Slashdot Mirror
Alexa, Siri and Google Assistant Desperately Want To Help You Do Your Routine -- But it Takes Too Much Programming and There Are Still Too Many Holes (wsj.com)
Google's Assistant and Amazon's Alexa are rapidly increasing their reach, and Apple's Siri is supposedly getting smarter. But all of these AI assistants are still too clumsy in day to day. David Pierce, writing for WSJ: My virtual assistant desperately wants to help me. Google Assistant, Amazon's Alexa, Apple's Siri -- even Samsung's Bixby and others -- have begun allowing users to set up "routines" that combine many actions into a single command. Shout "OK Google, good morning!" at your smart speaker and it can (in theory) open the blinds, turn on the lights, show you traffic and your calendar and turn on NPR. Tell Alexa to start a dance party, and watch it turn on the disco ball and fire up the "Glitter and Glowsticks" playlist. These routines embody what virtual assistants are meant to do, connecting all our gadgets and services and making everything work together. All you have to do is ask. And maybe not even that -- these tools aim to get to know you so well, they'll anticipate your needs. But these multistep systems are complicated to create, and they often require buying "smart" accessories and memorizing specific phrases.
In most cases, voice-controlled assistants have hit a wall where they perform a specific set of tasks well and not much else. They may be crazy ambitious, but they aren't ready to take on real work. If you are willing to do some finagling, there are already ways to make your devices and services work together better. Tools like IFTTT and Zapier let you connect web services, so you can automatically save every photo you share on Instagram into a Dropbox folder, or file your sales contacts into a spreadsheet. [...] All these tools offer sample routines, and I recommend trying a few. If you want to create a specific routine from scratch, just know: It's hard. It feels like putting together Ikea furniture without the instructions -- most of the pieces are there, but good luck building something that stands up. [...] A sufficiently smart home should observe and adapt to your needs. That kind of proactive, thoughtful help is a long way off. It will require computers that understand far more about us than they do now.
In most cases, voice-controlled assistants have hit a wall where they perform a specific set of tasks well and not much else. They may be crazy ambitious, but they aren't ready to take on real work. If you are willing to do some finagling, there are already ways to make your devices and services work together better. Tools like IFTTT and Zapier let you connect web services, so you can automatically save every photo you share on Instagram into a Dropbox folder, or file your sales contacts into a spreadsheet. [...] All these tools offer sample routines, and I recommend trying a few. If you want to create a specific routine from scratch, just know: It's hard. It feels like putting together Ikea furniture without the instructions -- most of the pieces are there, but good luck building something that stands up. [...] A sufficiently smart home should observe and adapt to your needs. That kind of proactive, thoughtful help is a long way off. It will require computers that understand far more about us than they do now.
...because the AI we have now are parlor tricks and there is no strong AI yet.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
I go to a friend's house, and he's got an Alexa (not sure which product), and he'll ask it to play music, which I'm more than capable of doing with a button or two on my iPod, and he'll say "Alexa, put milk on the grocery list," which I can do with a pencil and the pad hanging on the fridge. Somehow the joyous fangledness of these devices has passed me by. I have no interest. Zero.
open the pod bay doors
The description could equally well be replaced by, say, a button (or a touch-screen button) on a dinky controller, or say a raspberry pi if you like throwing cycles around, that'll trigger doing a number of things. It's really just triggering a shopping lists of small-ish tasks. You could even do it remotely with an "app" that sends a command to that little domotics-controlling box.
What is it about this voice control that makes it so super-duper hard? Please do explain.
Ever met an assistant that didn't need a year of training? Any assistant anywhere?
Assistant coach, executive assistant, teacher's assistant, lab assistant?
How about a protege? Oh wait, that's actually an assistant-in-training. . .for years.
Sorry friend, but you won't get anyone/anything/anybody to do what you want without telling them, showing them, and correcting them. And that takes time, by you.
Tough.
Oh yeah, and if it doesn't require _you_ to teach your assistant what you want, then it isn't your assistant, you're it's assistant. By definition. Because it's telling you how to behave.
Training is simple, all you need to do is give it all your data, without restrictions, and unfettered 24x7 microphone and camera,gps and network access. This will be used to make the service better!
People artificially restricting their assistants are what is keeping AI behind.
"Desperately Want To Help You Do Your Routine"
all they want, all they care about, is the money they make off your data, and the commissions/sales generated.
they don't give a shit about your life or your routine.
lexa, Siri and Google Assistant Desperately Want YOUR DATA
ftfy
This shit gives me the Yikes.
Are people REALLY into putting these crap technology inside their houses for "helping doing stuff"? For Christ Sakes!
