[Madeirey]

Please forgive any questions that might seem silly, obvious, or otherwise stupid for I am still a high-schooler and a complete beginner when it comes to programming C++.

Would it be possible to program an AI to learn from the internet and your files? Could you tell it to go teach itself and "read" all about, say, physics, and then tell you what you want to know or even help you out with physics homework? Or, could you give it some of you files of research papers and could it learn about those subjects from that? Or, if you asked the AI a question about thermodynamics and it didn't know, could it bring up a webpage about thermodynamics and say, "I don't know, but maybe this will help you out."?

I have absolutely no idea how a person would go about doing this, but I don't even know if it is possible. Is it? And if it is, how would you go about doing this? Has anyone done this before?

[Freddy]

Questions promote discussion 

Some of this can be done now, some of it has been attempted or is being developed.

I won't give a definitive answer because that would take hours and I don't even know if I could, so I'll just rattle a few things off of the top of my head.

You might want to research NLP - Natural Language Processing.

In a simple way this is about bridging the gap between human language and machines. There can be lots of places where this might be used. You also get into things like voice recognition and so on. But it's one thing to hear a sound and identify it and another to understand the meaning of the sound. That's my simple view of it.

Here we talk a lot about chatbots which are supposed to mimic the way in which people engage in conversation. Some people stop there, because it takes years to achieve just this - especially if you are a lone enthusiast without a big research project. That's not to say money isn't being thrown at the problem, because it is.

Some people branch into assistants too, where the main aim is to make using devices easier, especially good for people with disabilities. You can search Google for UltraHal and also Denise by Guile3D to get a couple of examples. UltralHal is a pattern matcher chatbot, it doesn't really think - it uses a method of determining what response is most likely 'best' for whatever the user just said or asked. It can work well.

I mention those because in the past UltraHal could learn things from a text file, but was still tied to pattern matching. I'm not sure if Denise does this, but both programs (perhaps with a little modification) can take a query and if it does not know the answer it will load up a web page for the user. Or having looked up the information present it in a way more to the application's suiting. This function is actually pretty easy to program, it's not a lot more than a search at a web service. AIML (another one to look up) bots can do this too, it works well because many AIML bots are on the web anyway, and more increasingly in mobile devices.

Web companies and organisations will often provide an API or Application programming interface - this allows bot makers to do some nice tricks and integrate other features into their programs. So with regards to your question about looking up thermodynamics - your bot could search Wikipedia as well as Google. Places like YouTube also have an API you can hook into too so you could also show a video perhaps.

With regard to learning things from text files in more detail etc, I don't know much about that, but I know people are doing things like this. Perhaps one of our programmers can step in on that one. People do spend a lot of time parsing languages, it's not an over night thing. And it's debatable how far it has come.

Try looking for IBM's Watson to get a glimpse of what might be possible in the future to a wider audience.

I've rambled on enough, I hope that's given you a few ideas about what to research.

[Don Patrick]

Yes, these things are possible, but one of them with great difficulty.

To start with the easiest part: Various existing chatbots will quote the first paragraph of Wikipedia or access Google to launch an appropriate website if you type a question, as does the Siri phone app. By recognising keywords and common phrasings like "What is a..." or "tell me about..." the program can recognise what the subject is that should be searched for (whatever is on "..."). This is feasable for a highschooler with some effort.

The other part is very hard: Having an AI read a text, learn from it and answer direct questions about the contents. I know this because I spent two years programming AI that does exactly that, and not yet perfectly. IBM's Watson is another program that extracts answers to questions from Wikipedia texts, and it is considered state-of-the-art AI (it uses both grammar and statistical analysis of word matches). If there are more such programs I would love to hear about them. The only program I know that can actually help you with your homework is Wolfram Alpha.
Natural Language Processing is indeed the field to research for this.

As one example of how: I first built a vocabulary to recognise and categorise the words, and a knowledge database of text files to store facts in. I programmed hundreds of grammar and spelling rules to convert sentences to e.g. subject-verb-object facts, then a central function that saves or retrieves the requested facts from the database, then some more grammar to convert the returned facts back into sentences. That's about the least you'd need to answer straight questions in specific detail. Possibly more work than all your homework ever combined.

Alternatively, if you don't need the answers to be all -that- accurate, you might try programming an AI to extract a few keywords from the question, then search texts for sentences that have those words in them, then have it return those sentences in the form of a summary. This should be much easier, I'm almost tempted to try it myself.

[Art]

The answers are:
Yes, yes, yes, yes, yes, I'm not sure, yes.

I've recently read some articles that take issue as to whether these "bots" actually "Think" or not.
To say that they recall past information and based on pattern matching, they come up with the best answer.
Isn't that exactly what most of us do? So these bots don't think but we do?
If given a question then arriving at a correct or appropriate answer isn't thinking, then what is?

Is it more important to know HOW one got an answer or the fact that an answer was obtained?

While I think people tend to automatically judge and categorize entities as living or non-living when making the "It's not really thinking" comment I'm not so sure who's actually correct.

There are "expert systems" that "Think" only about one particular subject, automotive, medical, chemical, biological, computer related, etc.
These systems know practically everything about their area of expertise but know very little to nothing of anything else.

Some bots / AI are self-contained, drawing their knowledge from a database on their local computer's hard drive or server. There are others that can scour the net, usually from a number of sources like, Wikipedia, Google, Wolfram Alpha or other data-rich source, then retrieve the best of the best.

Some of these AI's are already alive and well and live in your smart phone, bringing you information whenever asked and usually via spoken word, from recipes, directions for navigation, Text messaging, Posting on Blogs, Facebook, Twitter and just about anywhere else. They can convert any known conversion of matter that one has questions about and can rapidly perform language translations.

Web sites have now tailored sites to suit you whenever you visit, knowing your preferences and what might appeal to you.

Almost scary but this is only the tip of the berg for now.

One thing for me would be for a bot to be able to obtain general knowledge but also to know it's user / creator, inside and out and to be able to carry on a decent "fairly intelligent" conversation, then to RETAIN info from that conversation to be stored and perhaps later recalled as needed.

I think it's coming, and sooner than we think!

Good luck with your research and continuing education. A lot of us never stop learning or asking questions or wanting to know how something works. It is the foundation of wisdom!

[Ben.F.Rayfield]

I read that Google's open source word2vec software is a good base for this kind of thing.

[ranch vermin]

Very good replies... ill just add a link from another guy  http://www.royvanrijn.com/blog/2014/04/creating-a-chatterbot/

[Freddy]

Thanks Ranch (I don't know whether to call you Ranch or Vermin - Ranch sounds more friendly  ).

But yes, good link with some clear explanations.

Welcome to the site btw 

[8pla.net]

Let's discuss what your first C++ chatbot source code might look like...

Code

#include <iostream>
#include <string>

using namespace std;

int main()
{
   string whatever;

   cout << "What do you command: ";

   getline(cin, whatever, '\n');

   cout << "I will do that, " << whatever << ", as you command, master." << endl;
}

Let's discuss what your first C++ chatbot ALPHA test might look like...

What do you command: physics
I will do that, physics, as you command, master.

What do you command: thermodynamics
I will do that, thermodynamics, as you command, master.

What do you command: mow the lawn
I will do that, mow the lawn, as you command, master.


Let's discuss improving your first C++ chatbot to see how it might look like...