setting off an alarm, so it seemed obvious. I literally wanted "if this (or this), then that"!
No such luck. The basic IFTTT setup couldn't do it at all, no existing app could do it, and the developer program was invite-only. I got in after quite a long time, and even then it wasn't obvious. IFTTT wanted to treat 'check weather' as a script output exclusively, which I couldn't feed into any other system. The best I could do was be told the weather when I woke up. So, I stuck with the phone that could already do that.
That specific situation might have improved, but the general ecosystem doesn't seem much better. The only smart-home features I see available that I would use are list-making, music/media playing, and quick reference. But the features I would actually value are dense integration between apps, floating scripts like the one you describe, and non-user-triggered events to turn active tasks into passive ones. Those seem to be the features which are least available, even when they would be easy to implement.
I have several, along with Google Home too. They all suffer from the same issue: it's awkward to use with third party skills. Unless they are the highly integrated ones (Spotify, Logitech Harmony) in which case it's great.
It's the same for Siri, the API is so restrictive.
The whole "tell XX to YY" isn't convenient.
One workaround on Google Home is to use IFTTT, and in that case you can customize the whole phrase and response, and then it's pretty nice, even though the latency is high.
"-Okay Google, open the blinds -Sure thing Commander" never gets old.
Absolutely. Until the voice assistants become a bit more intelligent with the language used to activate them, it feels so rigid and un-natural communicating with them.
I don't see a great future for voice assistants in the near future, until we truly solve the problem of intent (in a natural way) and accordingly can respond.
You don't have to do that - the system can usually infer what skill you want from your utterance. [0]. And the "open the blinds" thing can be done on Alexa with a custom routine (Alexa specific IFTT).
My Echo was an interesting novelty at first. But a couple years in now, it is essentially (in order of usefulness):
1. An alarm clock / kitchen timer
2. A thing that tells me the weather report during morning coffee, so I know what to wear
3. A DJ that my kids yell at to play pop music
If it died tomorrow, I'd probably just go back to using my phone for these 3 things rather than buy a new one. I can't imagine getting into it enough to explore third-party "skills".
Find the phone, unlock the phone, find and open the app, find and click the control. That requires some effort.
Saying commands, having physical buttons at the expected location in your home, NFC tags.. there's a bunch of more convenient ways to do extremely repetitive tasks that would otherwise take you more time to do.
I guess it still has value if you have a non-privacy reason not to use voice control on your phone.
Some people don't consistently have their phones on hand when they're at home, which would do it. And some people have problems with unwanted voice activation - my experience was that the voice-printing on "OK Google" was not actually all that personalized, and sometimes it did totally unexpected things like breaking out of navigation to make a phone call. Last time I tested it, keeping Google voice recognition on was also a massive battery drain, but I assume that's improved.
(That phone call was particularly ludicrous: Android offered me an unprompted 'helpful tip' that I could say things like "call mom", but since the tip fired when it was already listening for a navigation command, it accepted its own instruction.)
Same, although I'd say a timer and weather machine that I talk to everyday is a runaway hit. There are few other things I own besides my bed, phone, clothes and a couple major appliances that I use every single day.
I came to say the same. For my 15mo old son, I prefer asking Alexa to play music on the Echo because I don't have to ignore him (in a way) to go tapping on my phone -- which makes him very curious about what I'm doing.
As a plus, he babbles at the Echo when he wants to listen to music.
And that's a killer feature for me; it's so damn convenient to use a voice interface to add items to the shopping list as you run out of item that I'd probably replace it just for that.
It's possible to access the list via their API so theoretically you could build your own solution. I agree the Alexa app is not good for accessing the lists.
I had the same problem with google home. I already had a shared (with the SO) google keep note for my shopping list. I wrote a google keep integration with an IFTTT app to solve this, letting me do a "OK Google, we need X" to add X to the list.
https://github.com/Resinderate/shopping-list/blob/master/sho...
When I first got a Google home, the keep integration was built in. And then they moved it to Google Express shopping list, suddenly, which was infuriating. Ended up using ifttt with Asana. Will try with keep as I preferred keep for shared shopping and Todo lists.
Pretty much the same for me, although I'll add (I have Google Homes but pretty much the same thing):
1. Metric / imperial conversions, especially in the kitchen - this is one feature that is miles better than using a phone or computer if my hands are dirty cooking something and I want to know how many grams 12 ounces is or something like that.
2. Intercom between my 2 google homes, one in my kitchen and one in our converted attic playroom - it's so much nicer to use Google home to call my kids down for dinner vs. screaming up two flights of stairs.
3. Making quick phone calls
4. Finding my phone - I lose my phone constantly, and it's super handy that I can get my google home to make it ring even if it's on silent.
Sorry that should have been timers. If I tell Siri to set a timer (while a timer is running) it still asks if I want to change the timer or leave it alone.
It can definitely read recipes, although I've only ever had it read me recipes I've found by voice. I don't know how to explicitly tell it the recipe I want. It's still kind of cool as it goes step by step, although that can be a bit buggy, especially if you're also listening to music, but i still use it from time to time
2. "Hey Google, announce dinner is ready". It handles standard meals with a standard message. Otherwise it will play back a recording of what you said. You can also use "broadcast" instead of "announce"
3. "Hey Google, call my wife". or mom or person's name.
