Ai Dreams Forum
Member's Experiments & Projects => General Project Discussion => Topic started by: paphus on June 01, 2015, 02:51:35 pm
-
We are crowd funding on Kickstarter to create a new non-profit organization for the pursuit of "free open artificial intelligence for everyone".
https://www.kickstarter.com/projects/paphus/bot-libre (https://www.kickstarter.com/projects/paphus/bot-libre)
We would like to develop a free open source AI platform. To start the project off, we would be open sourcing our BOT libre platform into the project.
http://www.botlibre.com/ (http://www.botlibre.com/)
The project will primarily be Java based, both server and Android, but will also have SDKs for Objective-C/iOS, PHP, JavaScript, and other languages.
If anyone is interested in supporting the project, or becoming a developer on the project, we would appreciate your support.
-
(excuse me.) Its really repulsive advertising, why dont you skip the link, and just tell us what it does without the cruddy music.
-
I assume your talking about the video... I think it is kind of cute,
The end goal of the project is strong AI. The interm goals are to provide a set of AI services such as chat bots, virtual agents, virtual assistants, social media automation, email automation, game AI, etc.
Currently the BOT libre platform supports:
- chat bots / virtual agents
- avatars, text to speech
- live chat integration
- email automation (Email bots)
- Twitter automation (Twitterbots)
- AIML
- Self (state based language parsing language)
- integrated knowledge graph/database
- importing from Wiktionary, Freebase
- web API, Android SDK, JavaScript SDK
We also have under development, Facebook automation, AIML 2.0, ChatScript support, and iOS SDK.
-
So... thats what it automates.
what is the means of its automation?
-
What do you mean by means?
The purpose of automating chat, social media, or email are many.
The main commercial usage is customer service, to provide 24x7 service, help users, etc., also marketing, games, dating, fun. Basically increase presence, reduce wait times, reduce costs, improve usability, etc.
-
Nice project, I like that you have something concrete that works :)
What idea do you have on reaching strong AI? In particular, *maybe* I would be interested in the project if I could get explanations from the AI to questions like:
* What compounds under what circumstances should be mixed up to get artificial food?
* How the Universe works and what are its rules?
* What is life and how it relates to un-alive matter?
-
I meant means by what is the "intelligence method" for doing the chat. I like getting technical, its more fun, even if its simple or overdone, everything I do is that. :)
-
I quite liked the advert myself. Good luck with it.
Nice to see something that actually works instead of just all the theory and ideas that tends to dominate these boards nowadays.
-
very good work! thank you very much friends! I'm going to try. O0
-
RE: getting technical, and ideas for strong ai
The system is a hybrid right now, it has support for a lot of different things, from AIML and a little bit of ChatScript pattern matching, keyword response searching, language parsing, and a knowledge graph. It is written in Java. Each bot on the platform has its own Postgres database which stores everything in a graph like object database structure (you can browse its 'brain' from the Knowledge page).
For strong AI, I'm open for ideas. Currently I have been object modelling the various components on the human brain, and slowly flushing them out and integrating them.
-
I think I know what you mean by knowledge parsing, extracting it into a database, simple knowledge. But what does it do with it?
Ive got a version of knowledge parsing that I just thought of then.
The computer is looking at different ways to bring up the same following words.
So if it sees "the egg smashes" and "the hammer smashes" it then decides to say "the egg" is interchangable with "the hammer" because they both smash. So it gets a link to say either on the occasion of wanting to bring up the egg, or the hammer.
This is nonsense, but if you could do this so it made sense, it would be excellent for your chat bot. Its first learning is slow, but as it increases its knowledge of these interchangable sentences, it can use its last interchanges to increase the learning speed as it goes!!!
And the whole point is increase the key size of your markov chain without sacrificing interactivity.
-
Paphus, what are your groundings on induction, deduction and abduction. I think that strong AI should be able to do those three processes, but that opens a thousand new gaps. How hard/easy is to implement those in BOT libre?
-
> induction, deduction and abduction
The bots currently have a Comprehension process that is capable of some of these. It basically analyzes learned questions and responses with its knowledge base and attempts to generalize responses or input processing based on meanings of the words in the phrase and the relationships of the data. If it finds a generalization in will alter its language processing state machine, or create a formula response. For example the bots can be taught to count.
Much more work needs to be done in this area.
-
I like what I hear there, sounds like alot of effort went into it. (Sorry for being rude, but in my old age im rejecting alot of society.)
Strong AI would be able to learn on what it learnt, thats the big deal for me now.
-
Paphus, I'd like to share a paper with you. This is something I've been working on for a while, and it might be of interest for you (it is already shared on this site, but here it is again): a very technical paper (https://docs.google.com/document/d/1Q8AxXFFgvGb9cHbpyFCUIzz1IDBJu2Wc1q2yUuVdGEE/edit?usp=sharing)
The paper is a little bit outdated (I'll compose a fresh one if you want), but you can get a feeling of what can be done. I plan to release the essence as a javascript library for whatever AI use. I might go free and opensource with this one, I'm almost certain of it.
I just thought that this could be interesting for you, as you are pursuing strong AI and the paper explains induction (a sort of) and deduction. Abduction is just a reverse deduction, so you can consider it covered when you solve deduction. The most important thing is a knowledge base storing mechanism which I solved with tuples and choices. It seems that this method is general enough to hold any sort of information. Once you got it right, you can do whatever process on top of it. In this case just enter a data set and you get deduced all conclusions by defined rules.
-
Ivan, I like your style for generating equations for loose samples on a graph, but i dont think you can develop anything you cant just zap in from the sensor, symbolic logic wise, BECAUSE you cant justify anything new, anything that you would have justified would have been able to been simply recorded. youd just have noise logic left over that half passed the test, which wouldnt be anything useful.
-
Ivan, I like your style for generating equations for loose samples on a graph, but i dont think you can develop anything you cant just zap in from the sensor, symbolic logic wise, BECAUSE you cant justify anything new, anything that you would have justified would have been able to been simply recorded. youd just have noise logic left over that half passed the test, which wouldnt be anything useful.
I've given this a thought before and I think it will work like a charm. After providing parsing rules (which you generate by statistically analyzing input sensors or you provide hard-coded i.e. english grammar) you have to parse a list of bytes from a sensor. Then you get a notions tree (like abstract syntax tree) that recursively groups input bytes into bigger particles with which you can do an arithmetic of transforming expressions to deduce implicitly contained facts (i.e. you can predict other states). You can assign a keyboard terminal to a sensor, or any other time-depended input stream. Then it is all in discovering rules and parsing input data to generate a knowledge base.
-
Well I cant say ive put as much thought as you into it, maybe starting off from some hardcoded knowledge it can use with the sensor maybe something will develop?
Implement it, then you know.