Voice assistants failed because they serve their makers more than they help users
- Reference: 1671006665
- News link: https://www.theregister.co.uk/2022/12/14/voice_assistants_failed/
- Source link:
We know that the physical world, filled with exceptions and one-off events, continuously confounds even the most able humans. Expecting more out of a piece of software seems – from 2022's jaded point of view - a tragic example of believing your own hype. Robo-maids and self-driving vehicles fall over when they encounter the real world. Anything else would be either blind luck or black magic.
We had every reason to expect better outcomes in the purely symbolic worlds of information, knowledge and communication. The big three personal voice assistants – Amazon Alexa, Google Assistant and Siri – have been with us for the better part of a decade, trained by billions (possibly trillions) of tiny interactions, each of them improving language models that must now be the equal of anything developed anywhere.
[1]
Despite this, voice assistants have proven [2]almost completely pointless . Amazon recently admitted to [3]losing billions of dollars a year on Alexa hardware that promises much and delivers disappointingly little. Ditto Google. Apple has risked less on Siri, but even its ambitions for a HomePod speaker – with embedded Siri – fell far short of its imagined potential.
[4]
[5]
Ask anyone how they use these assistants for (everyone with a smartphone has access to at least one of them) and the answer almost invariably boils down to one of two tasks: playing music, or setting a timer.
While useful, these activities are such trivial uses of voice assistants you have to wonder why anyone ever bothered – and, in the face of such overwhelming indifference, why vendors continue to throw billions at their development.
[6]
Why haven't all voice assistants joined the vast graveyard of "tried it, didn't quite work out" technologies, like pen computing, the semantic web, or [7]blockchain ?
The answer is simple: voice assistants were never designed to serve user needs. The users of voice assistants aren't its customers – they're the product.
Back at the 2019 Consumer Electronics Show, I took a brief tour through a pavilion featuring Amazon's "Alexa Home" – brief, because with all those devices listening, waiting for a word on which to pounce, the pavilion felt more like a holding cell at the Ministry of Love than the home of the future. All of those microphones were waiting for the opportunity to harvest data – on activity or need or desire – so that this information could be on sold, analyzed, translated into some form of near-subliminal nudges, then fed back to the user. The whole system had been carefully thought out as a closed loop of feedbacks, driving users into patterns of consumption designed to benefit Amazon.
[8]
Even though most users had no conscious awareness of this architecture, they somehow instinctively rejected it. No one uses voice assistants as their makers had intended – everyone engages them at the bare minimum that provides some level of automation, without surrendering agency. How we did this – collectively and unconsciously – must puzzle the makers of these gadgets, who obviously believed we'd be wide open to the wonder of AI.
They believed this because those assistants had been designed as mimics of the [9]Knowledge Navigator – the Ur-demo of personal digital assistants, dreamed up by Apple's research team thirty years before it became possible. Using a conversational interface, the Knowledge Navigator helped its users find what they needed to find, learn what they needed to learn, and do what they needed to do. It gave them added agency in every situation.
That dream did not survive the rise of the internet as an advertising-revenue-driven medium. Agency lost out to the agencies, learning gave way to profiling, and finding became restricted to commercially advantageous search results. Cut to size on the Procrustean Bed of Late Capitalism, the heirs to the Knowledge Navigator promised much – but could only ever serve their masters, not their users.
Playing with the exciting and terrifying [10]ChatGPT , I've recovered my sense of what a Knowledge Navigator should be like. It's friendly, knowledgeable, and works toward my own ends. Yet someone has already pointed out that, because ChatGPT has been released for free, it's free to harvest all of our interactions: learning, refining its models, and – perhaps – creating the successor generation to these failed voice assistants.
Something smart enough to be engaging and friendly and helpful, while at the same time profiling, analyzing and nudging – that may be our future. Will we instinctively resist these new offerings as they serve us up to their masters? ®
Get our [11]Tech Resources
[1] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2Y5ms0s776Mq8H-xAkrN8rgAAAMI&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[2] https://www.theregister.com/2022/11/23/voice_assistants_fail/
[3] https://www.businessinsider.com/amazon-alexa-job-layoffs-rise-and-fall-2022-11
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Y5ms0s776Mq8H-xAkrN8rgAAAMI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Y5ms0s776Mq8H-xAkrN8rgAAAMI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44Y5ms0s776Mq8H-xAkrN8rgAAAMI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[7] https://www.theregister.com/2022/04/01/forrester_web3_criticism/
[8] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33Y5ms0s776Mq8H-xAkrN8rgAAAMI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[9] https://en.wikipedia.org/wiki/Knowledge%20Navigator
[10] https://chat.openai.com/chat
[11] https://whitepapers.theregister.com/
Re: "... they serve their makers more than they help users"
The idea that a middle man exists for the end user's benefit is patent nonsense.
