[sunama]

Here's a question:

Answering a question about Pizza will certainly be an interesting point of research, BUT, for commercial purposes, why would any business be interested in buying a product/service which is capable of carrying on a casual conversation, about Pizaa (or any other subject)?

2 years ago, I was having this discussion with somebody. One thing he told me (and I do agree with him) is that creating a chat bot (which can do casual chit only), is pointless, simply because this world has plenty of real human beings who can do a better job.

What we need is an NLP which can process useful information/data. Dealing with casual questions about a person who is hungry doesn't seem important, from a business/commercial angle.

If I was on the phone to a big company and telling them about an AI, I couldn't use the fact that my AI can carry out casual conversations and talk about Pizza. The CEO of any company is very unlikely to buy my product based on this.

When making the sale over the phone, I would find it a lot easier if I can sell a product which is able process information/data, which relates directly to their business.

My way of thinking is get the features which are marketable and are of monetary value to businesses, implemented first. THEN, if you have time and money available, spend some of it on "trick" features, eg. making casual chit chat about a random subject (pizza or anything else).

Perhaps I am focusing too much on the business/commercial angle.
What do you guys think?

[DaveMorton]

2 years ago, I was having this discussion with somebody. One thing he told me (and I do agree with him) is that creating a chat bot (which can do casual chit only), is pointless, simply because this world has plenty of real human beings who can do a better job.

I have to completely disagree with this for a variety of reasons, but I'll illustrate just one: Anything that helps even a single person to enjoy life without causing harm to themselves or others can never be pointless. Even discounting that, the sense of accomplishment that the chatbot's creator derives from it also invalidates pointlessness. And one must also consider those (admittedly sad) folks who don't like other people (I know a couple) who would much rather talk to a computer than another person.

[Freddy]

I think there's really two areas here, maybe more.  Businesses would probably mostly be interested in some kind of expert system that would spew out information all about their product and so who's going to visit that site just for a chat ?  I reckon the business would want to discourage that probably. 

And then you get your companion chatbot, like an AI pet where it's more likely that the programmer will widen the field for topics of conversation.

People must like them, if only for a while, there are thousands of chatbots out there now.  But there is a long long way to go before they can beat us in conversational skills.

[sunama]

2 years ago, I was having this discussion with somebody. One thing he told me (and I do agree with him) is that creating a chat bot (which can do casual chit only), is pointless, simply because this world has plenty of real human beings who can do a better job.

I have to completely disagree with this for a variety of reasons, but I'll illustrate just one: Anything that helps even a single person to enjoy life without causing harm to themselves or others can never be pointless. Even discounting that, the sense of accomplishment that the chatbot's creator derives from it also invalidates pointlessness. And one must also consider those (admittedly sad) folks who don't like other people (I know a couple) who would much rather talk to a computer than another person.

I hear what you are saying and from a non-business perspective and I would agree. After all, a lot of people play with chat bots.
If a programmer is making a chat bot for a hobby...no problem.

But, what if the programmer is developing his idea into a business. Would this (casual chit chat) system be of any worth?
If you were making pitch for such a system to a potential buyer and are asking him to pay money for it, what exactly would your sales pitch be?

[DaveMorton]

Well, since I've already stated in another thread that I have no experience in the area of business, I can't answer your question with any degree of confidence. However, I can point out that there are several companies out there (Pandorabots and Personality Forge, just to name a couple) that at least make enough to "leave the lights on", so while a single conversational chatbot may not be a sound business investment, a hosting environment for a collection of them most certainly is.

[sunama]

However, I can point out that there are several companies out there (Pandorabots and Personality Forge, just to name a couple) that at least make enough to "leave the lights on", so while a single conversational chatbot may not be a sound business investment, a hosting environment for a collection of them most certainly is.

That's a good point.

[TheMikh28]

The way I see it, the chatbot market is saturated.  The natural language processing software market in its present state is saturated.  Any attempt at breaking into the respective markets nowadays is merely a drop in the bucket.  Even more niche corners of the markets are getting full.

At this point, making a pitch to businesses isn't necessarily going to get a project of this nature picked up.  It has to stand out, generate buzz, have some success on its own to attract the attention of potential investors.  Either that, or it has to have some serious brilliant innovation that both wows and demonstrates applicable relevance.