For an elderly relative of mine who no longer has the physical dexterity they used to, this has been a killer feature of voice assistants for them. No longer do they need to fiddle with awkward buttons to change the radio station or find out what the weather is, they just... ask and it happens.
I can turn lights off without getting off my bed, and damn, has it made my life crazy simple. I was in India for few months last year and I realized a large part of me getting up early was to just switch on the geyser, open the door for the cook and turn the lights on. I would get back to bed after and check up fb till the water got hot. Just the fact that Echo can do these things without making me get off the bed will be a crazy value addition to my current decently privileged life.
The problem is the interface. Voice commands and their responses are linear, one dimensional. It’s difficult to represent complex interaction within that scope. Think of all the investment that has gone into telephone based automated customer support. The best interface conceived so far is the dreaded phone tree. That’s essentially the same interface smart speakers are exposing.
The opportunity is to figure out how to better utilize the voice based medium. No one has done it yet. When they do, it will also likely improve the experience around screen readers and accessibility.
I think you're right, but on the other hand, a skill is faster created than a complex 2D UI.
Sure, I won't use a conversational interface to build the next Photoshop.
I could tell it stuff like, what I ate and it calculates my kcals or macros and tells me how much I have left to eat this day or what other stuff I should eat to hit my macros etc.
I could tell it what I bought and it would categorize the bills.
Some things are just too bothersome to do with my hands.
The medium of voice is very rich and capable when talking to a human, so this is something limited by the intelligence of the system you're interacting with.
Is it? I mean, some folks do great talking. Lots of them. Most are in narrative communication. If you get much beyond that, you jump to physical interesting quickly. Even telling is benefited with gestures. Consider how vague most spoken directions are. When you can augment with pointing, things are easier.
This is right. No none has yet written “The Design of Everyday Things” for voice interaction. We just don’t know what works yet so we are redoing what was done before.
I don't find it appealing to use voice as an interface. I'd much rather have a button to press.
I think the new feature in iOS where it guesses what I want to do (send a message to ABC, for example) based on previous patterns is promising. A whole screen of these actions would be great.
Well, imagine a goal-based system. The first utterance to a voice assistant provides some inputs and a goal, like booking an Uber, but if you haven't provided all the inputs it needs to meet the goal, the assistant knows how to ask for more information, and also saves enough state to let you modify (or cancel) requests after the fact. This is arguably more advanced than a phone tree; you're not just giving keywords to advance along the branches.
Despite often seeming like a thin veneer of natural language processing on top of a search engine, I'm sure most users think of their voice assistants, or are willing them to be, a general AI agent. They don't want to have to care what app/skill/web-scrape is involved in enabling a response.
The whole skills ecosystem feels like an awkward stopgap on the path to AI. The language required to invoke them feels particularly clunky - "Alexa, ask ThingFinder about a thing" - and then, is the user supposed to be talking to ThingFinder now, or Alexa? She still sounds like Alexa, but she doesn't seem quite herself.
As developer choosing an invocation name is fraught with difficulty. There are only so many natural sounding names for something which does a particular thing, without incongruously inserting some invented branding word onto it - "Alexa open Tidy Tide Tables". If someone's already using the most natural name you are free to use exactly the same, but then who knows whose skill will be launched? And you'd better make sure your skill's name doesn't clash with anything else in the world at large, like the entire history of music for example, "Alexa play Wicked Game". It's all a bit of a mess.
Yeah, that's a good point. People are good at mentally compartmentalizing behaviors with personalities. If Alexa doesn't allow each agent to have a personality, people are going to have a hard time keeping the behaviors straight.
Kinda reminds me of Neuromancer, where a superintelligent AI was broken into two pieces to avoid detection. One part was good at personality, and the other part had to mimic people in order to communicate. It was very unnerving for people to talk to an AI that was copying the personality of someone they knew.
Anecdotally a developer friend of mine strongly suspects, most people, are like him and not actually using these home speakers very much.
I’ve been working on a an Audio App that reads any articles to you for iOS/Android and planned to bring it to the Echo and he seems to suggest my efforts are totally wasted. Although their market reach seems to be massive, likely related to their cheap price, not surprised by this article that their actual usage is incredibly low. Most people seem to buy them and then forget about them.
As a shameless self plug, if you would like to check out my app that reads articles to you using beautiful sounding AI/ML, find it here:
I use the same features in google home but long for proper integration with Google Keep.
I had used home to populate my shopping list but now that has disappeared from the home app and you have to go to shoppinglist.google.com
I don't understand why i cant tell Home to add an item to a specific Keep list.
Playing songs/podcasts from Spotify on Echo while controlling it from my phone/computer is my favourite use of the speaker now. Ironically this action involves no voice commands whatsoever except occasionally telling the speaker to pause/resume/adjust volume when I'm away from my phone/computer.
As a counterpoint, thanks to an Echo in most rooms of my house and Hue lights everywhere I almost never touch a light switch. Likewise I generally use the Echo timers instead of oven timers for cooking, and I like using them as Spotify speakers. And I can't remember the last time I left my house before I asked the Echo about the weather. Sometimes my wife even uses it as an intercom if I'm in my office, but I concede that feature is a bit janky.
Whenever my wife and I go away we usually remark to each other that it suddenly feels very backwards to have to do things without the Echo. That might not be the best word, but it definitely feels like we're missing something integral to our home life when we don't have them around.