No it isn't.
Amazon specifically won their massive market share by being very, very good for consumers. They were often a bit cheaper than shops, but not the cheapest prices on the internet. But they beat the shops because you don't have to go out to get your stuff, and they beat the cheaper websites because they had a vast range all in one place and really good customer service.
They've become bigger, and their site more user-hostile in my opinion. Search on the site seems designed to show you lots of stuff you probably didn't want in order to merge the original excellent Amazon giant web store with some sort of ebay/Etsy type thing - but the customer service remains good.
Amazon as a middle man are providing a service to several different sets of clients. Running a web shop at scale isn't that easy, so if you're a manufacturer you may prefer to just concentrate on making stuff. Rather than also doing the messy bits of retail. Customer service is expensive.
It's a bit like the aritcle author's dig at "late capitalism". Who says it's ending anytime soon? Captitalism isn't how the economy works anyway - it's a description of who owns stuff. The economy is a mixed market economy with varying levels of government intervention in different sectors - and so far nothing better has come along.
Perhaps voice assistants have so-far failed because of evil capitalist running-dogs? Or perhaps it's just because voice isn't a very good interface for anything at all complex.
Plus its an integration issue. The makers of the voice kit are all internet companies. But the stuff we want to control is all made by industrial conglomerates - and it takes time for the two to get together and get organised. Which is a similar reason why few of us have home-automation - a lot of the tech has been around for decades - and became very cheap in the last one but that doesn't solve all the practical issues of getting it installed and integrated. Or even just getting people used to the idea.
Re: "... they serve their makers more than they help users"
It is not from the benevolence of the butcher, the brewer, or the baker, that we expect our dinner, but from their regard to their own interest.
Lights, FireTV and heating
Ours are used almost exclusively to turn on plugs (Alexa, turn on the living room lights) or to ask to rewind the firestick by 30 seconds because the wife talked through an important bit, or to ask her (Alexa) to turn on the heating. Oh and the drop in feature is handy for asking the kids to come down for dinner.
For this, we find it very useful. Maybe we’ve got lazy but it would be pain not to have it nowadays and have to manually turn on a switch!
Can see how this is of zero benefit to Amazon though
Re: Lights, FireTV and heating
I think the most use our ever got was my children asking it to make animal noises. This then descended into the inevitable burps and farts, which even the kids got bored of this eventually.
Fundamentally I think they fail because there's no reasoning with them. There's no prior knowledge in the conversation of anything you've previously asked. It makes the same mistakes over and over again, and every instruction is a new separate command.
One thing I think they're undervalued as is as a basic radio - at least once you've found a station you want there's no more needing to arguing with it.
Re: Lights, FireTV and heating
"Oh and the drop in feature is handy for asking the kids to come down for dinner."
My dad had a similar feature available to him twenty-odd years ago. We called it "bellowing". I can appreciate it - it's strangely satisfying having a good excuse to try and make the whole house rattle with your voice.
Don't know.
My sister is an electronics gadget freak, you know the type, smart 'phone, smart watch, laptop, firestick, tablets galore, home theatre speaker setup, game consoles and so on.
She did buy an Amazon Echo and used it for a few weeks, just getting the weather, playing a bit of music and trivial stuff like that. Then I saw that the Echo had been unplugged and now it has disappeared into a drawer somewhere. Unused, unloved and a complete waste of money.
If someone like my sister has turned against this sort of device when I would say that she was the ideal target for such things than there must be something about them that just turns people off. I have no idea what that would be but it seems as if the makers of these things do not either.
Re: Don't know.
Speech isn't a great interface. It's fine for simple on-off commands. But then we don't have much home automation. Also what's the point of telling the washing machine or dishwasher to turn on by voice, when you're already standing right next to it to press the button, having just loaded it. Plus you need to select which program to use (probably only 1 of 2 of the 30 on offer...)
Asking Alexa for the weather is dead easy. But what about the answer? If you're in LA the answer might be hot and blue skies all day plus peak temperature. But if you're in Britain you need to know that it will rain this morning and be sunny this afternoon, with a chance of showers, and a tiny chance of snow overnight. That's much easier to see as a little graph. So you could buy the Amazon Alexa Show (or whatever it's now called) which is a home speaker with a little screen. But why not just get your phone out.