I really don't see the point with these devices. Useless for me.
Lame.
In Putinist Russia, Alexa, Siri and Google Assistant DO YOU!
Schroedinger's Brexit: The UK is both in and out of the EU at the same time!
NO!! Get the f*** out of my life! My phone will not babysit me like my f***ing socialism gouvernment do (I'm from Canada)!
Will $CURRENT_YEAR be the year of the Linux Desktop?
In most cases, voice-controlled assistants have hit a wall where they perform a specific set of tasks well and not much else. They may be crazy ambitious, but they aren't ready to take on real work.
(bold mine...)
When these things can't do much more than repetitive tasks, they are gimmicks designed to fleece "zealots" of their hard earned cash in my opinion.
Those that used to use some of these gadgets at my office threw them away long ago after realizing that they had no real utility. Unfortunately, that wasn't before they parted with serious hard earned cash.
for an assistant named HAL 9000
Politics is Treachery, Religion is Brainwashing
It's very cool that you can set lights to come on at dusk.
It's also very cool that you can set lights to turn on when you come home.
In fact, I've got it set to turn on most of the house lights when I get home to full light and then off and to turn on to warm light at dusk (or a night light after 10pm)
Here's the kicker - the entire system doesn't understand previous states - so if you're out of the house after dusk the lights come on to warm light but when you come home, the lights come on at full on then turn off and STAY OFF - because that was the last script ran. There's no rules priority or return to previous state capability.
Auto-off is just as strange as well as the max "auto off" time for any light scene is 1 hour (at least through the various UIs). So I can't have a "lights on at sunset for 6 hours" scenario - I have to have TWO separate configurations - lights on at sunset, then lights off at midnight.
AI seems to be the new catch phrase for what is essentially sophisticated automation.
If I have to specifically tell devices about my life for them to be useful - that's not AI.
True AI will not require programming - it will learn on its own.
I think you are confused. I can hire an assistant today and say "throw me a party" and give some simple parameters and there will be a party of some sort. If the assistant is any good (and has access to the same resources your smartphone has) it will scour for music and social preferences and invite the correct people. At a minimum the person would ASK follow up questions to fill in any gaps. Even with little to no training. i.e. a day one assistant.
Try turning on your phone for the first time and telling it to throw you a party. It is just going to provide some quip back about how she doesn't party.
Don't forget about me. A little sign in here, a touch of WiFi there, and I can spy on you while don't absolutely nothing of value.
If you install lots of me, I still won't be of any use, but at least it'll be mildly amusing.
The difference is that while human assistants need training, they learn many many job related skills on their own without explicit instruction.
When machines can anticipate human needs without explicit instruction AI will be useful.
Explicitly training AI is no different than programming a computer with a different interface.
I bet you wear Prada, don't you?
Aah, change is good. -- Rafiki
Yeah, but it ain't easy. -- Simba
Over Christmas vacation, I saw my eight year old nephew use Google to avoid parental blocking on YouTube videos his parents didn't want him to watch, and use ever more creative ways to even hack my sister's phone to do the same thing.
Be careful what you ask for.
-- Tigger warning: This post may contain tiggers! --
Stop being a faggot, can't you?
Ok Google/Hey Siri,
What time is it?
Call it an "always on microphone".
I gave up on "home automation" when X10 was still a thing. The only thing these big players have added beyond clap-on/clap-off is a computer to spy on you. These things claiming to be "AI", yet they haven't the slightest clue that a holiday calls for alternative actions, UNLESS you meticulously program it to have alternative actions on a holiday. "Let's have a party" has completely different meaning to the average socialized human depending on if it is Oct 30 vs Dec 30
Aah, change is good. -- Rafiki
Yeah, but it ain't easy. -- Simba
They can shove them up their respective asses.
I think you'll find that if you hire a human assistant, and ask them to throw a party without explaining everything you do and don't want, and invite your boss, you'll lose your job.
It may be a good long time before computers have "common sense". But I wonder if enough pattern matching can be thrown at the applications that they are "good enough" for common tasks? If they sample enough requests from enough people they can probably narrow down an appropriate action based on similar responses from many other people.
How far can brute-force pattern matching carry these things? Putting common sense into these is of course the ideal, but until that Great Barrier is cracked, companies will try to push the limits of pattern matching. It will be interesting to see how far it can go.
Regardless, it may require a lot of personal info like the name of your family members, marriage status, what rooms are in your house, typical schedules (work, school, etc.), if you have stairs, if you have pets, etc.
Then again, a human butler would also need to know such to avoid asking redundant questions and to understand basic context. The difference is a butler is less likely to share such info with pesky marketers and dictators.