That being said, it's expensive to set up Hue lights everywhere and an Echo in every room. My setup might be one of the reasons I use it so much.
I think the problem with your app on Alexa is the interaction model - if your app could push to my Alexa so I can open an article on my phone and then say "Alexa read this article", or use the iOS share menu and share to your app on my Alexa that'd be pretty nice. But I don't see myself saying "Alexa, tell Articulu to read ...." if that's the interaction model it'd force.
Totally agreed with GP. Needing to somehow uniquely identify an article over voice assistant will be painful. However, something like Pocket, where you can push articles/pages into the record and call them back by voice or tag would be cool, especially with the ability to queue articles in a desired order.
I agree with you -- it's a fad product with limited utility.
The inability of any of these solutions to identify the user in a meaningful way or interact usefully without involving the whole room kneecaps the utility of the products beyond actions that you take in a public space.
Turning on/off lights, wireless speaker and replacing the landline are long term probably the killer apps. Not trivial, but no smartphone either.
Pretty much same, the Wink integration lets us turn on/off lights and do other things like “Turn on Movie Theater” which dims certain lights and turns off all others.
We use Alexa to play music when we just want something going in the background as well. Additionally we use timers and the grocery list.
Wife uses the daily news rundown.
Kids ask Alexa questions (what’s the fastest bird), have it make fart sounds, and that’s about it.
I got a pair of buzzers to play the quiz game, it was so horribly janky they’ve been used twice. I scan the the apps list and nothing strikes and as worth even trying.
I use the flash briefing thing every day. I suspect most people don’t know about it. I wake up, stumble into the shower and say “Alexa play the news” on the way. By the time I finish brushing my teeth I got my curated news read to me. Pretty nice.
Yeah I really wonder if these speakers are a sort of fad that folks feel is super cool and gravitate too ... and pretty quickly find there isn't a lot of use.
I am surprised there isn't a demo available on the site, at least there wasn't an obvious one I saw. I imagine your conversion rate would improve from that landing page if customers could hear a sample!
Maybe the problem is with the entire concept of distinct, discrete apps. Voice as an “interface” is based around conversation, not functional blocks and clickable items. A truly enjoyable and useful Alexa would have a conversation with you - even if a practical and transactional one - and the various functions and “apps” would be woven seamlessly into that, just as a human assistant might learn a new skill and work it helpfully into the conversation.
I think the whole voice activation hype inflated itself where people are naturally wanting to see the next big thing and of course people used this as an example - the title says it all.
Furthermore we need to look at what makes voice less restrictive and that’s making it instant. They are still reliant on it being cloud / web based which creates a lag and makes it awkward and nothing like ‘talking to something or someone ‘
I think we still don't have a good model for how voice UI needs to work. The equivalent of GUI for mouse-and-keyboard machines or tap-and-swipe interfaces for mobile.
If we did have a UI model, I imagine we'd be controlling car stereos and driver-mode mobile phones with voice already.
I wonder if any of these 80k apps have the "magic sauce" but just haven't applied it to the right problem.
Controlling a mobile phone while driving while using your voice is definitely a thing.
And I suspect if you have your phone hooked up to your car stereo, or integrated using Google or Apple's in-car docks, then using your voice to play music is also possible.
I think your point about not having a good model is correct, though, for anything other than basic uses (phoning, messaging, playing music, navigation, weather, etc)
The most frequent use cases for us in order of use-fullness:
1. Spotify
2. TV control (with Harmony) - great for the kids so they don't break the remotes
3. Home monitoring when we are away (drop in on the echo spot).
But for me home security seems to be an unexplored opportunity for Amazon (and Google). Those microphones are very sensitive - and if paired with motion sense on other devices (a Nest or something) start to give you powerful intrusion detection tools and alerting. How hard would it be for them to recognise the sound of breaking glass or the presence of someone in the room at a time they shouldn't be there?
Obviously solving the pet issue is a challenge, but I think it's an interesting use case to explore.
Yep. It's a pain to have to say "Alexa, ask Chevrolet to start my car". Why can't I just say "Alexa, start my car". Clearly it's the only car related skill...
While I feel Alexa has a lot of potential from the couple POC apps I've developed, the device class itself simply isn't where I need it to be to fully capitalize on it yet.
Alexa is just an input/output device. It is limited, which is OK but that means there will not be a 'runaway hit'. There's just not enough of a surface area for something so comprehensive.
It's like saying 'There are 80k mouse enabled apps and no runaway hit'.
What it can do is have an app for nearly anything, though, that makes sense for the form factor. It's on its way. The next step of its evolution is to look at what apps work and what conventions can be pulled from those as a general standard. Users will be much happier when they can nearly instantly download an Alexa enabled interface for an app and have it work intuitively. And this is doubly important for Alexa because there isn't deep feedback like you have with a mouse where you can see the things you aren't doing to gather hints at what's possible -- you just have to know or, at least know how to find the answer, like a command line.
Which gives me an idea 'man for Alexa'. At least we can standardize a help menu.
That's the thing about USPs...most are idealized possibilities not yet realized.
I mean, look at VR and AR and, hell, even AI.
But AI is not useless even though it hasn't reached it generalized intelligence promise. It is adding tremendous value even though it has landed in the limited middle.
I don't mind learning Alexa's syntax and what it expects of me. I get value from what it can do well. As long as that's true, I think it can miss its more grandiose promises and still be a huge success.
