[sunama]

One thing you have to be careful about is complexity of your NLP.
You must keep a close eye on performance (I assume this is why you are using C++, ie for speed).
My belief is that your AI will need A  LOT of knowledge. This will mean a HUGE database.
The computer(s) that your AI will require will have to be very powerful.

My own AI does not go into such great depth as your AI, as I believe that the current state of computer hardware is prohibiting this.

Basically, your code will have to be very very efficient. Creating efficient (depending on your skill level), will take a greater amount of time. This has its own dangers.

Is it not possible for you to team up with another c++ programmer and both of you will develop the AI?

I would hate to see your idea/algorithms/concepts vanish without a trace and years later, see the likes of Google make a US$Billion from an advanced AI, which essentially does the same thing as yours aims to do.

[victorshulist]

"One thing you have to be careful about is complexity of your NLP."
-------- I'll do whatever complexity is necessary to get the job done.     As I said, I don't want just another bot.

"You must keep a close eye on performance (I assume this is why you are using C++, ie for speed)."
-------- Yes, the engine is exceedingly high performance currently.   30+ word sentences in 0.10 seconds... on a slow (1.6 GHz) single core laptop !!!

My belief is that your AI will need A  LOT of knowledge. This will mean a HUGE database.
The computer(s) that your AI will require will have to be very powerful.
--------  Any system capable of true NLU will need a lot of knowledge.  This is not going to discourage me        Also see comment above.    Can always go multiprocessor, split up work into multiple threads to run on say an 8 core system (2 cpu x 4 core, or even 4 cpu x 4 core). 

My own AI does not go into such great depth as your AI, as I believe that the current state of computer hardware is prohibiting this.
------- see comment above.   Efficient algorithms.    The C++ I write is practically as low level as assembly language, runs exceedingly fast.

Basically, your code will have to be very very efficient. Creating efficient (depending on your skill level), will take a greater amount of time. This has its own dangers.
--------- again, same comment as above.

Is it not possible for you to team up with another c++ programmer and both of you will develop the AI?
-------- so far all other developers are doggedly follow the same old paths, pattern matching, statistics based, and other methods that I don't agree on.   So yes, if I could find developers that would just do what they're told lol

I would hate to see your idea/algorithms/concepts vanish without a trace and years later, see the likes of Google make a US$Billion from an advanced AI, which essentially does the same thing as yours aims to do.
--------Agreed!


So yes, the complexity of the system is fairly deep.  But it takes what it takes to get the job done.    Clearly, Eliza/Alice type level technology will never give us true NLU.    As I progress through the project, I am creating documentation and ALL levels, so I can keep everything straight in  my mind !

[sunama]

"You must keep a close eye on performance (I assume this is why you are using C++, ie for speed)."
-------- Yes, the engine is exceedingly high performance currently.   30+ word sentences in 0.10 seconds... on a slow (1.6 GHz) single core laptop !!!

This is VERY impressive.

I timed my own AI, through a simulator (inputting the retrieving information, to/from a system client) and I tend to average around 0,07s/query.

Bear in mind that the above are simple queries and I am using lots of multi threading to achieve that speed.
Intel Core2Duo E6600@3.3GHZ.

This is in VB.NET.

In order to get the speed that you are seeing, you must be writing very efficient code.

One of the reasons why I decided to not use C++ is that a higher level language like VB.NET is a lot more forgiving with errors and I believe that it is possible to write code faster. WIth C++, you have to be very precise.

I have looked in to converting my code from VB.NET to C++ (as C++ will be faster), however, as my program has about 26k lines of code, it would probably take me about 1 month to do the entire conversion (using conversion software). I may do this, but not yet.

Have you tried looking for a coding partner by advertising on Facebook or Google? What I'm saying is that the faster you can get a prototype ready, the faster you can get a webclient up and look for funding to take this technology to the next level. A coding partner who is good at C++ can help you in this. And I don't even think he needs to know how the program works. You just need to make empty methods for him to code, telling him what the input/output must be. In time, he will learn how the program works.

[victorshulist]

*
Some more examples
**


<user says> jack went to a fun dance

<Grace says> Yes, I understand.  I have stored that information in 'temp-sqa-test'

<user says> did jack go to a dance

<Grace says> If you mean *fun*, then yes.

<user says> did jack go to a big fun dance

<Grace says> Yes, but I'm not sure if it was *big*

user says> victor was going to a big party

<Grace says> Yes, I understand.  I have stored that information in 'temp-sqa-test'

<user says> did victor go to a party

<Grace says> If you mean *big*, then yes, I'm assuming so.

<user says> did victor go to a big fun party

<Grace says> Yes, I'm assuming so, but I'm not sure if it was *fun*

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

so notice , Grace pays attention to absolutely all words.   In the above she responds "assuming so" because , after all, we said  "WAS GOING TO" . .. that implies on our way, but the person didn't necessarily get there!

That is different than saying "WENT TO".

