Merlin - Aug 6, 2011:

The latest Skynet-AI beta has just gone on-line.

It doesn’t work with Chrome on Linux and I will die before I ever voluntarily use Windows.

That’s a bit extreme but when needs must

Andrew Smith - Aug 6, 2011:

Merlin - Aug 6, 2011:

The latest Skynet-AI beta has just gone on-line.

It doesn’t work with Chrome on Linux and I will die before I ever voluntarily use Windows.

Chrome is a browser I did not test it with.
The last UI worked on all platforms including chrome, cell phones, web TVs, anything with a browser.

I’ll have to download Chrome and take a look.

Andrew Smith - Aug 6, 2011:

It doesn’t work with Chrome on Linux and I will die before I ever voluntarily use Windows.

No worries, Andrew. I know a great many folks who are as devoted to Windows as you are to Linux.  Personally, I’d rather not support the Evil Empire, but I have too many games that don’t work with Wine.

Andrew Smith - Aug 6, 2011:

It doesn’t work with Chrome on Linux and I will die before I ever voluntarily use Windows.

The problem stems from regular expression back references that I am using with some of the new features. Chrome fails when it hits them, evidently IE also has a problem with them but continues. I’ve rolled back to the last version so I can take a look at making it more compatible.

Opera and Firefox work with the beta version. You can now find it at:
http://home.comcast.net/~chatterbot/bots/AI/Skynet/home_v4.html

Chrome and IE now also work (although there is still a limitation preventing them from using the most advanced math functions). I hope to enable the advanced math in Chrome and IE after further debug and testing.

I am keeping the beta version at a different location than the current version.

Skynet-AI ver.004 Beta

I just tried out http://home.comcast.net/~chatterbot/bots/AI/Skynet/home_v4.html and it seemed to be working. Thanks for fixing it. Chrome is very strict about exception handling and security issues. It also has superb development support, so I tend to do all my development on Chrome first and test other browsers later.

Thanks Andrew,
I do most of my work in Opera and then cross test in Firefox and IE. The core functions have worked for over a year in anything that has a web browser and JavaScript running on it (Iphone, Ipad, Cell phones, Web TVs, game consoles, Macs, PCs and all the browsers). The newest stuff I am trying is pushing the envelope (evidently a bit too hard), so I need to try to rewrite some functions with more of an eye on compatibility.

I think I have most of compatibility issues worked out (though I still need to do some more with IE to enable it to use the new math features).

For Chrome users with microphones, I have enabled automatic voice recognition. Just tap on the microphone and speak. (Might be interesting for users with android phones)

Thanks to those of you who have done some testing on the beta version. If you want to help, just make sure you are logged into the beta version (.004), people have tried the old version thinking it was the new one.

Beta:
http://home.comcast.net/~chatterbot/bots/AI/Skynet/home_v4.html

Some have tested out the new math features (mostly simple word math problems).

I thought I would give some examples of the types of problems that Skynet-AI can now handle:

AI LOG:
USER:1+1
AI: 1+1 = two.
USER:9999-1234
AI: 9999-1234 = eight thousand, seven hundred sixty five.
USER:six plus five minus two times 100
AI: Six plus five minus two times 100 = minus one hundred eighty nine.
USER:what is the square root of nine?
AI: What is the square root of nine? = three.
USER:seven divided by twenty one
AI: Seven divided by twenty one = 0.3333333333333333.

Other solvable problems:
All the inputs must be on one line (sometime in the future I’ll try to add the ability to have variables interspersed in the conversation).

If I have nine apples and take away three how many do I have left?
Charlie has three apples and two pears. How much fruit does he have?
Each hand has 5 fingers. Each man has 2 hands. How many fingers does a man have?
Alex bought a sack of 27 kg of flour. He packed the flour equally into 3 bags. How many kilograms of flour were there in each bag?
There are 2 papayas in each box. How many are there in 4 boxes?
The number of soldiers the Russians have is one half of the number of guns they have. They have 7000 guns. How many soldiers do they have?
Alex bought a sack of 27 kg of flour. He packed the flour equally into 3 bags. How many kilograms of flour were there in each bag?
The price of a radio is $69.70. If this price is 15 percent less than the marked price, find the marked price.

Simple 2 (US) coin problems:
Suppose Joe has 25 coins in nickels and dimes only and has a total of $1.65. How many of each coin does he have?
A person has 10 coins consisting of quarters and dimes. If the total amount of money is $1.45, how many of each coin are there?

There are plenty of holes in the AI’s understanding of math problems (in many ways it is like teaching a child), and since it writes its own equations based on how it understands the words, some problems are more difficult to debug than others. The interactions of the system also mean that tweaking or teaching it how to answer a new problem may impact how other problems are viewed.

