As an AI, why must Alexa be so DUMB

[logan74k]

Quick rant but hopefully opens up a discussion here:

The Echo has great voice recognition in my experience - so why, in 2016, is Alexa so painfully simple?? 

For example, my Echo is tied to my Nest thermostat.  It's set up and works as it's meant to when I say "Alexa, set Nest to 72"   To my dismay however, in order for the system to work, I must use that exact phrase in the correct order detailing precisely what I want to happen. 

This is not AI.  It's a program with rigid input requirements and rigid output options.  It's got some initial 'wow' factor in that "I just turned on my TV with my voice!" sense, but the magic quickly falls away when you try your damnedest to make something happen multiple times, and keep getting, "HMM, I'M NOT QUITE SURE HOW TO HELP YOU WITH THAT"  ... check the app, yep, Alexa heard exactly what I said, but "turn the heat up" means nothing, because those exact words were not programmed specifically.  Really, Alexa?  "Turn on the fan" has no meaning?  You have nothing connected which has a fan?  "It's cold in here" - "Sorry, I can't find the answer to the question I heard"  SMH.

I'd just love to know why Alexa can't be programmed to understand various ways of saying things and use them all as input phrases to trigger a desired output.  It doesn't have to be truly smart, I just wish the program allowed for variance in what you say. 

[mike27oct]

You are correct, it is not true AI like IBM's Watson computer. but the Echo does not cost as much as Watson, either.  You get what you pay for, so get over it, learn to live with the limitations.  I have had mine for a couple of years, and in some ways it has gotten smarter, but it is still pretty darn rigid and dumb in many ways, but no complaints from me as I am getting what I paid for.

[logan74k]

That's awesome that you feel you've gotten a good value and the system has met your aspirations.

The issues with the Watson comparison I see are 1: Many facets of Watson are available to us online, free, now, and 2: Watson seems to be a completely different animal in terms of programming.  It'd be great, but Alexa doesn't need to decipher abstract links between disparate ideas on the fly, it just needs to be able to recognize when it's being asked to do something it is fully capable of doing.

Since you bring up the value angle I wonder: How much do you feel we must pay for a cloud based voice controller which works for someone who hasn't been "programmed" to use it?

[kevb]

Well, "it's cold in here" isn't a request, it's a statement. "Turn the heat up"... to what temperature? That said, it is possible, through IFTTT or a home automation hub, to allow Alexa to start events when those phrases are said. My wife often says "it's cold in here"- it doesn't mean the heat comes on.

[DParker]

Well, "it's cold in here" isn't a request, it's a statement. "Turn the heat up"... to what temperature? That said, it is possible, through IFTTT or a home automation hub, to allow Alexa to start events when those phrases are said. My wife often says "it's cold in here"- it doesn't mean the heat comes on.

You can also say, "Turn the temperature up/down" to increase/decrease  the thermostat setting by 2° F.

Alexa is, in fact, real AI.  It isn't human-level AI, but AI nonetheless.  That term encompasses much more than most people think.

[logan74k]

I would argue that when I'm directly addressing the 'bot that controls my home automation, "it's cold in here" is a request.  Even if I were talking to a human, that statement would be looking to open up a conversation or direct an action toward correcting the issue. 

Dparker I'm sure technically Alexa can be categorized as Artificial Intelligence, though I'd love to know what criteria it meets to do so.  Alexa certainly wouldn't be able to make sense of anything it's never heard before.  I'm blissfully unaware of the true complexity of programming a system like Alexa.  I can't claim to be as brainy as a top level silicon valley programmer by a long shot.  All that said, nearly everything I've gotten my echo to do can be explained with simple, text-based if/then sequences.  If programming were truly as basic as that, Alexa could be vastly improved simply by adding a bunch of phrases that mean the same thing to qualify as the trigger.  So I'm sure there's something in there that makes that really difficult, and that question is the root of my frustration here -  I just wish I could get a glimpse of what the real blockage is, programming wise.  Theories are welcome.

[strayfish]

I'd say it's ongoing. At a very basic level, you might find as I did that Hush and Shut Up work just as well as Stop, and I'm noticing other flexibilities that seem to be working their way into the system. Sometimes though, it seems inexplicably obtuse and needs the exact words or it won't action the instruction. Much like some people, in my experience

[Snafs]

One would feel that the use of understanding of key words, hot, cold, light, on, off, together with words of items, could be calculated upon and some mild understanding of the phrase based upon assembling key words, as opposed to the full sentence would be possible.

Alexa, it's dark in the kitchen = we have the place kitchen, and the phrase dark, so perhaps a reply would be "would you like me to turn the kitchen light on for you"

You may have a number of lights, you could just say, turn the light on. Now alexa knows you want a light on, but not which one, so she could ask "which light" so you reply "kitchen"

I'd not imagine this is (as they say) rocket science, just very basic assembly of simple words into actions.
But would avoid the need for you to remember exact phrases.

It would make a lot of difference, if you gave Alexa a command, she was not sure which one, or there were mutliple possibilities due to your phrase, so she spoke back with some options.

Like your Partner in RL.

Can you get my sweater.

A human would reply, which one do you want?

The one with the snowman on the front, you reply.

[strayfish]

Can you get my sweater.