[victorshulist]

Yes, well in my case it was not just moving to C++.   It was many changes to the algorithm's efficency.

Yes, C++ is very strict.   However, the positive side of that is that you catch potential bugs faster.   But at the expensive of slower development.    Scripting languages are preferable for proof of concepts.   

C++ developer to help? Hum, that sounds like a good idea in general.  He/She would need minimum 10 years coding experience

[Art]

You should definitely make friends with another member here named Bruce Wilcox. He too believes that AIML had its time and place but not necessarily in ther advancement of NLP or advancing AI.

Not that Alice and AIML are bad by any means but to say that there are other avenues in which to explore that better suits his proposed target.

Bruce wrote the software AI in the Blue Mars project as well as code for many GO games which are quite complex to say the least. Bruce's chatbot caught the attention of a lot of us here as Suzette was quite the breath of fresh air in our community.

He's a really nice person and quite gifted in the field of software and AI. Give him a shout!

All the best!

[sunama]

He/She would need minimum 10 years coding experience

If you start placing stipulations on the qualities of your coding partner, you will reduce the number of applicants. I've learnt that the hard way.

Don't forget that many start-ups which sprout in Silicon Valley are based on software developers who are fresh out of University...with less than 10 years coding experience.

One problem I've found with the older guys is that they are less open to new ideas.

They only know 1 way (their way) or its the high way. Generally, youngsters are open to new/different ideas.

I think in an earlier post you stated that the people you have approached are all stuck on sentence (and key word) matching.

My belief is that as long as the individual is super intelligent, he will learn very quickly and will probably even offer you techniques on how to improve your own algorithms. 2 minds are almost always better than 1.

In saying all of the above, you are the boss of  your creation, so only you can make this call.

I still say that developing a large piece of software on your own is very dangerous and will leave you vulnerable to larger companies who will start development of your style of software after you and will race to prototype stage, much faster, leaving you in the dust. While you are still a few months away from prototype, they will already have funding in place and will have hired even more staff to work on and develop the software. From the point onwards, you are finished.

The way I will be getting around this is as follows:

1. Create the prototype (this will have bugs, but it needs to be just good enough to demonstrate that my idea actually works and will be enough to impress possible investors). VB.NET will allow me to reach prototype stage faster.
2. Start looking for funding based on what I have.
3. If I need something more advanced, I can always license some algorithms/code from other coders, like yourself, for example. These additional algorithms will be tagged onto the original AI. Possible investors shall be informed of this possibility during pitches.
4. After funding has been attained, hire some programmers who will start working on coding the AI, full time. This will accelerate the development of the AI and even if another company starts developing their own (similar) AI, they will be racing, on equal footing with me. It won't be easy for them to "overtake" me.

If I decide to do this all by myself, right the way through (say on a 3-4 year time-line), I won't stand a chance and could get overtaken very quickly and with ease, by a company or group of coders.

The other alternative is to hire a coder, or a group of coders, using your own money and get the prototype completed within 6-12 months. The problem with this idea though, is that you have to have a lot of cash available.

[victorshulist]

For populating, and maintaining the database, files, rules, sure.  But  I'll do the core coding for sure

I think it is time to break the chain of "just get the thing out the door" and everything ends up angry about buggy crappie software.

Now for sure, I would have enough non-core coding work for 6 people working full time !!

[sunama]

OK, here is a question.

As far as commercial applications go, where do you feel your AI (once the prototype is working as expected) will be applicable.

For example, in my case, the entire is being developed in order to answer questions using natural language. I call it an information retrieval system.

As the AI develops, so will its understanding and it will be able to do a whole lot more.

A simple example, would be for a user (or officer of a company), to enter facts about their company. Eg. we sell item x and item y. The price of item x is £1. The price of item y is £2. We offer discounts when condition C occurs.

A user, who is seeking information about this company can then ask (using natural language),

"how much is item x?"
OR
"what is the price of item y?"
OR
"what do i have to do to get a discount?"

It is easy to see that this system can replace the role of human beings if a customer is asking basic questions about a company's services/products.

Initially, my prototype won't be any good at idle chit chat. For that you need to see clever-bot, which is good at idle chit chat but is crap at everything else.

So, in what way can your AI be used, in a commercial application?
Why would a company want to buy your product?

[victorshulist]

That is a very very good question !!

So originally this whole concept of a conversational computer program has fascinated me.    The whole concept of he Turing test.   And for the most part, it is still a fascinating academic challenge.    A year ago I think my goal was a "chit chat" system but I think the last few months I have changed my mind and decided to first accomplish something more useful, practical.

So, that was a little context or background.   To answer your question, I am aiming almost exactly at what you outlined.    Information retrieval is a high priority.   Everyday at work if I want to find a document someone just says  "oh .. it's on the fileshare"....  I feel like telling them to go to h***  because I know the nightmare of surfing the file system HOPING I will find what I want.    The world NEEDS natural language I.R. system.   And I mean a powerful one.   One that can, if it is uncertain, can ask you questions, and actually have a dialog with you in order to "zero in" and know exactly what you want, and find it.

