#4 The 3d internet is coming - an intuitive text search

[BenRayfield]

Audivolv servers - Natural Langauge Mouse Interface (NLMI) clients

http://sourceforge.net/projects/natlangmouse

Natural Language Mouse Interface includes 2 subprojects that need to be merged (after the old 1 is simplified). Both use html divs to move things on screen in realtime. The old code uses html tables, and the new code is faster and only puts 1 word in each html div. The new code is 3d and the old code is 2d, because 1 word can be changed with html css font-size property at runtime. Both change color.

The newest version includes a pure Javascript 3d constraint-based physics engine (around 800 lines) which supports any 3d shape made of words and constraints between the words about distances and which words should be between which others in 3d space. It runs fast enough that you could play it like a game with 40 words on the screen, and I could probably make it faster if it needs to be, but 40 is enough for what I'm planning, because people can not pay attention to more than 40 words at once, and the words can be replaced at runtime based on mouse movements, AI, and physics.

Plans: Using that physics and color and mouse system as a user interface in a web browser, and connecting through Ajax to many servers working together running Audivolv to organize it, the whole system will work the following way:

User goes to http://audivolv.com and sees many words moving in 3d and changing colors.

User moves the mouse over some words and sees it moves the words and changes their color. User is confused at first but quickly learns that moving the mouse slowly and rubbing it on words joins those words into a 3d shape (whatever shape they are in relative to eachother) and they become brighter for maybe 10 seconds.

User moves the mouse fast and is confused why it causes a small number of words (less words than moving the mouse slowly) to move quickly in the direction the mouse moves. User figures out why it is that way when those fast moving words change color to whatever color words it hits and is more likely to join with those.

User figures out how to use the 3d physics, color, and mouse system quickly with no explanation. For most people, it can not be explained well, but it will be intuitive when they use it. (I only have a demo of the 3d physics system now, not a search engine yet.)

Every word will be exactly 1 color, which can change gradually or quickly. When the mouse moves slowly or not at all, whatever color its over gets brighter and spreads to a larger area of words (in 3d space) gradually. Color is normalized by statistics (average, standard deviation, etc) in a colorspace (alpha/transparent, red, green, blue) to make sure no 1 color occurs too much more than any other.

Color is important to the search algorithm. You could think of it as 6 colors: red, green, blue, and the 3 between them. But it will flow gradually between those and have brightness changes. For each of the 6 main colors, that defines an approximate group of words which have some 3d distance to eachother. Every part of the text (except html tags, javascript, etc) of every webpage on the internet can also be represented as distances between certain words. My search engine will be able to search the whole internet for relative distances between words (and ideas represented by words, by using AI) of 40 words at a time (or however many words are on the screen).

In Google, Yahoo, or Msn search, you have to type the words to search, then click a few results and see if its what you wanted, but in my new search engine can be used only by moving the mouse and your intuition of word colors and locations and speeds in 3d space. Words that are closer or the same color will more often mean you want to see those words closer in the webpage search results. Words that are opposite colors and near will represent the NOT operator or something like that.

A more complex example:

Rub the mouse on 2 groups of words. The first group is "politician, airplane" and the second group you rub your mouse on to make a 3d object is "payment, loophole, and 15 other words". Then you use the mouse to accelerate 1 of those 3d objects (made of colored words) toward the other. When they hit eachother, a few sentences and other words appear, including: "lobbied money to pay for his private pilots license" and words you might find in the definitions of all those words but more likely if its related to both groups. It will be like smashing 2 rocks together (made of words in 3d) and pieces of the rocks (new search results in the form of sentences and words, which can also be used the same way in 3d) fly apart at the place you hit them together.

Its a physics system, and words are like particles that make up bigger objects. Forget the instructions. It will be intuitive and fast to find whatever you're looking for or what you did not know you wanted to find in a vague unconscious psychology way. It will use the psychology and mouse interactions of many people to form a collective-intelligence of many people to use their brainpower so I need less computers to run the server-side AI (Audivolv, and maybe some customized AI for this combination with Natural Language Mouse Interface).

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

Collective intelligence is a relatively new area of reserach and has never been used in a search engine to this extreme. Its not a "social network". Its a "content network", more like http://digg.com is a "content network" instead of a "social network". Unlike http://digg.com my new search engine will use Human psychology more directly.

It won't feel like a webpage. It will feel like an intelligent 3d game and paint program and search engine all at the same time.