What if you type in an expression with Quaternions?

Just more rules and logic my friend   All in due time.  For me though, conversation is top priority.    Then, once system knows the parse tree you want, if a nested term is an integer, real number, complex number, quaternion, or roman numerals, etc, it will be handled in the IFLO, where domain specific logic and knowledge is handled.

Complex numbers, that’s a looooooooong time ago!!!! Didn’t know they still existed

Victor Shulist - Feb 2, 2011:

Merlin - Jan 31, 2011:

This also lets some of us throw grammar completely out the window and attempt to identify meaning from the input in a “fuzzy” fashion.

I wish you luck in that ‘attempt’.  Is the solution forthcoming ?  I mean do you have any ideas on how to go from fuzzy to crisp?

The solution is already built into Skynet-AI. Crisper rules (neurons) have a higher priority and thereby “shadow” fuzzier rules. You can see an examples of it in the “Learn” function and the word math problems.

In word math, if the AI can recognize the math problem and solve it, it returns an answer. If not, it determines whether to send it to the “cascading search engines” or a fuzzier response (sometimes both).

Assuming I am successful in my version .004 development I expect to be able to have the AI dynamically add to the knowledge base in real-time much more extensively than it does now.

Most people, most I know anyway, speak and write with proper grammar.

True, unless you put your bot up on the web. Then, almost every conversation either has spelling errors, bad grammar or text speak.

Also, keep in mind, that the core CLUES engine doesn’t care or differentiate between proper and improper grammar,

Maybe you are using a “fuzzy” system after all.

Merlin - Feb 2, 2011:

Most people, most I know anyway, speak and write with proper grammar.

True, unless you put your bot up on the web. Then, almost every conversation either has spelling errors, bad grammar or text speak.

Also, keep in mind, that the core CLUES engine doesn’t care or differentiate between proper and improper grammar,

Maybe you are using a “fuzzy” system after all.

Yes, I think people have a certain connotation with the word ‘grammar’.  It was my wife that actually suggested that I put more focus on grammar in the bot. 

Now when people hear the word “grammar”, they think of their school days and your strict grade school teacher.    There are many things in grammar that are completely unnecessary, and that CLUES doesn’t care at all about.

For example, agreement of subject and verb rules, CLUES doesn’t care at all about that.  Also you can be a ‘wise guy’ and say things like “Yes I are” instead of “Yes I am”.  So all kinds of things like that are tolerated.  (Even “how r u”, “how ru”—“text speak”).

So when I speak of CLUES being centered around grammar, perhaps I shouldn’t use that term, instead, it uses grammar production rules as * possibilities* that is, HINTS of what the structure can be.  Those “possible structures” can be anything from a set of “grade school” grammar rules, or street grammar.

So it is already tolerant of grammar, but not about spelling (not yet).

That functionality won’t be a problem (calculating closet word match, then generate parse tree for each possibility), then using semantic inference that it already has, it can pick the parse tree with a specific spelling of the word that correlated with data in its KB.

So, it just amounts to having “layers of permutations” to go through on the front end. 

Merlin - Jan 31, 2011:

I have been able to integrate the transformational logic into my test platform so that I can handle structured word problems:


The price of a radio is 69.70. If this price is 15 percent less than the
marked price, find the marked price.
SKYNET-The answer is: 82


merlin, I had this talk with skynet

ME:  my brother’s name is paul

Skynet AI : Hello paul. Are we capable of living peacefully together?

ME:  LEARN my brother’s name is paul

Skynet AI : I’ve got that paul.

Or do you have to edit something on the server side, for it to pick that up? 

Actually I had the priority of the learning rule/neuron set to low. Until recently not many visitors were using it so it didn’t matter. I have turned it up now. I have tested a few different learning modes to try to get a feel for which might work best.

Skynet thought your name was Paul. the “my name is” neuron is very fuzzy and I really need to add a crisper rule to deal with the “my *‘s name is *”. Some of the conversation have been funny when Skynet has started calling people by other names.

My best friend’s name is Henry.
I have only one best friend.
Henry’s wife is Jane.
Jane has a 50 dollar radio.

My radio is double the cost of my best friend’s wife’s radio.

How much is my radio?

That would be cool. Along those lines, I found that to do those types of things I needed to rethink how I was adding learned neurons.

Merlin - Feb 3, 2011:

....and I really need to add a crisper rule to deal with the “my *‘s name is *”. Some of the conversation have been funny when Skynet has started calling people by other names.

Crispiness and fuzziness- the proper amount of each, seems to be a challenge we are both facing

Victor Shulist - Feb 4, 2011:

Merlin - Feb 4, 2011:

(If Skynet-AI can solve a math problem would you say it “understands” and “reasons” about it?).