For me, developing an artificially intelligent entity is a means to an end.  Ideas, principles, algorithms, processes, etc. - "technology" - all get refined through tons of scrutiny as they are developed, incorporated, and used in the project, and they may in turn be incorporated into other projects.  My project is far from complete, but its core libraries that I've written are (and have been for over a couple years now) the lifeblood of at least half the various projects on my server.  Some of these libraries and tools, should I desire to do so, can be made and distributed as standalone software (if not just useful libraries) themselves.

I'm not an entrepreneur myself though, and I don't particularly desire to go down that path with my own projects, but having a public library of powerful self-developed (and likely open-source) software -- which I would ultimately like to offer -- certainly does not hurt my prospects in other ambitions, such as employment or research/academic opportunities.

Just checked out the present state of the AI Core.  From what I understand, its conversational capabilities are currently at the most basic (AIML-esque), and it incorporates a simple grammatical parsing and association-forming / logic engine.  For the most part it feels very hard-coded rather than learned, like a natural-language-processing expert system of a sort moreso than an artificially intelligent entity.  This is neither the first nor the last of its kind, and in my opinion the system isn't truly at a stage where it will attract real investment unless I'm missing something significant.  If this project is to flourish, it has to offer something substantially "more" than what it currently does.  I am, however, curious as to what components of the project the 40,000+ lines of code are being allocated to.

I apologize if I come off as harsh.

Just my two bits worth.  </rant>

[sunama]

Mikh28, you are not being harsh, but honest. There is nothing wrong with that.

You are indeed correct in that there are plenty of NLP systems around, which have or are being created by some bright minds.

What I am seeing though is that the use of NLP systems in industry, seems to be very sparse.

I have seen some websites offering a (basic) NLP system, for use as web assistant. Apart from that, can you give me any areas where NLP is being used commercially (and not just as some academic/research experiment)? I would very much like to know.

My idea is to actually get various NLP systems to market.

I am currently taking a "pause" as it were, from developing the AICore as I am now looking at NLP from a business angle.

What I am seeing are some bright people, producing and implementing some good algorithms. I am seeing them place their NLP's on websites where other users from around the world can play with them. This is nice, but it isn't how you turn the idea into a business.

Think about it, NLP is something so unbelievably amazing yet nobody has yet managed to market the technology properly.

I shall give you an example:

You phone up a customer service centre and you have to...press 1 for customer services, press 2 for billing, press 3 for payment, press 4 for... you get the picture.
To program this sort of system is ridiculously easy and should've been phased out about 5 years ago. Yet, for some strange reason, this antiquated system is still in operation today.
If the state of NLP is so saturated, why are we not seeing these simple systems (eg. the press x for y department), being superseded by a more advanced NLP?

My belief is that while academics know what can be done with NLP, businesses don't...hence we have some great technology which isn't being utilized by commerce.

The situation is similar to if a small bunch of people around the world would have created an engine which uses 75% less fuel/energy than current vehicles. Yet the engine is not actually being put to commercial use. This just wouldn't happen. But this is the current situation with NLP.

Business/Commerce
At present, my time is currently being spent looking at NLP from a business angle (not as an academic or programmer) and I am seeing some huge gaps in the (commercial) market. Where NLP systems can be used in such a way that businesses can save themselves money.

I am currently thinking of changing my business model to something which involves licensing algorithms from different people. This means that rather than wasting time on programming a product which other's are already working on, I want to let them continue working on their program, while I concentrate on monetising their creations and give them a percentage. This means that they start getting a revenue stream for their NLPs, while they develop them. This is a plan which I have been thinking about for the last week or so, so I am still refining it.

I have already contacted Bragi and Victor (from these forums, whose NLPs I hold in high regard) and am also contacting other programmers who have good NLP's.

I have also contacted various business people and I am evolving my original business plan.

You are right in saying that, in order to gain funding, whoever it is, shall have to distinguish themselves from the competition. The best way to do this is bring the NLP system to market and successfully sell it to the market...gaining revenue...and proving that they can bring this (business) idea forward. Once this happens, investors will take that person, seriously.

With regards to having the "hard-coded feel". I assure you, my AICore doesn't have many words hard coded. The problem is that its database is currently very sparse. I need to populate its database with more facts and sentences.