> Alexa is just an input/output device. It is limited, which is OK but that means there will not be a 'runaway hit'.
I like the way you say that. My Google Home is a remote control for my mouth. It's like licking a keyboard one key at a time in the dark. Simple queries are easy, but any non-trivial query just aint't gonna happen at the moment.
I don't believe your comparison really works. A mouse is useless on its own. An Alexa device can operate on its own.
I do agree with your latter sentiment. It does feel like a command line at times. More so like trying to figure out a text adventure game. Zork would be very difficult if you didn't know the basic functions/words. That's what Alexa feels like most of the time for me. I'd love a help menu.
It's more useful with third-party apps but those are still a part of the Alexa system. That's like saying a computer is useless without software installed. I wouldnt use my computer very often if I wasnt able to install software but that doesnt mean it isnt functional without it.
Plus you would still have the entire Amazon app ecosystem. While their apps may not seem perfect they do have a free music section that could at least partially replace spotify.
Other first-party apps on Alexa allow you to access lot's of other Amazon services like purchasing items from their website. That's a lot of functionality even if we are excluding third-party applications.
But the mouse does have a runaway hit: Microsoft Windows.
That being said, it took 22 years to get from Engelbart's original mouse to the first version of Windows that really took off (3.0), so perhaps we're just too early on in this product cycle for the hit to have emerged yet.
I'd say the Alexa is more like birth of the personal computer. Computers weren't very good in the mid 1970's.
Now that companies realize there's a product here, there will be an arm's race. Look for big improvements in next decade. Lots of people and billions of dollars are about to go into making these products better.
To take this analogy further, it is interesting that by say 1976, we seem to have already realized that what you need to make voice devices actually useful is a screen and a GUI.
At any rate, being connected to the cloud will allow for faster iteration. Once it becomes a solved problem, then you can more easily remove the network.
That’s why you don’t see rapid iteration in the rest of the world.
Google, Apple, Amazon can update their voice devices continuously without a local software release. If the next generation voice algorithms needs twice as much hardware, your $30 device will still work because the processing is in the cloud.
Video games, for example, have been trying to move to the cloud. Put all the code in the cloud and just send the pixels
I don't want rapid (rabid?) iteration, I want shit that works.
"The cloud" is little more than a hyped-up, glorified business objective; human beings shouldn't have to be connected to the hive mind to enjoy the full benefits of technology. Nothing smacks of SV-style elitism more than the proliferation of "the cloud"
And you can pry my locally-installed videogames from my cold-dead hands. I hope every single streaming startup in that sector fails spectacularly.
Which is fine, as my life is perfectly great without. I like your assumption that all innovation is supposedly good...
And it's 100 million amazon devices because Amazon pushes them relentlessly and has been for some time. Their utility is questionable, even the article mentions that.
>I'd say the Alexa is more like birth of the personal computer.
There were advancements in Text-To-Speech and/or speech recognition, stretching back to the early days of PC/Mac e.g. DragonDictate[1] using Hidden Markov Models, IBM ViaVoice[2], latterly Nuance Dragon NaturallySpeaking[3]. Although, they were not perfect in their early iterations. However, they got progressively better and fairly impressive once trained properly.
I would rather draw comparisons with their current day counterparts like Amazon Polly or Lyrebird[4] et al., than associate voice assistants with a paradigm shift.
Disclaimer: I haven't played with this since 2016, so may be out of date.
Having played with the Alexa and Google home. My take on why there are no hits is that it's very clunky to invoke apps, especially for one-off queries.
"Hey, Alexa/Google, Ask <keyword> to do <blah>" or some variation on that. Starting an app and then interacting with it works fine for some use cases but not all.
It really needs to be a flow of install app for given type of task, say searching for a hotel, and then like Android, there are defaults. So I say "hey alexa, find me a hotel room in blah" and it just knows to hand off to whatever my favourite is.
So long as all the core easy commands are only accessible to Google/Amazon, this is going to continue to be a problem.
And it hasn't really changed since 2016 so that is still mostly spot on.
I think the real problem though is that people aren't good at remembering voice commands, either how to invoke the command or that the command even exists. Without visual cues it's easy to forget everything.
You're not going to have a runaway hit when it's predicated on everyone remembering that special thing to say. It's like trying to have a runaway hit that had to be based on a keyboard shortcut.
All of those services are great for the basics, e.g. How's the weather / set a timer / remind me in N minutes / order toilet paper ... Everything else is a bit tougher in my opinion.
Agreed, there needs to be a paradigm shift in registering intents. I should not create a skill that the user needs to remember a command for. Instead, my skill should be able to hook into intents and return structured data. For example:
"Me: Alexa, find me a hotel near the beach in Destin, Florida that is available next week"
"Alexa: Hotels.com has a three star hotel available for $987 for six nights, and Airbnb has a four bedroom condo available for the same dates. Would you like to book one of these, or hear more options?"
"Me: Book the first hotel."
...and so on...
In this instance, Amazon has to define the schema for user intention, and hotels.com and Airbnb must provide interfaces to that intent, responding with normalized/standardized data that Alexa can then reconfigure into an appropriate response. Amazon has to throw their weight around to get other companies to play the game, and it's on them to organize the returned data into an appropriate response. This is probably where a discussion about the semantic web would be useful :)
Using Alexa for tasks like this is a qualitatively different experience. Hands-free/voice control can be very useful in that your hands can be busy, and you don't need to find or walk to and manipulate a device or remote.
