Echo is not really an AI

[jwlv]

From my own testing, it would seem that Echo isn't really an AI under the hood. It does not have natural language processing capabilities. It behaves more like pattern matching or regular expression matching. For example, I can say "Alexa, shama lama ding dong what's the weather?" Alexa would match the keywords "what" and "weather" and then give me the weather. Everything else I said is ignored.

This would also explain why it is very picky when it comes to smart devices. If you have devices named "kitchen light", "bedroom light", "master bedroom light", "garage light" and so on, they all become keywords. And the fact that all these keywords contain the word "light" would cause Alexa to ask "which device do you want" or it can't find the correct device at all.

Certain keywords are weighted more heavily than others. Some heavily weighted keywords are: weather, turn on, turn off, open, close, time, where, etc. 

Example: If I say "Alexa, close McDonalds" the response I would get is the location and hours of the nearest McDonalds. Alexa matched the keyword "close" so it thinks that I want to know when the store closes. However, if I say "open McDonalds", it thinks that the "open" keyword is to be used to start a skill rather than getting the location and hours of the nearest McDonalds. I bet if I were to name a device "weather light" Alexa wouldn't find it because "weather" is much more heavily weighted against any device names.

So in my opinion Alexa is dumb. It doesn't do any real natural language processing. But it does control smart devices quite well if you name your smart devices appropriately.

Google Home on the other hand has really good natural language processing. They've been doing it on their search engine and on Android for years. But support for smart devices is very lacking. They may catch up eventually.

[DParker]

For example, I can say "Alexa, shama lama ding dong what's the weather?" Alexa would match the keywords "what" and "weather" and then give me the weather. Everything else I said is ignored.

How is that substantially different from the way a human processes similar speech?  Your example contains a recognizable request for information, along with some irrelevant extra introductory verbiage that does not mask or alter what it is that you're asking for.  As a result the extra verbiage is ignored and the meaningful portion is acted upon.  Person-to-person communication  of this sort is exceedingly common.

This would also explain why it is very picky when it comes to smart devices. If you have devices named "kitchen light", "bedroom light", "master bedroom light", "garage light" and so on, they all become keywords. And the fact that all these keywords contain the word "light" would cause Alexa to ask "which device do you want" or it can't find the correct device at all.

Again, I don't see how this is an example of non-AI language processing.  If I'm assisting you with working on your car and you simply say, "Hand me a wrench"...while I'm looking at a tool box full of different-sized wrenches...I'm going to respond by asking, "What size?" unless I have some way of already knowing what size you need.  Similarly, if you have a dozen different Alexa-controllable lights in your home and she asks which one you're referring to when you say, "Alexa, turn on the light"...that's not an issue with weak language processing.  That's a problem with an ambiguous request on your part.  Every light in my home has the word "light" in its name, and yet Alexa has no trouble at all understanding what I want when my requests contain the same amount of clarifying information a human would expect from me.

Certain keywords are weighted more heavily than others. Some heavily weighted keywords are: weather, turn on, turn off, open, close, time, where, etc. 

Example: If I say "Alexa, close McDonalds" the response I would get is the location and hours of the nearest McDonalds. Alexa matched the keyword "close" so it thinks that I want to know when the store closes. However, if I say "open McDonalds", it thinks that the "open" keyword is to be used to start a skill rather than getting the location and hours of the nearest McDonalds. I bet if I were to name a device "weather light" Alexa wouldn't find it because "weather" is much more heavily weighted against any device names.

That might be true.  But it's still not all that different from the way humans process spoken language.  It's fairly normal for people to key in on certain parts of what you say and either ignore or mishear other parts based on their own priorities/biases regarding information.

So in my opinion Alexa is dumb. It doesn't do any real natural language processing. But it does control smart devices quite well if you name your smart devices appropriately.

Google Home on the other hand has really good natural language processing. They've been doing it on their search engine and on Android for years. But support for smart devices is very lacking. They may catch up eventually.

