Introducing ArckonArckon is a computer program with artificial intelligence (AI), basically a digital brain. The program communicates through typed text, learns from statements and answers questions. Arckon is not built for casual conversation but for learning, reasoning, and solving problems through intelligent discussion.
Ultimately such a program might serve as an expert advisor on issues such as climate change, but at this time Arckon is mainly a talkable knowledge database. The program is not publicly accessible as this is a development project that is not meant for public entertainment.
Demonstration► 2014: Arguments & opinions
► 2013: Language skills (rehearsed demo) ♪
copyright © 2014. Do not redistribute or host this content elsewhere without permission.
What can he do besides talk?• Learn new words (automatically or by explaining).
• Learn new facts (automatically)
• Learn by reading text documents (300 words per second)
• Basic reasoning (inference, generalizations, etc)
• Summarise documents & webpages (sentence extraction)
• Visual object tracking, colour recognition, motion alarm
• Virtual assistant functions: Schedule tasks & reminders, tell time, do math, play music, open files & programs, automate computer tasks, weather forecast, custom tv guide.
• Spell correction, word games, anagram decryption
• Voice interface (Windows speech)
One example of practical use is to have Arckon read a large text document, making him an instant expert, after which he can give you direct answers to questions about the subject matter.
How does it work?Arckon was built from scratch with C++ and uses a custom structured database of text files.
Arckon understands English language through a vocabulary, word morphology and grammar rules.
He extracts a.o. the subject-verb-object facts, then checks and/or saves them in his database.
If Arckon does not know the answer to a question, he will attempt to infer it from related facts.
Optionally Arckon may engage a set of social rules to keep conversation rolling. All potential answers are checked for their relevance to recent context and sorted by priority. Finally, Arckon formulates his own sentences by feeding facts through a grammatical template. The only pre-programmed answers are two dozen standard manners, like "Thank you" and "You're welcome".
ConceptThis project is about creating human-level intelligent thought processes. I am however not concerned with the undebatable definition of "intelligence", the biological composition of a human brain, or recreating all aspects of humanity like emotion. I only recreate logical processes as I perceive them in the only brain that I know intimately: My own.
Rather than programming exact instructions or gambling with unsupervised learning, I try to handle each aspect with the most efficient means:
• Rules: I did not learn languages on my own, I was taught grammar and spelling rules in school for each one. It is only logical to equip AI with these rules from the start and save time.
• Learning & reasoning: The knowledge of the world is too vast and variable to enter manually, so the AI needs the ability to learn and adjust, while intelligent reasoning helps fill in the blanks.
• Supervision: Human language is inherently flawed and interpretable. Communicating in human language will inevitably lead to misunderstandings, so it is necessary to occasionally correct the AI as one would a child.
Re-inventing wheelsMost of what I have been doing has apparently already been done in the 70's, 80's and 90's. Nevertheless it took me but a year's time to come up with these same methods in a working program. The point is simply that I learn five times faster through practice than through study, because that's just the way my brain is wired. (re)inventing may seem unnecessary, but it is in fact very insightful and keeps me motivated, a vital factor for any project of this size.
As I have no official credentials in the field of computer sciences, I don't expect anyone to take my word. Occasionally I compete in the Loebner Prize contest for artificial intelligence to demonstrate my progress, and I have also entered the Winograd Schema Challenge. I describe Arckon's performances in these events in my blog.