Again, "Alexa turn on the heating" - easy. Alexa, program the heating to come on at 6am week days, 7am weekends with the thermostat set to 21°. On week days turn it off again at 8, at weekends lower it to 18° after 9 then...
I strongly doubt that voice assistants are sitting unused because people are instinctively resisting profiling. People don't seem to be instintively resisting profiling when using Google, or Facebook, or indeed anything else.
Rather, I think the problem is simply that they just don't have that many uses. Most of the use cases that their developers thought out can be accomplished both faster and more reliably with a screen and keyboard.
Look at what the two common voice assistant tasks are. "Play music" and "Set a timer" are requests that are useful when I'm busy doing something else away from my device. I.e., they are tasks not worthy of my primary attention. Also, they are both tasks where if the voice assistant screws up, I can immediately tell (when I hear the first few notes, or when it reads back the timer wrong) and I've lost nothing - not even the five seconds enunciating the query, because, as mentioned, I was doing something else anyway. There just aren't that many tasks with that profile.
Why would I ask Assistant to search Google for something, when I can get the results faster by typing the query AND without wasting time repeating it three times until the voice recognition gets it right?
Why would I ask Alexa to buy something on Amazon, when I can search Amazon faster on the website AND without risking ordering the wrong thing by mistake?
I bet that the reason Apple didn't actually make the Knowledge Navigator, and the reason they didn't bet the farm on Siri, is that if there's one thing Apple is good at, it's UX engineering. They understand that if you give users a really cool way to do in ten seconds something that can be done in five seconds in the boring old way, they'll switch back to the boring old way in an hour.
What voice assistants need in order to actually get used is to be like a real live PA. Someone who, when asked to Google something, will not just Google what I said and read the first hit, but also know from previous context what I'm really looking for, then read through the query results, refine the query and exclude irrelevant hits, collate the information and present it to me in a synthetic fashion. Someone who, when asked to buy a gizmo, will not just search Amazon and order the first hit for 'gizmo', but know the specific details I want, search results to find a gizmo that matches them, do price-comparison, read reviews and figure out which ones are reliable, and pick the most convenient shipping option based on my schedule.
ChatGPT looks like it could do that at first glance... but it can't. First of all, a model like ChatGPT is trained once and does not learn further after training. Sure, it can tell you about WW2, but go ask it about something that happened the day before yesterday; it won't know. And retraining is computationally extremely expensive.
Secondly, ChatGPT always sounds convincing, but is actually wrong a lot. Fine for a laugh, but I wouldn't trust it to do anything that I actually want to get right.
In short, voice assistants are a no-starter until we actually have strong AI. I'm not holding my breath.
Alexa, create 10 El Reg comment accounts and upvote the above post from each.
The same applies to tv series and games
Series exist to make the creators money, NOT to tell a story. This is evident when they set out with no end in sight and either run out of budget and have to rush a conclusion (if you get one), or flop so hard at the end because there was no overarching story. I started watching Foundation and felt intellectually insulted by the end of the season*, it was so frustrating. Walking dead was padded so hard with filler episodes. etc
Started playing Death Stranding and near the beginning during a cut scene three Monster energy drinks are on a table - in some futuristic post apocalyptic society?? I can't un-see it and it pulls me straight out of the fantasy to remind me that the purpose of this game is to make money, not to tell a story (even though its been paid for!?). Again, felt insulting.
It is so hard to find tech that is geared towards the users' benefit, even if paid for. I wanted to enjoy this stuff so hard
Books are still a refuge
* it was brilliant till about episode 8 when the writers suddenly went 'oh shit, we need to wrap this up' and suddenly everything that happened before was irrelevant. They closed off all the tension built to that point and now have a nice clean slate for season 2. Rather than telling a story it fell into the seasonal story arc theme which fits around budgeting, not storytelling. Having not read Foundation I don't know what happens and whilst this may follow the books, it was a shit way to experience it.
Re: The same applies to tv series and games
I have some bad news I'm afraid, the Foundation books and Foundation TV series are only really related by the title and the names of some of the characters.
Re: The same applies to tv series and games
... which is OK, because whether there is another foundations series or not, is really irrelevant. I had the same impression watching series 1, only worse, because I had read the books (ages ago). I was impressed with the series... 'watchable' aspect, i.e. settings, landscapes, etc., the acting was quite professional (most of them tried hard and respect for that), the plot was... I never really bothered to match it to what I vaguely remembered from the books, and I knew it would be shit simply because the show must go on, i.e. several loose ends to continue in series 2, should the decision-makers make the decision. This 'plotting' is so evident in tv series, that it puts me off letting be captured, as I do with books. With books (other than serialized crap) they try to sell you a story. With tv series, they try to sell you drugs and see if you get hooked so they can keep selling more of the same. Fuck them.