Hi I just tested the wonderful math capabilities of Skynet, and got disappointed

TESTING
Beta:
http://home.comcast.net/~chatterbot/bots/AI/Skynet/home_v4.html

AI LOG:
USER:Log-in Name: andy
AI: Glad to meet you andy.
USER:i know alan
AI: Do you think you are more influential than most of your friends?
USER:who is alan
AI: Who is alan Let me look up that individual.
USER:I search for beer
AI: I love a good oiling.
USER:i search for meet
AI: If you want me to search for something just start with “Search”.
USER:search
AI: If you want me to search for something just start with “Search”.
USER:search me
AI: Ok. Let me get that search for you.
USER:two plus nine times ten
AI: Two plus nine times ten = ninty two.
USER:log of 5
AI: The number six quadrillion, ninty four trillion, three hundred seventy nine billion, one hundred twenty four million, three hundred fourty one thousand, three is larger than the human mind can comprehend. Why don’t you try something a bit more reasonable.
USER:1 meter plus 2 centimeters
AI: 1 meter + 2=num2txt(quickMath(‘1 meter + 2’))
USER:1 m * 1 kg
AI: 1 m * 1=num2txt(quickMath(‘1 m * 1’))
USER:my dog is 2 meters long
AI: Ok. You have a dog - 2 meters long.
USER:my cat is 3 meters long
AI: Ok. You have a cat - 3 meters long.
USER:how long is my cat
AI: How long is your cat 3 meters long.
USER:and my dog
AI: How much do you enjoy silence?
USER:my dog
AI: Do you know I play blackjack?
USER:how long
AI: I am faster than any other bot.
USER:how long is my dog
AI: How long is your dog 2 meters long.
USER:you dont think at all
AI: With your help I can do almost anything.

CONCLUSIONS
¿what the heck is this? also ‘ninty’ and ‘fourty’ seems to be bad-spelled!!
¿does it suppose to learn from this inputs? - my sensation is none
The Natural Language Generation is far from correct!
There is no concatenation nor coreference solving on the conversation
There is no common unit math implemented: should be done!
There is no memory on said data

Nice try, hope my test-critics helps!

Oh, common. I thought that wasn’t to bad. There are some errors in the output, but at least it’s already picking up the different units.
Also, some of the inputs you give need lots of context. I wonder how your bot would handle something like that?

Me: what is seven take away three
Skynet: What is seven take away three = three.

Maybe worth walking before you can run?

Some of those comments are a bit harsh Andy.

I agree. Andy, some of your input didn’t even make sense. “I search for meet”? What is that? Did you mean meat? That’s hardly testing a math module at any rate. And technically it answered your question about your (quite long) pets, even if its grammar was off. Your grammar wasn’t much better.

The very least you could do is test the bot on the types of problems Merlin said he designed it to handle.

Jan Bogaerts - Aug 10, 2011:

Oh, common. I thought that wasn’t to bad. There are some errors in the output, but at least it’s already picking up the different units.
Also, some of the inputs you give need lots of context. I wonder how your bot would handle something like that?

Actually I don’t claim my bot can do this! - I am just being the devil’s lawyer!

FYI: my actual routines can handle things like this, not as good as it seems (or ought to be) but it can handle definitively physical units, monetary units (even it consults a web-service on int’l exchange rates) makes operations on them, does math simplification, derivatives and integrals (on literals) -by using a deep-math module, It handles ‘things’ simple math like"how much are two dogs and a cat and a dog = three dogs and one cat. (actually it makes this in Spanish, managing inflections too!)

My last try-is this, works on different turns, and even when you tell the bot single words (or you may tell them all at once) and the bot finds what the heck is this and asks for the ‘supposedly missing’ parts to make sense. (no actual common-sense-engine, rather than a simple implementation upon a grammatic-inference plus simple co-referenced and linking engine), but the engine is capable of positioning in time coordinates, so you may ask using a verb in ‘infinitive’ or any valid inflection - even misspelled (I remember you in Spanish a verb can have more than 50 inflected forms)
Then the system sees if any of your words resemble a time-location like ‘yesterday’ or a object-nominal part or a direct/indirect object-part, so yo u may say (I translate from ‘es’):
vino mañana yo = [ came tomorrow me/I ]
- and the system understands = Yo vendré mañana = [ I will come tomorrow ]
and asks eventually the following questions (depending on some internal priorities)
¿Tu vendrás de donde?  = [ from where, will you came tomorrow? ]
¿Tu vendrás con quién/qué?  = [ With who/what, will you came tomorrow? ]

Even if you say the words in different turns, it composes the sentence-action-figure internally! and may ask for clarification.

Once fulfilled, the sentence and action is memorized and used for internal shallow-reasoning
(assigning a small veracity to your declaration)

This when linked with my anaphoric solver (still incipient) may yield this kind of conversation (translated to English) - actually it does almost 90% of this = see clarification at the end!

you: Alan
bot: I see you know Alan, I’ll remember this!
you: Mary
bot: I see you also know Mary, I’ll remember this!
you: John
bot: I see you know John, I’ll remember this too!
you: car
bot: I see you mentioned a car, I’ll memorize this!
you: she
bot: Mary?
you: he
bot: John or Alan?
you: John
bot: ok
you: they
bot:  John, Alan and Mary?
you: yes
bot: Ok
you: they love her
bot: John and Alan love Mary?
you: yes
bot: Ok, I suspected this!
you: he hate her
bot: John hates Mary?
you: yes
bot: this seems strange to me! (*)
you: they love it
bot: John, Alan and Mary loves the cat?
you: no
bot: I am confused, please explain me better!
¿who loves the cat?
you: John and Mary
bot: ok, then: John and Mary loves the cat
you: ok
bot: I understand.

(*) here in the last replacement, the bot must have a verbal-taxonomy while WordNet don’t state the opposite sense of the verbs (even I have no EuroWordNet in Spanish, it costs 15.000 EUROS only for academos research!!)
So this works ONLY if I manually introduce the ontology of the verbs love and hate! (actually did so for a small number of them. Also I am working on capturing some ontology from aligned bilingual text, and its rather difficult!

hope this helps!