A human would reply, which one do you want?

The one with the snowman on the front, you reply.

At the moment, it's the dogmatic human that says, 'Yes I can,' then does nothing until you ask them to 'please get my sweater'!

[Snafs]

In respect to the excellent posting above.
I do agree, in general that the 'ask for the right thing, in the right way and it works' argument is valid.

However, if you stop and think for a moment, long term, we will end up with these, so called "simple devices" being vastly harder to use than any other device.

Other devices, be they Washing machines, Radio's, Cars, Phones, pretty much anything have visual clues of the function we want.
We may only need a visual reminder sometimes, or to read just the 1st couple of words and recall what we need.

But to end up with literally thousands of commands you must, recall perfectly, without any visual clue, it going to be impossible for many, without some AI feedback being added into these systems to help us out.

Open the door = don't understand
Please open the door = don't understand
unlock the door = don't understand
open THIS door = don't understand
Please open THIS door = don't understand
Open my back door = don't understand
Unlock my back door = don't understand.
Unlock and open my back door = Ok. Door is now open.

Such a scenario, whilst seeming silly, is NOT the world we want to find ourselves in.
Esp is you have say 1000 such commands in years to time to memorise

Ideally you would say something, perhaps a little wrong, and the "device" will ask you back, to check what you mean.

We could end up where some just think, you know what, it's easier to just use the key, or press the switch than to remember all this.

[DParker]

But to end up with literally thousands of commands you must, recall perfectly...

But as I've already pointed out, that's a false premise.  All you need do is phrase your command in a clear and unambiguous manner, including information that common sense should tell you is required.  "Turn off the downstairs lights" or "Turn the downstairs lights off (or even just "Downstairs lights off") doesn't require me to memorize anything.  I want the downstairs lights to be turned off, and use of any one of the listed phrases (or even other variants I haven't thought of) should be intuitively obvious.

Granted, a clarification inquiry from Alexa in the event of an abiguous command is helpful, but she already does that in many cases.

[logan74k]

Open the door = don't understand
Please open the door = don't understand
unlock the door = don't understand
open THIS door = don't understand
Please open THIS door = don't understand
Open my back door = don't understand
Unlock my back door = don't understand.
Unlock and open my back door = Ok. Door is now open.

This is just what I mean, and what using Alexa has been like for me in more than a few instances.  I think the "ask for the right thing in the right way" argument is great, if 100% of Alexa users are young, bright, logic-minded people, possibly with a programming background, with time to learn precisely the phrases the program can respond to.  It would be worlds more convenient to get to a system that was programmed to work with us, rather than a system we need to be programmed to use.

Granted, there are a few allowances in word order that Alexa can accommodate.  But taking DParker's example above "Turn off the downstairs lights", you could sit there all day trying to say it different ways and nothing might happen, simply because Alexa can only fathom the downstairs lights as "main floor" lights.  Or maybe "basement" lights.  It's simple context like this that Alexa can't distinguish.  The person who set up the device might not always be the person trying to turn off the lights downstairs.

More clarifying questions would be an excellent start, but maybe the programmers or execs don't want Alexa seen to be asking 'dumb' questions all the time if they aren't confident the line of questioning the program is capable of would lead to the correct action.  Possibly it would just reveal more clearly how limited the system is in understanding.

[Snafs]

I have to follow up on what's just been mentioned.

It's very VERY easy for young/middle aged, even some senior people who still have their full faculties to be unable to place themselves in a position of someone else, who may not be so great.

Example, My neighbour is an elderly lady, who's daughter bought her an iPad. You know, the most simple device to use, tap on a icon and something happens.
She was totally and utterly confused by it, flashing colours, icons scrolling left and right, if you did not tap just right things did not work correctly.

It's way easier to just let an old person use a light switch than to try and explain to them just how to tell alexa to turn on the light as they will not remember the right phrase.

Really, the ideal would be getting near human responses, which I know we are a million miles away from.
Humans, after being asked a question, often ask for clarification to make sure they understood right. We do it all the time.

We generally do not give full exact instructions when we talk.

Honey, can you get me a beer from the fridge please.
Sure dear, which one?
A bud please.
Ok dear.

We after don't say, Wife, can you get me a bottle of Budweiser from the fridge.

Likewise....

Honey, can you bring me down a sweater when you come downstairs.
ok, dear, which one would you like.
The thick one please as it's cold outside
ok, will do.

I'd say we often throw out an initial request and leave out some detail that MAY need clarification.
If may not. You may only have one beer and one sweater.

Alexa, turn on the light please.
Sure, which one.
The dining room
Ok, sure.

[kevb]

"Alexa, turn on the light please.
Sure, which one.
The dining room
Ok, sure." 

Doesn't it do this now? If I say Echo, turn out the light, it responds "I found many devices that match that name. Which one did you want?" Basement light. "Ok"

[Celeritas670]

Some very intelligent comments here, on all sides.

Regarding the claim that Alexa is not Watson, but Watson costs much more than Alexa, consider this.  Alexa is cloud based.  Amazon only needs one Waston.  With millions of users, this could be very cost effective eventually.  It's coming, probably sooner than most of us think. 

Just imagine: "Alexa, construct and send a witty reply to the previous post."