Probably not.  It depends how flexible it is, I’d have to test it

Here is a good example of what I would consider truly understanding and reasoning via NLP.  It is an electronics problem.

user: R1 and R2 are connected in parallel.
user: The voltage across R1 is 100 volts.
user: The resistance of R2 is 25 ohms.
user: What is the current through the resister that is in parallel with R1?

AI : 4 amperes

So you have natural language understanding, correlation of facts, and deduction of the answer.  The system had to simplify “the resister that is in parallel with R1” to mean “R2” (since we only said one resistor was in parallel with it), then deduce that, since R1 and R2 in parallel, R1’s voltage = R2’s voltage = 100 volts, and then lookup of the values, and then the actual simple math (ohms law (for simple DC) : current = voltage divided by resistance).

I would consider your example potentially understanding an “electronics” problem. Before Skynet-AI gets there I hope to handle more broad math problems.

C R Hunt - Feb 7, 2011:

Victor Shulist - Feb 7, 2011:

But I think it will be very useful, even if not considered ‘thought’, if the program can figure out which combinations of logic modules (in this example, the only “logic module” was basic ohms law).  But later i want logic modules, not only with electronics, but with any other topic.

What I’m looking forward to is the bot taking basic modules, like an “algebra module”, and building more complex modules, such as an “ohm’s law” module, on its own based on natural language instructions. How cool would that be?

CR - I agree! I feel the building block approach has the best chance of making progress towards a general AI. I think this is similar to how we teach our children.

Sticking with math, when building Skynet-AI I started with basic math and then added the ability to do a bidirectional transform to and from natural language. I hope to expand this function and add additional types of natural language word math problems.

The most basic type of math problem has all the information necessary to solve it in the problem. For example:
2+2=
What is two plus two?

More complex problems require multi-line input and predefined knowledge.

Assuming I have handled the translation correctly for addition, subtraction, multiplication and division; I am now trying to expand the range by adding:

Additional symbols/predefined constants: PI for example
Logic: Is two plus two four?
Predefined formulas for a specific domain/type of word math problem:
Unit conversions: How many feet are in three meters
Geometry: If one side of a square is two feet long, what is the length of the perimeter?
Date math, purchasing, etc.

Maybe someday even an “electronics” module.

It’ll be great to see where your bot goes with mathematics, Merlin. To be able to understand, manipulate, and correctly respond to math questions is basically equivalent to responding “intelligently” to a restricted set of natural language. (Restricted in the sense that math problems generally aren’t as free-form in composition as most text.) That’ll be a huge accomplishment.

I’m taking the “module” approach to the way my bot translates complex sentences into “simple” sentences. If it’s given a complex sentence and its equivalent set of simple sentences, it currently sets up a sort of generalized mapping onto the simple sentences that it can then apply to other sentences of the same grammatical type.

However this mapping currently doesn’t allow for word stemming, conjugation, or part of speech morphing, which might be useful in practice. Therefore I plan to build modules that perform these type of operations and let the bot figure out which modules to apply to which words/phrases of the complex sentence in order to arrive at the simple sentences. (Potentially computationally intensive, but I’ve got some ideas on how to do this “intelligently”.) Then, the bot will save the set of instructions it came up with (which modules to use and in which order) so that it can use its own instructions on future sentences. I can’t wait to implement this..!

C R Hunt - Feb 7, 2011:

It’ll be great to see where your bot goes with mathematics, Merlin. To be able to understand, manipulate, and correctly respond to math questions is basically equivalent to responding “intelligently” to a restricted set of natural language. (Restricted in the sense that math problems generally aren’t as free-form in composition as most text.) That’ll be a huge accomplishment.

One of the reasons I chose math is because it is both a restricted set and it has been proven that computers are already good at math given that the input is formulated in the correct way. I believe the same could be said of “time”. Virtually every computer now has a clock and a math chip. In the future, every bot should be able to handle both these functions at least to some basic level.

Merlin - Feb 7, 2011:

Additional symbols/predefined constants: PI for example
Logic: Is two plus two four?
Predefined formulas for a specific domain/type of word math problem:
Unit conversions: How many feet are in three meters
Geometry: If one side of a square is two feet long, what is the length of the perimeter?
Date math, purchasing, etc.

Maybe someday even an “electronics” module.

Excellent, I think that would be a very logical next step.  The reasoning being, you would be tackling more complex language constructs, bridging the gap between math formulas and natural language (I mean natural language applied to the physical world).

So, are you planning to enter Skynet-AI in the chatterbox challenge this year?