There are also many features which I have not yet switched on.
There is an algorithm in there which allows it to read stuff from the internet, automatically, when it believes it is appropriate, but right now, I haven't switched it on, as I want to bring certain features on-line, one at a time...not all at the same time.

At present, the AICore is able to tell (with around 80% certainty if a fact is true or false (by looking up on the internet). It can do this automatically, but I have not switched this algorithm on.

There are also some other algorithms which have not yet been switched on - too many to speak of - and this is the reason why there is so much code (everything is learnt and very little is hard-coded). Had I hard-coded a lot sentences, the number of lines of code, would've been significantly lower.

By the way Mikh, do you have a website for your work...I'd like to take a look at it.

[TheMikh28]

You are indeed correct in that there are plenty of NLP systems around, which have or are being created by some bright minds.

What I am seeing though is that the use of NLP systems in industry, seems to be very sparse.

You phone up a customer service centre and you have to...press 1 for customer services, press 2 for billing, press 3 for payment, press 4 for... you get the picture.
To program this sort of system is ridiculously easy and should've been phased out about 5 years ago. Yet, for some strange reason, this antiquated system is still in operation today.

If the state of NLP is so saturated, why are we not seeing these simple systems (eg. the press x for y department), being superseded by a more advanced NLP?

Businesses used to be extremely enthusiastic about systems that incorporated natural language processing and learned knowledge.  This was back in the late 1970s (?), 1980s, and into the early 1990s.  The systems (expert systems) proved far too tedious to develop and maintain, however, and the market collapsed.

Similarly, voice recognition (e.g., via telephones) was experimented with by businesses, and the results weren't as impressive as they'd hoped, and as such, telephone support that relies on voice recognition has picked up a terrible reputation.

While technology in both fields have significantly improved over the years, businesses are reluctant to invest in these new technologies due to what has been experienced with past technologies.  In addition to this, these technologies don't bring the businesses significantly higher returns (and would rather not spend money to replace/upgrade their "software capital" which works just fine), so they see no reason to adopt it.

So, what one finds are a lot of different technologies being offered by developers, and very few being picked up by businesses.  That's the saturated market I mentioned.

I am currently thinking of changing my business model to something which involves licensing algorithms from different people.

Algorithms cannot be licensed.  Software, code, and processes, however, can.

You are right in saying that, in order to gain funding, whoever it is, shall have to distinguish themselves from the competition. The best way to do this is bring the NLP system to market and successfully sell it to the market...gaining revenue...and proving that they can bring this (business) idea forward. Once this happens, investors will take that person, seriously.

Promoting and selling the product on the market is what every competitor is trying to do, though, and some are successfully doing.  Few attract significant investment though.  That is why I'm stressing the need to offer a product that goes above and beyond in excellence -- so that the product can not only stand out and sell well from good marketing, but also be attractive entities that may wish to invest in new technologies, who have their pick of any product available.  Something about the software -- and not just how it's marketed -- has to make it exceptionally worthwhile and revolutionary in order for it to succeed.

With regards to having the "hard-coded feel". I assure you, my AICore doesn't have many words hard coded. The problem is that its database is currently very sparse. I need to populate its database with more facts and sentences.

I suppose my comments about it feeling hard-coded reflected a personal bias against the concept moreso than anything, since I dropped the hard-coding of words entirely a couple years ago (we discussed this in the past).  Depending on the nature of what you're developing, hard-coding may be necessary, but I feel as though hard-coding words/phrases into an artificial conversational agent is a concept/practice that is quickly becoming outdated.

The only text that is definitively hard-coded into my project is the space character, for parsing between words, and I would ultimately like to drop that as well; it's just impractical to do so anytime soon.

There are also many features which I have not yet switched on.
There is an algorithm in there which allows it to read stuff from the internet, automatically, when it believes it is appropriate, but right now, I haven't switched it on, as I want to bring certain features on-line, one at a time...not all at the same time.

At present, the AICore is able to tell (with around 80% certainty if a fact is true or false (by looking up on the internet). It can do this automatically, but I have not switched this algorithm on.

There are also some other algorithms which have not yet been switched on - too many to speak of - and this is the reason why there is so much code (everything is learnt and very little is hard-coded). Had I hard-coded a lot sentences, the number of lines of code, would've been significantly lower.