Re: The same applies to tv series and games
Books are still a refuge
Sadly writers of books have adopted the same technique. I've lost count of series where you can see exactly when the author realised (s)he was on a winner, and started churning out long books of episodic tales which didn't advance the story, and left all the characters neatly lined up at the end for the next sequel. Robert Jordan and Terry Brooks are two that come to mind.
My voice assistant doesn't understand me....
Like the author of the article I can see no real pont in "Hey Google", "OK Siri" or whatever you have to use to wake up the Microsoft offering - for anything other than 'ok google, turn on the christmas lights'.
The voice recognition (for me) is fairly inaccurate and so I spend more time yelling at the assisant for failing to carry out a verbal instruction, that even my 3 year old would understand, than I would opening an app and doing the same thing using my fingers.
Re: My voice assistant doesn't understand me....
It is traditional at this point in a forum chat to add the following link
https://www.youtube.com/watch?v=NMS2VnDveP8
Bandwidth
> voice assistants were never designed to serve user needs. The users of voice assistants aren't its customers – they're the product.
While I enjoy a bit of tinfoil-hattery as much as every other paranoid nervous wreck, I think the actual answer is much simpler.
Voice interfaces are too slow. To compound that problem, they are inherently sequential - even worse than video.
An average person listens (for example to an audio book) at about 150 words a minute. 5 letters to the average word and that's 100 bits per second.
People read at nearly double that speed and even better, we can skip all the boring (or irrelevant) parts.
The one benefit that audio does have is that it does not require the user to be focused on the task in the same way as looking at a page or screen does. Hence it is handy for alerts (which could well be what hearing was evolved for) and for giving short, precise, commands. But once you get more than one person speaking at a time, while people can filter that - albeit with some effort - it seems that the machines cannot.
nonsense
So voice assistants where mysteriously and instinctively rejected? Sorry to prick this bubble, but there is no mystery in that, and it has nothing to do with our instinctive concern for our privacy - the same people are cheerfully using facebook and such which are a much bigger privacy sinkhole. The truth is that voice assistants are rejected because they fail dismally at all but simplest of tasks. The best they can do is, instead of learning to understand your language, to train YOU to say specific phrases to get certain results - just like using a programming language with a fixed syntax. As PAs they are complete and utter crap.
Oh, and disabling google apps' access to the microphone is a very good idea.
Re: nonsense
" train YOU to say specific phrases to get certain results - just like using a programming language with a fixed syntax
And just like the "new UI" which forces you to click on a new and unfamiliar set of icons concealed on unmarked pop-up lists at random points on the screen.
All these issues are inevitable outcomes of handing engineering problems with ergonomic considerations to self-congratulatory whizz kids who know nothing about either engineering or ergonomics.
This is a bad take
How exactly does Apple use Siri data? Why lump them in with Amazon and Google?
Ad company and retail company create services to serve their ad and retail businesses. Film at 11.
They don't do Scottish accents
Mandatory viewing
https://www.youtube.com/watch?v=HbDnxzrbxn4
Not trusty
Something smart enough to be engaging and friendly and helpful, while at the same time profiling, analyzing and nudging ...
I'm sure there are many different reasons for people to drop or not use voice assistants, but that quote must account for a lot. Replace "some thing " with "some one ", and you have the sort of person that you avoid once you realise that they are trying to manipulate or con you.
If a machine pretending to be a human is to be successful, it should at least behave like the sort of human that people can trust - one that doesn't have an agenda, and understands the meaning of confidentiality.
Re: Not trusty
like the sort of human that people can trust - one that doesn't have an agenda, and understands the meaning of confidentiality
Would you expect to find such people at Apple or Google?
"finding became restricted to commercially advantageous search results"
Sad that all challengers to the search crown, a tarnished and tawdry bauble, have followed the same "screw the end user" design as Google
The real reason why voice activated is rubbish
I can so believe this:-
https://www.youtube.com/watch?v=nwPtcqcqz00
https://www.youtube.com/watch?v=ZzSzkAuKPe0
and never forgetting
https://www.youtube.com/watch?v=HbDnxzrbxn4
Last night I was dwelling on the very relevant topic of using something as 'intelligent' as Chat.GPT in Alexa or Siri or Google. Yeah, it would be great to be able to actually use these assistants in a natural, non-consistent way, asking questions, getting things turned off or on, or even providing ideas and inspiration.