Table-ized A.I.
These virtual assistants don't want anything, they are literally incapable of wanting. However, the corporations behind them desperately want to integrate them into people's lives. To call their motives nefarious is an understatement because they are downright diabolical. Consider, really consider what they are trying to do with these devices. The growing reliance on smartphones was mostly a fluke that they exploited but this is an intentional effort to do something similar but exploit it in minimalist fashion. It seems hyperbolic on the firsthand but when you give it some thought about the total lack of boundaries these devices have (remotely updated without consent to do anything) then you can see the tip of the iceberg.
Anons need not reply. Questions end with a question mark.
Reminds me of Dragon Naturally Speaking that my boss in a law firm had me install on his computer -- complete with a legal vocabulary.
Comes time to train the goddam thing and he tries to get ME to do it.
After some discussion that failed to inform, I just did what he said. I told him to get up so I could do it for him. He said he had work to do, so install it on my computer; train it, and get back to him when it was ready.
I explained, slowly, how that doesn't work but he insisted.
I did as he said and when he was ready, I told him to go into my office. He said he wanted it on his computer. I reminded him of how that went down.
He huffed off to my computer room, sat down and started dictating into Word. Of course, the translation was stupid. He asked how in hell it would ever work and I said, it's trained to my voice.
"I'll sit in here and you tell me what you want and I'll inform my computer."
He didn't even ask me to uninstall it from his computer.
It little behooves the best of us to comment on the rest of us.
Pretty sure China will make these listening devices mandatory in all homes and businesses.
... for personal computers.
"Look! It even balances my checkbook!!!"
It little behooves the best of us to comment on the rest of us.
There are already better answers for these things. This is what your smart home hub is for. You are always going to have a poor experience on these things if you are connecting your voice assistant to things like hue, lifx, nest, harmony hub, etc directly. The answer is to integrate all those devices into a flexible and open smart controller like Home assist, Vera, etc and then integrate that and only that with your voice assistant(s).
That way you get a clean and modular system. You'll still have to learn the commands to some extent but because everything is presented to the voice assistant via one plugin/skill/service/whatever the syntax across devices will be uniform and consistent. Have an Alexa and GA or Sirii? Np, those things just talk to the same smart controller, state remains consistent and uniform, Everything and it's dog is including smart features but if you set up multiple things to control the same devices directly you are just asking for trouble.
Besides, if you are ever going to be serious about a smart home you can't use wifi for all your devices, wifi doesn't scale well to high client counts.
programming simple things is hard. and even the most simple task can have bugs in the code.
but people swear most jobs will be "automated" in the "near future"
Stop being morons that intentionally bring more and more surveillance devices into the sanctity and safety of your own home and just do your own shit instead.
I pity inanimate objects because they cannot move. From specks of dust to paperweights. Or a penny sealed in resin. Plastic santas in perpetual underwater snowstorms. Sculptures that appear to be moving but aren't. I feel sorry for them all.
A thing cannot depererately anything.
If I go to somebody else's house that has one of these things, all I have to do is to say, "Alexa, order one case of XY lube. Ship overnight." or some silly shit like that. The stupid things get unplugged every time I do that.
I don't respond to AC's.
Desperate to spy on you and then send ads to you.
None of them are getting into my house. I value my privacy too much.
Wake me up when there is some real A.I. in systems not this poor representation of it. DSS is not A.I.
I bought a Google assistant a few months ago, just after it became available in my native language. The one thing that surprised me is that I can't teach it anything. It's supposed to learn, but I can't speed up the learning process by explaining it what I want. Why can't I use my voice to tell it what kind of routines I want? Why is it not possible to teach it synonyms?
Most assistants just do what they are told. It's not like they are autonomous or they'r ust replace the job holder they're assisting.
You could get any moron off the street with half a brain for any of those positions.
"Get coffee", "Grade papers", "Take notes at this meeting", "Pipeette 50 ml of A into B, repeat 500 times".
I know someone in each of these roles, they were almost all wet-behind-the-ears randos with no field specific experience.
That is not at all about training.
That is about task assignment.
Please don't conflate the two.
They remain what they have been from day one: good for grins and giggles, party games - and very little more. The example given in the summary is a good one: I can do those tasks myself to my satisfaction with very little effort. For the most part, Alexa, Google Assistant, etc. solve problems that don't really exist, or that are so easy to solve with very little effort. They are utterly useless when it comes to trying to get them to do things where even a minimum of ambiguity is involved. Or even when no ambiguity is present: it seems to be beyond Google Assistant's capabilities to make phone calls using the mechanism I specify - e.g. Hangouts (or something else) as opposed to my voice plan. Plus it is pathetic (and comical) how little understanding is involved: ask them NOT to give you the weather forecast under any circumstances, and they will all immediately proceed to give you the weather forecast.