That makes sense.

By the way Mikh, do you have a website for your work...I'd like to take a look at it.

I presently do not have any of my projects or code available online, though I'm looking to upload them to a website or something in the near future.

I'm also planning on creating a topic here sooner or later just to outline the specific details of algorithms, functions, classes, processes, etc. in my project and discuss them with other users.

[sunama]

I'm also planning on creating a topic here sooner or later just to outline the specific details of algorithms, functions, classes, processes, etc. in my project and discuss them with other users.

I remember chatting with about some ideas, via email, I think.
Specifically, it was about word relationships.
It might've been a year ago.

You can post your discussion about your ideas in the "concepts and ideas" sub section. I am happy to talk about any theories/ideas that I have.

[Genesis]

I checked out your AI core. I decided to give it one of my 'context' tests. Though the test is usually for programs aimed at AGI. I wanted to still how yours would respond to it. It tells you the level the chatterbot is in.

The first is a time test.

Bot: Hey
Me: What time is it?
Bot: The time is 01:28:8. Does this answer your question?
Me: Yes.
Bot: Great.
Me: How long has passed since then?
Bot: The answer to the question (how long has passed since then?) is not known.
Me: How long has passed since I asked you what the time was?
Bot: The answer to the question (how long has passed since I asked you what the time was?) is not known.

The second is a 'friend' test.

Bot: Hey
Me: My friend Eliza finally came over.
Bot: Does this information apply to you: my friend eliza finally came over.
Me: Yes.
Bot: Information saved.
Me: Was she late?
Bot: The answer to the question (was she late?) is not known.
Me: Where is Eliza?
Bot: The answer to the question (where is eliza?) is not known.
Me: Who came over?
Bot: Phones are used for communication over long distances (this may be true. AICore shall attempt to verify this soon). Does this answer your question?

As you can see it failed both tests. Its definitely needs more work.
Good luck.

[Bragi]

AGI: Association for Geographic Information, Analytical Graphics, Inc, American Gunsmithing Institute, Adjusted gross income ?

[lrh9]

For us language is a learned ability.

How do you know this? Perhaps you meant 'speech' (aka the control of the muscles round the vocal chords and mouth) instead of language. Don't forget, infants are born with a brain, not an empty box that needs filling.

I meant language. I don't have any formal proof. It is difficult to empirically establish a link between any brain activity and behavior. I just use the strong assertion because it is more convenient than prefacing everything with "I think...". However, I'm pretty confident for a few reasons.

Most early behavior is instinctual. Very young people are notorious for grasping things (Palmer grasp reflex) and as they get older trying to eat the things they grab. There isn't any intentionality behind crying. Merely a reaction to unpleasant stimuli.

http://en.wikipedia.org/wiki/Primitive_reflexes

Most behaviors are learned rather than inherited.

sunama, I agree that manually creating computational knowledge about a sufficiently broad amount of things is impossible, but one of the curious features about Python is that it is possible to dynamically create classes during runtime. One of the ideas I'd like to explore is to create a program that uses a very simple object-oriented English syntax to create data and form relations from conversation.

E.g. two of the most common sentences in object-orientation are "x is a y" (this says that object x is of class (aka type) y) and "x has a y" (this says that object x has an attribute (aka characteristic) y). It almost seems like they can account for a lot of what people are trying to do with chat bots.

[Merlin]

lrh9,
I agree with you on:
""x is a y" (this says that object x is of class (aka type) y) and "x has a y" (this says that object x has an attribute (aka characteristic) y). It almost seems like they can account for a lot of what people are trying to do with chat bots."

This was one of the first dynamic learning capabilities I put into Skynet-AI. That functionality is fundamental for the AI to learn and start "grounding" itself. How to store, retrieve and manipulate that information is a key building block.

[sunama]

The first is a time test.


The second is a 'friend' test.


As you can see it failed both tests. Its definitely needs more work.
Good luck.

Time test...yep. That answer will come after, I have developed the casual chat algorithm.
I shall put that on my to-do list.

Friend test:
This involves the "context" of the current conversation...not yet implemented.
This is already on my to-do list, but isn't a high priority as I may use somebody elses algorithm, thus allowing me to spend my time on something else.

Thanks for the feedback. Much appreciated and at some point both those tests shall be handled.