I believe Merlin said that the advanced math capabilities are not yet implemented in the online version of Skynet.

As for the input “what is 2 add 2”, I would never phrase the question that way. I think the more common way (for Americans) to ask would be “what is 2 plus 2”. However, I can’t speak for other native speakers. (And I agree that the bot should ideally be able to handle the question the way you posed it.)

Not sure what you mean about American, the question is in perfect English.

In America, nobody would say “2 add 2.”  They say “2 plus 2”.  I believe Merlin is American.  Thus your question would be a good test for a bot written by an American as the “2 add 2” terminology would not have been encountered before.  As it would be a new concept stated that way, a bot capable of figuring out what that phrase means would be displaying flexibility.  A brittle AI agent would treat the phrase possibly as one string instead of say, “metaphorizing” what it already knows about encountering patterns like [ something] operator [something].

Ah I see, my apologies. Yes, in England “What is 2 add 2” would be perfectly acceptable. I assumed that as Merlin said, “I have been able to integrate the transformational logic into my test platform so that I can handle structured word problems:” and gave some examples that this part of the bot was working.

Grace will consider

“What is 2 add 2?”
“What is 2 plus 2?”
“What is 2 + 2?”

Even

“What is 2+2?”  (“2+2” as one “word”)

even

“What is 2 and 2?” (Yes, to a programmer that could be considered doing a Logical AND of 2 and 2 which is equal to 2, but average people sometimes use ‘and’ for ‘plus’ in this way).

for “What is 2 and 2? ” it can be told that “and” is synonym for “plus” (when considered in that context), and in other contexts, NOT a synonym.

Same concept as

“What is 2 - 2?”
“What is 2 minus 2?”
“What is 2 take away 2?”

Or

“What is 10 / 5 ?”
“What is 10 divided by 5?”
Or

“How many times can you put 5 into 10 ?”

Showing that, although the bot is based on grammar, there will be plenty of room for ‘fuzziness’

Actually, a future functionality will take into account if the user is a programmer, and then know to use the logical-AND meaning of ‘and’ and reply “2 and 2 = 2 if you mean the logical and”

Steve Worswick - Feb 1, 2011:

USER:what is 2 add 2?
AI: Let me search for an answer. Sometimes, it takes a couple of seconds. (Chatbot opens a webpage which defines the number 2 instead of giving the answer).

CR and Toby are correct I am American and some of my language bias comes from that. No one has ever tried the “add” expression in that way before. I have added “add” and now you can add using “add”.

Victor Shulist - Feb 1, 2011:

Grace will consider

“What is 2 add 2?”
“What is 2 plus 2?”
“What is 2 + 2?”
. . .

The on-line version of the Math module has some bugs but has been very effective in handling the types of math input that most visitors submit. Mostly this has been a platform to test simple math expressions with English to Math to English conversions.

Here is how the on-line version handles your and some other examples:

Toby Graves - Feb 1, 2011:

Not sure what you mean about American, the question is in perfect English.

In America, nobody would say “2 add 2.”  They say “2 plus 2”.  I believe Merlin is American.  Thus your question would be a good test for a bot written by an American as the “2 add 2” terminology would not have been encountered before.  As it would be a new concept stated that way, a bot capable of figuring out what that phrase means would be displaying flexibility.  A brittle AI agent would treat the phrase possibly as one string instead of say, “metaphorizing” what it already knows about encountering patterns like [ something] operator [something].

Toby,
It is a good test for unseen input. It also exposes some of the strategies I use to attempt question answering. Skynet-AI includes a concept of “Cascading Search Engines”.

It has the ability to test questions against one search engine and if it does not find an answer it tries a secondary engine. In addition, depending on the type of query, it will direct the query against what it thinks is the search engine that could provide the best results.

The response:
“AI: Let me search for an answer. Sometimes, it takes a couple of seconds. (Chatbot opens a webpage which defines the number 2 instead of giving the answer).”
Comes up after the AI has failed to parse the math problem, attempted to derive an answer from a primary search engine, and then defaulted to a backup engine.

At least it gave an answer that was math related.
If you typed the same thing into Google the first page is about a Turkish Ariline. (Google would have also recognized “What is 2 plus 2” and responded with “4”). One of the problems with search engines is they do not have the concept of “I don’t know”. You are on your own to try to figure out if what they give you is correct or not.

“I have been able to integrate the transformational logic into my test platform so that I can handle structured word problems:” and gave some examples that this part of the bot was working.

Steve,
The test platform is not on-line. I test and develop off-line so that I don’t introduce unexpected problems into the bot before I am ready. I do continuously update Skynet-AI, but I expect to only add these enhancements in the next version.

I haven’t decided if I am going to add all of this into Skynet-AI (which plays a certain character) or launch a new bot (which I tend to do every year or so). With Skynet, input is a single line text box. If I keep that style, I would like the bot to be able to answer the questions even if some of the information is intermingled in earlier parts of the conversation. The alternative is to go into a “Math Mode” that allows multi-line text entry and can more easily handle word math problems. When I am happy with the results I may start testing with a Math Mode.

I also want to fully explore “date/time math”, where the bot has a much better concept of time.

Jan Bogaerts - Feb 1, 2011:

I am confused. I thought the JAIL platform that you are using for skynet-AI is based on regular expressions?

Jan,
Jail (JavaScript Artificial Intelligence Language) does use Regular Expressions, but that is only one part of what I call a “neuron”.

I’ll try to explain a bit more over on the Jail Thread

Funny intermezzo: ever tried: ‘what is 2 plus 2’ on Google???

http://www.google.com/search?hl=en&q=what+is+2+plus+2&sourceid=navclient-ff&rlz=1B3GGGL_en-GBNL349NL351&ie=UTF-8

Interesting: ‘what is 2 add 2’ doesn’t work.

Nice, Google got all the following correct:

what is five times eight

fourty eight divided by seventy

what is half of (5 times 9)

ninty two times 3 - 4

three times sqaure root of 25

sine of 88 divided by square root of 9

sine of the cosine of square root of 888 divided by 72

sine of the cosine of the tangent of (88 divided by 7 plus 2) times 4

it assumed angle in radians which is cool.

It even seems to know about complex numbers:

-55i x (72 + 14i)

This is a good example of the “Cascading Search Engine” approach I spoke of earlier.
Although the basic Math AI in Skynet-AI could not parse the input, the primary search engine was able to digest the result and return a correct answer.

One of the reasons I have focused on math is that all of the computers can handle this equation if formatted correctly. Yet, if entered in natural language, few can handle it now.

Victor Shulist - Feb 1, 2011:

It even seems to know about complex numbers:

-55i x (72 + 14i)

Indeed, Google’s calculating capabilities are impressive. I use it as my primary calculator and unit converter. (It even converts to smoots!) The calculator also knows most physical constants. If I need Planck’s constant, all I need is to type “h”. For Boltzmann’s constant, “k”! (This last one comes up a lot. Too bad it assumes SI…Ah, well. Still wonderful!)

hehehe….. CR:  I knew you’d bite for that complex numbers example, being our physics expert. 

Yes, golden ratio, speed of light in vacuum, e, Josephson constant, magnetic flux quantum, von kiltzing constant, etc etc. 

My niece is going through university, second year, majoring in physics, she knows about my bot; I told her, yes, you can help educate it with your plethora of physics knowledge.