If you're cooking, the timer on your oven should literally be right in front of your face.
Every use case I hear parroted to justify the installation of audio surveillance devices in our homes sounds like a solution desperately in search of a problem. What do you plan to do with the milliseconds of time you may have saved?
Sure, but I can ask Alexa what the weather is like while actively doing something else, like say before brushing my teeth or while in the shower. Or while in my kids closet, picking out clothes for the next day.
And while I can set a single timer on my microwave/oven, or set them on my phone, being able to set a timer, hands free, while cooking is pretty damn convenient. I can set a timer while my hands are dirty from mixing meat for meatloaf or handling chicken.
There's something to be said about the convenience factor, and that's what most technological innovations are about, making life easier or more convenient.
I disagree...the killer app is a smart speaker (i.e. Amazon Music + voice control). That in an of itself is pretty useful, especially when you can get a reasonably good one for under a hundred bucks now.
Outside of that, Alexa isn't the platform the industry thinks it is but rather the equivalent of the touch screen on a smart phone. It's another form of input.
There needs to be much better and more seamless integration with 3rd party apps for it to really be useful.
You are right, I would say (Music + Voice Control) as you can configure other services (Spotify for instance), there are a lot of negative comments here, I wonder if it is not due to the population of Commenters on Hackernews that would tend to be younger.
If you have a child it is nearly a nanny that helps your child doing his/her homework, plus you can connect several Alexa (from friends) and interact with them, if you have also several Alexa you can communicate in different rooms (speaks/broadcast)
I find it quicker to enter commands on my phone for most things. One skill I thought would be really useful was "Alexa, hows my commute?" but without a way to go in and 'star' specific trains it would tell me about trains leaving in the next few minutes which isn't helpful!
I also cant often remember what I'm supposed to say and I sometimes forget my words as if I've been put on the spot somehow. Overall its not a pleasant user experience.
I've spent a lot of time thinking about monetization on voice-assist apps. This is a challenging problem, and I suspect it's holding AAA quality 3rd party support for the platforms behind. There is no stand alone (from a mobileapp/webapp) voice-assist app that has brought in significant revenue. I'm going to break down some of the most common attempts at revenue and describe their pitfalls.
* Sell Audio (Spotify, Audible, MP3 style): If the goal is to get into this business than you'll have much better reach if you sell on smartphones. The voice-assist app will be supplementary at best.
* Sell products (Amazon/Ebay style): People aren't generally comfortable buying products with voice-assist. You'll have to compete directly with Amazon.
* Market products (In-line advertising style): You'll have to generate a lot of original content to plug sponsored products. The original content will have to be very compelling. It's probably more appealing to host the original content somewhere else primarily.
* Sell audio advertisement: What make the voice-assist platform compelling for content consumption? How can you create a very engaging voice-assist app? Will people tolerate audio advertisements? They tend to consume more time and be less interesting than visual ads.
226 comments
[ 3.4 ms ] story [ 235 ms ] threadhttps://www.cnbc.com/2018/12/21/african-grey-parrot-rocco-wa...
A device with multiple microphones is more useful from the couch, especially when I need to yell into the kitchen. I don’t want to feel the need to keep my phone within earshot at all times.
I wish Apple had such a $30 device for Siri but it’s unlikely.
edit Also, I completely agree. I wish there was a HomePod Dot type offering. I've been sprinkling HomePods throughout the home, but it's expensive and also obtrusive in places where you don't want something that bulky.
Adding things to a shopping list is the other one that works nicely.
That music integration with sonos sounds really nice.
Nothing groundbreaking though.
The first thing I wanted to do was to add a feature so I could add a task to my to-do list software which is not supported by Alexa. It turns out that you cannot construct a sentence along the lines of "Tell Asana to add a task: <task>". You can't actually have a 'slot' which contains a freeform piece of text, even if it is the last piece of text in the sentence.
The Alexa API differs between regions, so the North America version of the API supports this but the EU version doesn't. It was removed from the NA version for a while but placed back after a bit of an uproar.
I think you could develop some more useful stuff using Alexa if only this feature was consistently available. I cannot think of a good reason why it isn't.
I now just mainly have it as a Spotify speaker, and occasionally I use it as an expensive egg timer. I normally have it on mute because I find it activates and starts recording private conversations. I don't see it getting better.
What?!
I'm very anti these devices, now I'm even moreso!
Is that a misfire? It reacted to the trigger phrase. Have you trained it to only react to your voice (I know that's an option for Google assistant, but I didn't do it for mine).
You can at least change Alexa's wake word to a couple other options ("Echo" or "Computer").
What about AMAZON.SearchQuery?
However I'm just thinking of getting rid of my Alexa devices anyway - I was discussing political events with my partner last night, spotted the Alexa device light up (I think I said "a letter") and noticed shortly after that I was moderating my speech and not referencing controversial topics. I later just set it on mute.
The most useful 'basic' behavior I could think of for a smart home device was to check the weather and trigger my alarm earlier in bad weather or heavy traffic. I knew IFTTT was capable of running scripts then a trigger happened, checking the weather and traffic, and setting off an alarm, so it seemed obvious. I literally wanted "if this (or this), then that"!