In summary, they are cute gimmicks, but, as of today, far, far less useful than we have been told.
How about breaking it into steps? Start out by having a control center that you tell it what textual commands do: a command-line interface. Thus, "Dim the front lights" will trigger a set of macros that dim the lights that you select and set to your selected brightness amount.
When the text commands are tested and work right, you THEN hook it up to the voice assistant. It may take some feedback training to make sure it interprets them right. It may pronounce a list of multiple candidate matches and ask which you intended if it has problems finding a "clean" match. After time it could learn your pronunciation pattern for a given command and stop giving you those multiple options. You can also set up a "synonym list" to cross-translate similar ways of saying the same thing. You may also want to create custom feedback responses to make sure it interpreted you correctly, such as "All front 4 lights have been dimmed to 40 percent" or the like. Or just play Beatles tune snippets that you associate with your lights. "She came in through the bathroom window" for example. (Inspired by an overzealous climbing fan.)
You may need to have the voice assistant distinguish between your custom commands versus general pre-canned commands like "what's the weather tomorrow"? Maybe a name like "Ralph" will trigger the custom bot, and "Alexa" the standard commercial bot.
Sure, it's a fair amount of work, but most "auto-house" people are fiddling hobbyists anyhow. Package them with good samples and virtual testing kits. If the interfaces are standardized, then each vender doesn't have to provide a voice assistant, just a command-line interface to hook up to another vendor's voice assistant.
In general I believe AI will have to learn to component-tize its processes in order to better tune, study, and dissect results. It's why alternatives to neural-nets, such as factor tables (see sig below), should be explored more. Neural nets have arguably been the first big practical AI breakthrough, but we seem to be hitting a wall with them and need look around wider for other tools.
I took an AI course in the late 1980's, and there were several techniques described that one could apply. No one technique can probably do the job by itself; good AI will probably take multiple techniques, and the industry needs practice glueing multiple techniques to triangulate answers and split the load up. Modularizing techniques is to both better understand & tune the intermediate results, and to allow mixing and matching the best-of-breed or best-fit-for-the-job. Modularization is what made the PC clones take off: you could buy monitors, disk drives, printers, add-on cards, etc. from many different vendors. AI seems still in the mid-70's where vendors tried to control the whole personal computing environment.
Lone-wolf bots won't cut it. Even if lone-wolf bots end up working, they may be too difficult to extract their reasoning steps from. It may be great job security for the one mop-headed professor who understands the gizmo, but if AI is to spread, then tool-type specialists without PhD's will need to be leveraged.
Table-ized A.I.
No problem, he can always slip in via the back passage now.
When Australians imitate the New Zealand accent, they sound like South Africans to the NZ ear. The South African accent, even the sea-level one is distinctly different to the NZ ear. Also the NZ accent shifts from north to south (it did this in pre-European times too) and with the level of education.
The key command in Dragon Dictate is "scrap that" to delete faulty or mis-heard text. For kiwis DD hears the would-be command as "scrip that", so rendering the program beyond useless.
Who said this party is for a fucking job? Moron.
Plus - try just "explaining" in the same way you would to a human to this "AI" and see what happens. The required detail and pain of providing the explanation is the difference. You see - intelligent humans already have an intuition. This AI is just pre-programmed shit with voice recognition. With this AI YOU are the assistant setting up a party that can be repeated over and over. With a human assistant you can provide general parameters and the assistant can figure out the more mundane details. With this AI you needs to provide everything.
Smart humans can figure shit out. A CEO doesn't (have to) plan every detail of his holiday party. He can tell an employee "throw a holiday party" and something appropriate will happen.
Very easily
These are smart devices, not intelligent devices.
WARNING: Smartphones have side effects--most of them undocumented.
So, after decades of work by our brightest minds and biggest egos, we can not build a thinking program? Maybe until then, some of the smaller problems could be reviewed? Like 3D Printing Medicine?
You're full of shit.
I'm from Texas and the program works.
It little behooves the best of us to comment on the rest of us.
What's wrong with that? It's done all the time on Slashdot and intellectual discussions. I'm curious how it would change the original premise if such did happen, hypothetically.
I guess what I trying to get at is whether or not "wanting" changes anything, and what the meaning of "want" is.
How can "want" or "desire" be objectively measured? Where are the Want-O-Meters in the labs? To discuss the original statement, I believe that has to be addressed. Does a bacterium "want" food? If not, does a worm? Does an octopus?
