[nemoaristos]

I was amazed to see the new AIML 2.0 proposed, an amazing upgrade, this will revolutionize A.I., allowing developers and hobbyists to make Siri and Google Now style A.I. from Pandorabots for all their mobile devices such as Smartphone and iPad etc.

Pandorabots is now yet again a world leader in A.I. - can it do a similar thing with R.O.S. Robot Operating System?

On the OOB Tags for mobile devices I have been wondering when a similar R.O.S. Tag system will come out for Pandorabots allowing robotics using R.O.S. to be immediately upgraded with voice command A.I. via Pandorabots - Basically Siri for R.O.S. although obviously Siri is a derivative of A.L.I.C.E.

A "bridge code" like the lines of code that links Second Life with Pandorabots, that links CallMom with Smartphones and Tablets - Can a similar bridging code be made quickly that allows current and future R.O.S. robots link up with Pandorabots to grant instant Artificial Intelligence to current and future R.O.S. robots?

Pandorabots could be a global leader in a Robotics revolution starting off from very simple rudimentary code to link R.O.S. and AIML on the cloud with Linux.

A basic Android app to allow any R.O.S. system to be patched into CallMom style Pandorabot for voice operated R.O.S. commands and to provide instant AI personality to any R.O.S. Robot.

[Bragi]

this will revolutionize A.I.
Pandorabots is now yet again a world leader in A.I.

Not really. AIMl is simply catching up with newer platforms, but it's still a lot behind.

[squarebear]

I disagree. Blame the artist not the palette.

Is the piano behind as a musical instrument because you can only play Twinkle Twinkle Little Star with one finger? Listen to a Chopin piano concerto to see what it can do.

[Bragi]

I'm not talking about how the platform is used, I'm referencing to the features it offers.
If you only have a hammer.... you can still make this:

But you're still limited to the things you make, for instance:

[squarebear]

From what I have seen of the complaints raised by people who have tried AIML, their issues can usually be solved if you code it correctly. I haven't come across a problem on these or other chatbot boards that I can't solve using AIML.

[Bragi]

So can you use AIML to perform statistical predictions or recognize visual patterns?

[squarebear]

I see no reason why not, if your AIML code is fed with the right data. In fact I use AIML to send me statistical data about the people who are talking to my bot, so I can see who is swearing at it, talking about sex and so on. No predications as such but I see no reason why it isn't possible.

Visualisation would be possible but it isn't my field. However, if you had data such as:

01110
10001
01110
10001
01110

It would certainly be possible for AIML to recognise this as a number 8. I'm not saying it is the holy grail but is certainly capable of a lot more than people give it credit.

[Bragi]

Perhaps you should take a look at Richard's graphmaster definition. It simply doesn't allow statistical approaches due to how duplicates are handled. For instance, you can't use AIML to build a bayesian filter.

Granted, for certain, very simplistic images, you could use AIML to do extact pattern matching.

[Bragi]

also forgot to mention: missing values is another reason why AIML can't do statistical matching.

[squarebear]

also forgot to mention: missing values is another reason why AIML can't do statistical matching.

What sort of values Jan? If you mean assigning values to keywords and adding them up to see if a message qualified as spam or swearing, yes that is certainly possible with AIML.

[Bragi]

No, what I mean is the following:
I can best explain this with the 'titanic' example. If you have the age, sex, location of boarding, passengers class, location of the cabins, $ paid per ticket (some 1st class passengers paid less then those in 3th class),... values of all the passengers on the Titanic, you can fairly accurately predict the survival rate for given set of values. But sometimes you don't know all the values, for instance, age or sex could be missing. Statistical decision trees are able to handle this gracefully and still give a fairly accurate result. Of course, the more values it needs to fill in, the less accurate things get.
Without modifying AIML's specification about the graph tree, this can't be done.

[squarebear]

Ah I see. AIML is designed as a chatbot language rather than a general all purpose artificial intelligence solution. I dare say you could assign values to the things you mentioned to work things out like that but that's not really what it was designed for.

Age < 30 = 10 points
Age > 29 = 5 points
cabin near lifeboat = 50 points
cabin in middle of ship = 0 points
ticket price > $1000 = 20 points
ticket price < $500 = 0 points

Add them all up. Anyone with a score of 40 or more is more likely to survive. This would be a fairly straightforward problem for AIML, assuming the values were accurate and I see no need to worry about the graph. Unless I am missing something, I see no problem with this task. As I say though, it's more designed for chatbot issues though than a general solve anything language.

[Bragi]

Yes, you could calculate some heuristics and build some AIML categories around that. But that's the whole point of the statistical approach: it calculates this for you, usually much more accurate than you can. All you need to do is feed it some historical data before doing your tests. Furthermore, in the example of the titanic, it's indeed easy to pre-calculate this, but often that's simply not possible. That's when the category approach breaks.  (ex: a suggestion system like amazon uses. Not only the size is a problem, but also the ever changing nature of the data).
And yes, you are correct, AIML is not designed to do this. That's exactly why I brought up the graph. With a few basic, simple alterations to the structure of the tree and the algorithm to calculate the result, you can do this with AIML.  In concrete, here's how to go from a pattern matcher like AIML's to a statistical decision tree. And here you can find some more info on missing values (somewhere in the middle of the article).