No such luck. The basic IFTTT setup couldn't do it at all, no existing app could do it, and the developer program was invite-only. I got in after quite a long time, and even then it wasn't obvious. IFTTT wanted to treat 'check weather' as a script output exclusively, which I couldn't feed into any other system. The best I could do was be told the weather when I woke up. So, I stuck with the phone that could already do that.
That specific situation might have improved, but the general ecosystem doesn't seem much better. The only smart-home features I see available that I would use are list-making, music/media playing, and quick reference. But the features I would actually value are dense integration between apps, floating scripts like the one you describe, and non-user-triggered events to turn active tasks into passive ones. Those seem to be the features which are least available, even when they would be easy to implement.
It's the same for Siri, the API is so restrictive.
The whole "tell XX to YY" isn't convenient.
One workaround on Google Home is to use IFTTT, and in that case you can customize the whole phrase and response, and then it's pretty nice, even though the latency is high.
"-Okay Google, open the blinds -Sure thing Commander" never gets old.
I don't see a great future for voice assistants in the near future, until we truly solve the problem of intent (in a natural way) and accordingly can respond.
I can't imagine Amazon would hand the "Alexa, order more bread" keyword over to Instacart/Wal-Mart/whoever without a fight.
You don't have to do that - the system can usually infer what skill you want from your utterance. [0]. And the "open the blinds" thing can be done on Alexa with a custom routine (Alexa specific IFTT).
[0] https://developer.amazon.com/blogs/alexa/post/c870fd31-4f91-...
Disclaimer - interned on the team that built this
1. An alarm clock / kitchen timer
2. A thing that tells me the weather report during morning coffee, so I know what to wear
3. A DJ that my kids yell at to play pop music
If it died tomorrow, I'd probably just go back to using my phone for these 3 things rather than buy a new one. I can't imagine getting into it enough to explore third-party "skills".
I use it for: - I have to go in 1h but i only need 30 minutes to prepare the leave so timer for 30minutes - Pizza / food - short nap - for learning
And quite often.
We also user audible heavily. The kids live the boxcar children readings.
Find the phone, unlock the phone, find and open the app, find and click the control. That requires some effort.
Saying commands, having physical buttons at the expected location in your home, NFC tags.. there's a bunch of more convenient ways to do extremely repetitive tasks that would otherwise take you more time to do.
Some people don't consistently have their phones on hand when they're at home, which would do it. And some people have problems with unwanted voice activation - my experience was that the voice-printing on "OK Google" was not actually all that personalized, and sometimes it did totally unexpected things like breaking out of navigation to make a phone call. Last time I tested it, keeping Google voice recognition on was also a massive battery drain, but I assume that's improved.
(That phone call was particularly ludicrous: Android offered me an unprompted 'helpful tip' that I could say things like "call mom", but since the tip fired when it was already listening for a navigation command, it accepted its own instruction.)
As a plus, he babbles at the Echo when he wants to listen to music.
4. Shopping List.
And that's a killer feature for me; it's so damn convenient to use a voice interface to add items to the shopping list as you run out of item that I'd probably replace it just for that.
1. Metric / imperial conversions, especially in the kitchen - this is one feature that is miles better than using a phone or computer if my hands are dirty cooking something and I want to know how many grams 12 ounces is or something like that.
2. Intercom between my 2 google homes, one in my kitchen and one in our converted attic playroom - it's so much nicer to use Google home to call my kids down for dinner vs. screaming up two flights of stairs.
3. Making quick phone calls
4. Finding my phone - I lose my phone constantly, and it's super handy that I can get my google home to make it ring even if it's on silent.
* Setting multiple alarms (why can't apple do this??)
* Conversions
* Food questions
I think it would be rad if I could feed it a recipe and then have it read me the ingredients and instructions for each step. Maybe that exists?
It launched without that ability but an update sometime last year lets you set multiple alarms.
3. "Hey Google, call my wife". or mom or person's name.
4. "Hey Google, find my phone".
The opportunity is to figure out how to better utilize the voice based medium. No one has done it yet. When they do, it will also likely improve the experience around screen readers and accessibility.
Sure, I won't use a conversational interface to build the next Photoshop.
I could tell it stuff like, what I ate and it calculates my kcals or macros and tells me how much I have left to eat this day or what other stuff I should eat to hit my macros etc.
I could tell it what I bought and it would categorize the bills.
Some things are just too bothersome to do with my hands.
I think the new feature in iOS where it guesses what I want to do (send a message to ABC, for example) based on previous patterns is promising. A whole screen of these actions would be great.
The exposed API is extremely restrictive. No native SDK, all "apps" run on the cloud. Code - Compile - Test cycle is very slow.
The whole skills ecosystem feels like an awkward stopgap on the path to AI. The language required to invoke them feels particularly clunky - "Alexa, ask ThingFinder about a thing" - and then, is the user supposed to be talking to ThingFinder now, or Alexa? She still sounds like Alexa, but she doesn't seem quite herself.
As developer choosing an invocation name is fraught with difficulty. There are only so many natural sounding names for something which does a particular thing, without incongruously inserting some invented branding word onto it - "Alexa open Tidy Tide Tables". If someone's already using the most natural name you are free to use exactly the same, but then who knows whose skill will be launched? And you'd better make sure your skill's name doesn't clash with anything else in the world at large, like the entire history of music for example, "Alexa play Wicked Game". It's all a bit of a mess.