Table-ized A.I.
I've had a Samsung Galaxy S8 now for almost 18 months and I still haven't agreed to the Bixby terms of service which pop up from time to time (mostly as a result of accidentally pressing that damned Bixby button). It would be nice if I could remove the darned thing altogether (as well as some other shovelware) but since I can't, ain't no way I'm gonna agree to it. It can stay there in its (hopefully) unconfigured state forever.
I'm an IT guy and I still don't want a digital assistant doing anything for me. My phone is used for calls, texts, and the occasional map, nothing else. I don't really surf the net on the thing, as I despise tiny screens. I'm terribly old school, prefer the command line for most things except Web browsing and stuff like LDAP, and don't need to be modernized beyond what FreeBSD/OpenBSD offers. I miss mainframes, minis, and even Novell NetWare running NDS. All of this "let a program do something for you" is foreign to me. I like to be in control and the closest I will ever get to automation is writing a shell script, and this is only because I'm lazy and don't want to do the same thing 20 times.
I don't want to live in some dystopian future where AI knows everything about me and tracks me constantly. My iPhone has all of the location and other stuff disabled except for when I intentionally invoke it. When I come home, my phone goes into a corner in airplane mode.
Smeg-head.
look, companies have started just calling normal database automatic data processing programs as AI. like, it's so bad it has gone down to calling tiller databases ai. it's stupid.
you make a script to notify you that the disk is going to be full soon? congratulations you're now an AI developer in 2019.
And the fucking media is gobbling it up. Anyhow, 99% of the rise of "AI".
I mean, fuck, it's gone to the point where excel sheets are essentially said to be "AI" now. An access database from 1998 would be "deep learning".
like the fuck.. anyhow, there have been actual advantages in image/shape matching. but even then, if they really had that down _pat_ then tell me this: Why the fuck can't I have automatic OCR text translation for Thai writing? like, surely you could just automate the "ai" learning for it with a bunch of fonts? surely? if you can teach an "ai" to try enough combinations to win go and guess which trees to follow, can't you just make something with at least some commercial appeal and make it read thai?
world was created 5 seconds before this post as it is.
Further - "The job of an assistant is to know what i want in advance and just do it for me"
What a bunch of functionally challenged socially isolated managers. They're too important to drive the car even though they're the only one in it. Aaaaawwwww shucks.
The fundamental problem here is that they truly think they're selling the movie prop that dovetails directly into their pre-existing methodology of divesting all the responsibility and effort to 3rd parties but claiming all the credit. They actually don't know they're whiny selfish children and would be offended if it was pointed out.
so you're saying that the point you made wasn't as self evident as you were intending, you got mildly mocked for an obviously skewed comparison and now you're upset?
I suspect you've misunderstood the argument here :'D
Post earlier in the evening.
It little behooves the best of us to comment on the rest of us.
I bought an Echo Plus and am discovering the limitations - not just Alexa but the whole "smart home" ecosystem. One has to think like an architect while purchasing devices. First the devices are expensive, second the don't talk to each other, and third Alexa has several limitations regarding Routines. Which devices talk to each other, which ones will I control via the vendor hub vs Echo - including the Use Case and whether it will every fully work in a Routine.
For example - I have a group of lights (Philips Hue, without the Hue Hub). I can say "Alexa - set GroupA to 50%" and it works. But in a routine, one can't set the brightness of a group - only on individual devices.
Okay - let me add a light switch to that mix. Alexa doesn't talk to light switches. The Hue Hub does ("alexa buy me a $60 hue hub"). But I have "legacy" lights and am considering the purchase of 120V smart-switches. Wait - they don't talk to Hue. "Alexa - Vacation mode" requires me to stitch together Scenes on the Hue and control these switches from Alexa routines. And there's more. Alexa doesn't support Sunrise/Sunset (but Hue does). So Hue can turn lights on/off following the sun. Alexa can't drive this nor participate - so those "non-Hue" switches can't fully participate in Vacation mode.
I feel like I'm writing PROGRAMs as I setup routines. and I would REALLY like to have something like Switch instead to do it. The Routine UI is a pita.
And to top it off - Alexa doesn't let you Name routines. They are call named (by Alexa) "Every day at 6pm" or "Weekdays at 8 am" etc. and you can have multiples that trigger at the same time - therefore all have the same name. Now, what was I doing at 5:30 vs 6pm?
I can't imaging setting up a routine via "Hey Alexa - set up a routine..." The hunch stuff would be horrible. .. and if the power goes out -- all the Hue equipment boots up in the ON position. thank you.