Provided you do not ask Luis to book tickets to the MIddle East
Based on MSFT record of lately, it may book you one way tickets to Falluja or Tirkrit if you ask it to buy tickets to Riyadh.
Microsoft has pushed out a C# software development kit (SDK) for its in-beta language parsing API, LUIS. LUIS – the Language Understanding Intelligent Service – is another chunk of the chatbot capability Redmond is so keen on. It's a model-making environment which Microsoft reckons helps developers teach existing apps to …
A standalone thing or a pointless API to use Cortana?
I don't want to use so called "Cloud" voice control / response EVER. Alexis/Echo/Amazon, Apple Siri, MS Cortana, Samsung TVs, Barbie Dolls, etc are EVIL gatherers of private info. We had phones, PCs, car Radios etc with working voice control/recognition built in previously. These privacy theft, corporate services are damaging and setting back speech user interface development.
While there are arguments 'cloud' processed improves recognition without recourse to time consuming 'training' wizards to teach it how you speak, what I've found with the Google version is that it's not a trade off that works for me.
What does that leave as an excuse? Applying improvement metrics gathered from people using the service to improve the service. Sorry, I'm not 'people' and such a improvement process does little to improve the 'personal' in personal assistant in my opinion.
I also think a speech recognition service that requires online to work is of limited use as a accessibility feature (I don't know, is there a fallback?). It's not even as if all Cortana functions are available to everyone globally from what I've heard either.
Most people that have asked me about speech recognition over the last ten years quickly lose interest when they find they might have to train it up a little, which is one reason even those who know about the cloud privacy implications don't care.
LUIS parses a given sentence - supplied as text (from a chat client, Speech-to-Text functionality, or otherwise prepared and delivered) - and returns an intent and entities (eg assuming the developer has built a model for a music player then "I want to listen to Sade" would give me an intent of "PlaySong" and an entity of "Sade") - it leverages both specific domain training data (supplied by the developer in the form of annotated phrases) and then overlays that on the general Corpus that Microsoft have build for the language understanding models. The models can be a little more sophisticated (eg prompting for missing required items) but are fairly simple to train and interact with.
The developers application may know about you, but your interaction with LUIS is anonymised via the proxy that is the app you're interacting with (though depending on how that's architected there may be correlation possible via IP address)
The only intelligent response to that question is to ask "Which Paris?" I'm pretty sure that wouldn't be difficult for an AI to figure out if its database of places is good enough. If the DB is not good enough then that's a human failing, not an AI one. Sorry, your test of human redundancy is insufficient.
Anyway, if humans are redundant (presumably non-existent, or on the way there) then the question itself is redundant since there would be no one to ask it and no one to go there anyway. Indeed, a language parsing AI would itself be redundant because I'm pretty sure AI's would have better ways of talking to each other than verbal syntactical language.
Am I taking this too seriously?
A human wouldn't be able to make any more sense of Paris Texas than an AI would, or would be likely to make the same assumptions as the AI. The questioner would need to be more specific for anything to make sense of the question. Its just a database/knowledge thing. - does the human/AI know there's a film called that. I do, and I think its a doozing film, but the film is getting on a bit now and many people don't know its there. I'd expect an AI with a good general database to do better than a human in this case, so are we redundant? An AI can book tickets but it takes a human to enjoy the film or place. So, not yet.
why oh why MUST they only provide an API in C-pound???
it's like, couldn't they do something simple, like a C language API? [it's how they USED to do things, back when Win32 API was the way to go]
unfortunately, this is a trend that many of us have seen coming: the beginning of the end of the Win32 API, where ONLY "the METRO" or Universal [CR]apps can be written for Win-10-nic, almost like for WinRT boxen. They want this. they don't want Win32 API. It hurts them to see us STILL using it for SERIOUS applications. They want everyone using C-pound so they can CONTROL us and hook us into "their way" of doing things, like an evil company selling addictive substances.
yeah, well, tinfoil hat nonetheless, and black helicopters, too.
[don't say I didn't warn you]
C-SHARP, not POUND, is just a language, not the finished API. It compiles to something binary, or semi binary if its .NET. The API will be something binary too with a well defined interface, which isn't necessarily simple. The Win32 API is not simple and all other libraries such as .NET are still largely built on top of it. If you can write a C# program to talk to the API then surely you could just as well use something else too. It sounds like they are providing some C# templates or something (I haven't investigated in depth coz I have no intention of using this thing) to ease dev work, but there's nothing to stop you writing your own libraries and use your own languages to access the API. Even C++ can be used in a .NET environment.
Biting the hand that feeds IT © 1998–2019