Kinda reminds me of Neuromancer, where a superintelligent AI was broken into two pieces to avoid detection. One part was good at personality, and the other part had to mimic people in order to communicate. It was very unnerving for people to talk to an AI that was copying the personality of someone they knew.
I’ve been working on a an Audio App that reads any articles to you for iOS/Android and planned to bring it to the Echo and he seems to suggest my efforts are totally wasted. Although their market reach seems to be massive, likely related to their cheap price, not surprised by this article that their actual usage is incredibly low. Most people seem to buy them and then forget about them.
As a shameless self plug, if you would like to check out my app that reads articles to you using beautiful sounding AI/ML, find it here:
https://articulu.com
Those are by far my killer features why i use it and why i like it.
I haven't looked at any amazon skill store as i don't see any reason for it.
Whenever my wife and I go away we usually remark to each other that it suddenly feels very backwards to have to do things without the Echo. That might not be the best word, but it definitely feels like we're missing something integral to our home life when we don't have them around.
That being said, it's expensive to set up Hue lights everywhere and an Echo in every room. My setup might be one of the reasons I use it so much.
EDIT: Is there a way to buy the app, or do I really have to rent it for $60 a year?
The inability of any of these solutions to identify the user in a meaningful way or interact usefully without involving the whole room kneecaps the utility of the products beyond actions that you take in a public space.
Turning on/off lights, wireless speaker and replacing the landline are long term probably the killer apps. Not trivial, but no smartphone either.
Wife uses the daily news rundown.
Kids ask Alexa questions (what’s the fastest bird), have it make fart sounds, and that’s about it.
I got a pair of buzzers to play the quiz game, it was so horribly janky they’ve been used twice. I scan the the apps list and nothing strikes and as worth even trying.
Furthermore we need to look at what makes voice less restrictive and that’s making it instant. They are still reliant on it being cloud / web based which creates a lag and makes it awkward and nothing like ‘talking to something or someone ‘
A good toolbox is filled with balanced tools. That there isn't some hyper-addictive time-suck application isn't a bad thing.
However, if it's a case of no one actually using the device, then there might be cause for concern.
If we did have a UI model, I imagine we'd be controlling car stereos and driver-mode mobile phones with voice already.
I wonder if any of these 80k apps have the "magic sauce" but just haven't applied it to the right problem.
And I suspect if you have your phone hooked up to your car stereo, or integrated using Google or Apple's in-car docks, then using your voice to play music is also possible.
I think your point about not having a good model is correct, though, for anything other than basic uses (phoning, messaging, playing music, navigation, weather, etc)
1. Spotify
2. TV control (with Harmony) - great for the kids so they don't break the remotes
3. Home monitoring when we are away (drop in on the echo spot).
But for me home security seems to be an unexplored opportunity for Amazon (and Google). Those microphones are very sensitive - and if paired with motion sense on other devices (a Nest or something) start to give you powerful intrusion detection tools and alerting. How hard would it be for them to recognise the sound of breaking glass or the presence of someone in the room at a time they shouldn't be there?
Obviously solving the pet issue is a challenge, but I think it's an interesting use case to explore.
Saying "Alexa, ask [Whatever skill name] to do something" is really unnatural and ruins the whole user experience.
They need sort out the grammar to make it actually feel conversational.
It's like saying 'There are 80k mouse enabled apps and no runaway hit'.
What it can do is have an app for nearly anything, though, that makes sense for the form factor. It's on its way. The next step of its evolution is to look at what apps work and what conventions can be pulled from those as a general standard. Users will be much happier when they can nearly instantly download an Alexa enabled interface for an app and have it work intuitively. And this is doubly important for Alexa because there isn't deep feedback like you have with a mouse where you can see the things you aren't doing to gather hints at what's possible -- you just have to know or, at least know how to find the answer, like a command line.
Which gives me an idea 'man for Alexa'. At least we can standardize a help menu.
I mean, look at VR and AR and, hell, even AI.
But AI is not useless even though it hasn't reached it generalized intelligence promise. It is adding tremendous value even though it has landed in the limited middle.
I don't mind learning Alexa's syntax and what it expects of me. I get value from what it can do well. As long as that's true, I think it can miss its more grandiose promises and still be a huge success.
I like the way you say that. My Google Home is a remote control for my mouth. It's like licking a keyboard one key at a time in the dark. Simple queries are easy, but any non-trivial query just aint't gonna happen at the moment.
Ewww. Good analogy, but eww.
I do agree with your latter sentiment. It does feel like a command line at times. More so like trying to figure out a text adventure game. Zork would be very difficult if you didn't know the basic functions/words. That's what Alexa feels like most of the time for me. I'd love a help menu.
Can it?
I don't know how much I'd use Alexa if it weren't for Spotify. There are some native apps but its real value is in the 3rd party apps it connects to.
Plus you would still have the entire Amazon app ecosystem. While their apps may not seem perfect they do have a free music section that could at least partially replace spotify. Other first-party apps on Alexa allow you to access lot's of other Amazon services like purchasing items from their website. That's a lot of functionality even if we are excluding third-party applications.
Only if you conveniently ignore the AWS services it uses.
That being said, it took 22 years to get from Engelbart's original mouse to the first version of Windows that really took off (3.0), so perhaps we're just too early on in this product cycle for the hit to have emerged yet.
Now that companies realize there's a product here, there will be an arm's race. Look for big improvements in next decade. Lots of people and billions of dollars are about to go into making these products better.