Google is wonderful and we tried putting "Google Desktop Search" on our file server.  Not much good.

The reason is :
Google is ONE SHOT.

You type in your keywords,  cross your fingers, pray, whatever, and hit the button.  You either get it or you don't.   There is no interactive "clarification" questions.

So very powerful I.R. is one goal yes.

What I was thinking more is an interactive dialog system that can answer questions about a product, similiar to what you said.

user--  I'm interested in a cell phone
ai-- I see, now will you go pay-as-you-go, or monthly ?
user--  uh....  i'm sorry to admit , I don't know what that means
ai-- oh, of course, well, pay as you is when xxxxxxxxx and monthly is xxxxxxxxxx
user-- say I go monthly, what options do i have then?
ai -- well, then you have  A,  B, or C

I don't know.. I have no idea, I don't even own a cell phone, but you get the idea.     Before I would show an investor I think I would want very powerful I.R.  **but** with interactive conversation like that.

In addition to I.R.  I'd like to show the system's ability for troubleshooting...

user-- i can't surf the net, something is wrong
ai--  can you ping anything?
user-- what does that even mean?
ai-- try opening a dos window, and enter "ping www.cnn.com"
user--  what the heck is a "dos window"
ai-- i'm sorry, a dos window is where you can type commands. 
user-- oh, and how on earth do I "open a dos" window as you say?
ai--- to open a dos window, click START and then click RUN, then type "cmd" (without the double quotes) and press enter
user--- ok a black box appeared, now what?
ai--- good ! that is the dos window!  now click in that window, and type 
ai--         ping www.cnn.com
ai -- then press enter

[victorshulist]

Also, the system understanding natural language questions that involve "IF" will be applicable.

user-- "So, you're saying if I sign up for six months, I get the first month free? is that correct?"
ai-- that is correct, but you can sign up for a year and get 2 months free.

something like that.  Where the system can even evaluate NL "if" logic.

So I guess to sum up:

1-   customer service (finding information, cable TV plans, cell phone plans, or you want to locate a specific item on an electronics site)

2 --troubleshooting

I will also probably , in parallel, be working on a chitchat functionality and also as a tutor for educational purposes.    

And later on, perhaps something like a Tax adviser

user-- how much do I have to pay to the government this year?
ai-- as it stands right now, you have to pay $500
user-- can I purchase more RRSP or something ?
ai-- yes, buy an additional $5000 in RRSP and your balance due would be 0.00.
user-- hum.. I don't have that kind of money kicking around
ai-- well, about about $2500, then you'd only be paying in 125 $

I don't know.. but you get the idea

Actually, when I really think about this,  I think what I really want to build IS:

A system that can learn.

In other words, I don't want to care about the information.

I would have a product that can learn anything, in NL.   Then, a company can use that service to educate the system,  (not in programming, but in NL).   And when they have told it everything it needs to know, they can test it (like my examples above).

Once it has passed all the tests of KNOWLEDGE about their company, they would put it into production.

Thus, what I am building really is a "Natural language development system".  Where, instead of the system admins being Perl or C++ coders, they just teach it via English.

Thus, if the system gives false information, it wouldn't be my problem if the NL fact inputs were wrong.  Only if it was an internal (core language engine) issue, would the issue be handed to me.

[sunama]

Your phone sales example is moving away from information retrieval  and customer service and actually moving towards sales. It would be quite an achievement if you could create an online salesman. I think though, it would take you many years to get to that point. Sales is a very sensitive area and the last thing you would want is for your AI to recommend to a customer a much cheaper product (as opposed to the more profitable model, which the customer originally wanted to buy).

[victorshulist]

I appreciated the times when a salesperson didn't try to push a more expensive product on me just so he can maximize his profits.   I then developed a long term relationship with that company.   It's called TRUST.

But again, that choice would be up to the customer (my client, not the client's client).

However, if the goal *was* simply to sell the most expensive product, it would be like a game of chess.

Like chess, before the computer makes a move, it determines what sequences of moves is likely and which one of its moves could end in stalemate, checkmate for him, or checkmate for opponent.

In the same way, the system would have to consider, before I make this next comment, will it result in the customer changing his mind about which product to buy.

So yes, that is a lot more complex.

[sunama]

It is going to be hugely complex.

You are now talking about some very complex sales techniques. In order to teach/program these techniques, you may have to bring in a top salesman and have him give you techniques.

I think this development would be best left after your program is fully funded and is being sold/distributed as part of a company.

IMO, this would be a 4th of 5th generation product...perhaps 5-10 years in the future.

It would be truly mind blowing technology, considering that top salesman earn huge sums of money.

[victorshulist]

Agreed, on all points.  All in time.  Especially one man working a bit on evenings and weekends!

Advanced I.R. first.

And I do appreciate your conveying the sense of urgency to look for funding.