One of the problems with that as I see it is complexity. I'm sure it takes a lot more resources to keep something the equivalent of chat.GPT running for every iPhone user than to keep Siri running. And it's a lot less predictable. I've seen too many examples of chat.GPT being totally and utterly inept (ask it what 5+5 is, and tell it, "no, it's not 10, it's 11" and see how weird things get) and just plain wrong in its answers. Sure, it's right *most* of the time, but when it's wrong, it makes you wonder how these things are going to work if they were ever in control of anything.
I use my Alexa mainly to ask for the weather, set timers, turn lights off and on, add to a shopping list, and ask it how old actors are when we are watching a film. I use Siri mainly for setting alarms and timers, checking the weather and as a dictation device for sending quick messages to people. These are simple applications that are predictable. It'll either know or not know an answer. It'll possibly mishear an instruction. But It's not going to accidentally drive 300 miles to a post office when I asked it to nip down the road and buy some stamps, because it can't do that sort of thing yet. I'll be scared when it can.
Voice control
I had rudimentary voice controls (which basically amounted to 'call the number of this contact') on my phone (not even a smartphone) close to 2 decades ago. The convenience of being able to do some things just by voice, is a very interesting and useful feature. However 2 things to note - (1) all the voice processing was done locally. (2) It wasn't 'always-on' listening, you had to hold a button to activate it.
It could only recognise a couple of other commands beyond 'call' and a name to match against whatever was in the address book (which was probably limited to a couple of hundred entries and I'm pretty sure the length of the name was also limited), but then again it was at least a million times less powerful than a modern smartphone. It's 100% certain that a modern device can be built that can locally process and act on all manner of useful commands without sending any data anywhere. Yet nobody is building one, because they can't compete against the highly subsidised Siri, Alexa, Google.
Good to know that people are rejecting these spies en masse
I had only a few uses, both in-car ones.
Call a number, refuse an incoming call, have an SMS read, reply to an SMS. All of the worked well on my Nokia with Windows Phone. Each time an SMS came in, the phone politely asked if I wanted it read, and it could get and send an answer.
Since I switched to Android, this powerful OS can't perform easily such tasks. Android Auto is a joke, and now wants a compatible car. Google assistant works so badly I now looks for numbers with wheel controls and wait to read SMS.
My TomTom does a better job with its limited set of voice commands - limited but useful enough. Probably because it is interested only in tracking my trips, and not everything else.
Voice assistants failed because
I can't re-map the button to something even remotely useful (despite trying various software, etc). But hey, was it not a fantastic trap for milions of idiots, who have spent countless hours, years of their own time, of their own free will, to feed free information back to OS giants, to various government bodies and, naturally, to handset makers. And to their carefully selected business partners. And their business partners' partners' partners.
Has anyone tried ...
having Siri to chat to Alexa to GoogleAssistant
while Chat.GPT riffs off the results
Google Home has got worse since 2020
I have a brace of Google Homes that I mostly use as smart speakers, timers, lighting controllers and occasionally for answering questions when neither I nor Mrs Chimp wants to get their phone out. They're certainly useful when your hands are full or you need to set a timer with hands too wet or dirty for a phone, but for about two years they've been getting worse, not better. Voice commands are misunderstood more frequently, with a particular case that there's a 50/50 chance of what will happen to the lights when I ask Google to turn them "out" instead of "off". They might go off, they might stay on. This is not a system that's learning my speech patterns. My partner's Irish accent fares even worse.
I thought it might just be us, until I discovered the Google Home subreddit. It's full of users complaining that the performance of their devices has declined. Reports from the Alexa camp suggest that it's more responsive, but too frequently tries to sell you things based on your queries.
As for answering questions, it's fine if you want to know basic facts, but if you try to ask any sort of nested or conditional question, the best you can hope for is instructions to look at a link on your phone. Which is exactly what I don't want to do, or can't do, when I'm asking the question.
If Google can't make money off these things, then its history suggests that it will withdraw investment, performance will decline, users will abandon the product and they'll be consigned to the great Google Skip. Perhaps that journey has already begun.
"Procrustean Bed of Late Capitalism"
Nice!
"... they serve their makers more than they help users"
Of course they do -- that's what they're there for. The idea that a middle man exists for the end user's benefit is patent nonsense. It fundamentally exists to rake off a bit of the transaction value for itself.