Like taking the Internet away?
At any rate, being connected to the cloud will allow for faster iteration. Once it becomes a solved problem, then you can more easily remove the network.
Google, Apple, Amazon can update their voice devices continuously without a local software release. If the next generation voice algorithms needs twice as much hardware, your $30 device will still work because the processing is in the cloud.
Video games, for example, have been trying to move to the cloud. Put all the code in the cloud and just send the pixels
"The cloud" is little more than a hyped-up, glorified business objective; human beings shouldn't have to be connected to the hive mind to enjoy the full benefits of technology. Nothing smacks of SV-style elitism more than the proliferation of "the cloud"
And you can pry my locally-installed videogames from my cold-dead hands. I hope every single streaming startup in that sector fails spectacularly.
Hopefully, with 100 million Amazon devices, and growing fast, you’ll have your product within a decade or two.
And it's 100 million amazon devices because Amazon pushes them relentlessly and has been for some time. Their utility is questionable, even the article mentions that.
Did you know over 3 billion devices run Java?
Trying and failing for a decade or more, because people like low latency and being able to customise things.
There were advancements in Text-To-Speech and/or speech recognition, stretching back to the early days of PC/Mac e.g. DragonDictate[1] using Hidden Markov Models, IBM ViaVoice[2], latterly Nuance Dragon NaturallySpeaking[3]. Although, they were not perfect in their early iterations. However, they got progressively better and fairly impressive once trained properly.
I would rather draw comparisons with their current day counterparts like Amazon Polly or Lyrebird[4] et al., than associate voice assistants with a paradigm shift.
https://en.wikipedia.org/wiki/Speech_recognition#1970-1990
[1]https://en.wikipedia.org/wiki/DragonDictate
[2]https://en.wikipedia.org/wiki/IBM_ViaVoice
[3]https://en.wikipedia.org/wiki/Dragon_NaturallySpeaking
[4]https://lyrebird.ai/
Having played with the Alexa and Google home. My take on why there are no hits is that it's very clunky to invoke apps, especially for one-off queries.
"Hey, Alexa/Google, Ask <keyword> to do <blah>" or some variation on that. Starting an app and then interacting with it works fine for some use cases but not all.
It really needs to be a flow of install app for given type of task, say searching for a hotel, and then like Android, there are defaults. So I say "hey alexa, find me a hotel room in blah" and it just knows to hand off to whatever my favourite is.
So long as all the core easy commands are only accessible to Google/Amazon, this is going to continue to be a problem.
I think the real problem though is that people aren't good at remembering voice commands, either how to invoke the command or that the command even exists. Without visual cues it's easy to forget everything.
You're not going to have a runaway hit when it's predicated on everyone remembering that special thing to say. It's like trying to have a runaway hit that had to be based on a keyboard shortcut.
All of those services are great for the basics, e.g. How's the weather / set a timer / remind me in N minutes / order toilet paper ... Everything else is a bit tougher in my opinion.
"Me: Alexa, find me a hotel near the beach in Destin, Florida that is available next week"
"Alexa: Hotels.com has a three star hotel available for $987 for six nights, and Airbnb has a four bedroom condo available for the same dates. Would you like to book one of these, or hear more options?"
"Me: Book the first hotel."
...and so on...
In this instance, Amazon has to define the schema for user intention, and hotels.com and Airbnb must provide interfaces to that intent, responding with normalized/standardized data that Alexa can then reconfigure into an appropriate response. Amazon has to throw their weight around to get other companies to play the game, and it's on them to organize the returned data into an appropriate response. This is probably where a discussion about the semantic web would be useful :)
Every use case I hear parroted to justify the installation of audio surveillance devices in our homes sounds like a solution desperately in search of a problem. What do you plan to do with the milliseconds of time you may have saved?
And while I can set a single timer on my microwave/oven, or set them on my phone, being able to set a timer, hands free, while cooking is pretty damn convenient. I can set a timer while my hands are dirty from mixing meat for meatloaf or handling chicken.
There's something to be said about the convenience factor, and that's what most technological innovations are about, making life easier or more convenient.
Outside of that, Alexa isn't the platform the industry thinks it is but rather the equivalent of the touch screen on a smart phone. It's another form of input.
There needs to be much better and more seamless integration with 3rd party apps for it to really be useful.
If you have a child it is nearly a nanny that helps your child doing his/her homework, plus you can connect several Alexa (from friends) and interact with them, if you have also several Alexa you can communicate in different rooms (speaks/broadcast)
I also cant often remember what I'm supposed to say and I sometimes forget my words as if I've been put on the spot somehow. Overall its not a pleasant user experience.
* Sell Audio (Spotify, Audible, MP3 style): If the goal is to get into this business than you'll have much better reach if you sell on smartphones. The voice-assist app will be supplementary at best.
* Sell products (Amazon/Ebay style): People aren't generally comfortable buying products with voice-assist. You'll have to compete directly with Amazon.
* Market products (In-line advertising style): You'll have to generate a lot of original content to plug sponsored products. The original content will have to be very compelling. It's probably more appealing to host the original content somewhere else primarily.
* Sell audio advertisement: What make the voice-assist platform compelling for content consumption? How can you create a very engaging voice-assist app? Will people tolerate audio advertisements? They tend to consume more time and be less interesting than visual ads.