Do you read any GH forums, like the one over at SmartThings?  GH's language processing is definitely better than Alexa's, as is the search engine.  But the language processing difference is one of degree rather than of kind.  And in fact its ability to understand what you want when it comes to issuing device control commands is even worse than Alexa's, and requires that you be even more careful about naming things.

[jwlv]

To describe natural language processing (NLP) is rather complex.

This example has nothing to do with Alexa or Google, but serves as a simple example of how an AI NLP should handle these queries.
1. Is the window open?
2. Is Windows open?

A person would have no problems distinguishing the two questions.

A pattern matching algorithm (non-NLP) might see the keywords "window(s)" and "open". An NLP should be able to discern that when the user says "the window" they actually mean a physical window and not a computer running the Windows OS.

[DParker]

To describe natural language processing (NLP) is rather complex.

This example has nothing to do with Alexa or Google, but serves as a simple example of how an AI NLP should handle these queries.
1. Is the window open?
2. Is Windows open?

A person would have no problems distinguishing the two questions.

A pattern matching algorithm (non-NLP) might see the keywords "window(s)" and "open". An NLP should be able to discern that when the user says "the window" they actually mean a physical window and not a computer running the Windows OS.

Even that example doesn't work (FWIW, I was a subscriber to AI Expert magazine in the mid-'80s, so I'm not new to this topic).  Humans are easily able ti differentiate between the two phrases largely due to context clues, which could alter their interpretation.   For instance, you assume that someone would take, "Is the window open?" to refer to a physical window.  But that would not be true in the context of...say...a call between a user and tech support rep for a software app that runs on Windows.

Conversely, "Is windows open?" asked by someone who's English is poor and who is preparimg to paint a living room, would be easily interperatable as a question regarding the state of the room's ability to vent fumes.

Yes, those are somewhat strained examples, but they illustrate the point that human-like interpretation of speech is about a lot more than just analysis of words and syntax, and requires information that is just not available to something as comparatively simple and limited as Alexa and GH.  But that doesn't either of them non-AI.  After all, the "A" is for "Artificial".  What they're doing is simulating/approximating biological intelligence, not duplicating it.

[renegade600]

From my own testing, it would seem that Echo isn't really an AI under the hood. It does not have natural language processing capabilities. It behaves more like pattern matching or regular expression matching. For example, I can say "Alexa, shama lama ding dong what's the weather?" Alexa would match the keywords "what" and "weather" and then give me the weather. Everything else I said is ignored.

This would also explain why it is very picky when it comes to smart devices. If you have devices named "kitchen light", "bedroom light", "master bedroom light", "garage light" and so on, they all become keywords. And the fact that all these keywords contain the word "light" would cause Alexa to ask "which device do you want" or it can't find the correct device at all.

Certain keywords are weighted more heavily than others. Some heavily weighted keywords are: weather, turn on, turn off, open, close, time, where, etc. 

Example: If I say "Alexa, close McDonalds" the response I would get is the location and hours of the nearest McDonalds. Alexa matched the keyword "close" so it thinks that I want to know when the store closes. However, if I say "open McDonalds", it thinks that the "open" keyword is to be used to start a skill rather than getting the location and hours of the nearest McDonalds. I bet if I were to name a device "weather light" Alexa wouldn't find it because "weather" is much more heavily weighted against any device names.

So in my opinion Alexa is dumb. It doesn't do any real natural language processing. But it does control smart devices quite well if you name your smart devices appropriately.

Google Home on the other hand has really good natural language processing. They've been doing it on their search engine and on Android for years. But support for smart devices is very lacking. They may catch up eventually.

based on your definition, a human baby is not human

[jwlv]

based on your definition, a human baby is not human

By my definition, a human baby is not an AI.

[PaulKTF]

I don't care if it's A.I or not as long as it does what it's supposed to.

And it does.