REGAI - Blueprinting regex-AIs

[NeonSturm]

News:

regex function reference : not available jet
regex accelerated processor : not available jet

current tasks:

• task name - description
  • (url=post)user(/url) #These peoples work on the issue, links thread-ops + posts where possible.
• Thread-OP - keep this thread up to date and think about it more.
  • NeonSturm
• regai_commands
• regai_commonsense
• regai_parser
• regai_cc
• regai_speak

Intro:

Before I started this thread, I posted somewhere else and I think it fits here too:

, but if the "display-block interface" improved enough along with "display-block regex-AI" the might soon be able to.

If you don't beg shine to improve AI but instead suggest to implement the interface improvements for display-blocks to game-variables where they aren't sufficient to allow us to handle it with regex-logic-AI, we could reuse it in other regex parsers unlike shine's obfuscated code.

I will support such a suggestion when I see it and maybe it's time for some Community Content: proper regex function references (similarly to what PHP, Java and X2/"X3-Reunion"-Macros provide).
I will try some tasks for the AI-part once regex is done (researching on human language support for a few years now), but I am not used to work with regex jet.

There is another relevant thread:

RegEx compilation thread
But until now the OP only has one easy-to-use regex "Line parsing" for removing first/last line.
What we need is much more.

I like to quote the PHP manual, because it is very useful, clean and easy to use - samples:

• "chop — Alias of rtrim"
  
• "explode — Split a string by string"
• "substr — Return part of a string"
  • "string substr ( string $string , int $start [, int $length ] )"
    • for example, $start can be negative and counts from the end then which is great after you used strpos (to "find the position of the first occurrence of a substring in a string")

( …to be … continued … in the next post because of expected post lengths in the future )
[doublepost=1487092649,1487092591][/doublepost]( … continue … to be … well, I can blame the forum for future info fragmentation )

Major Tasks:

Note: if you find something that looks like it's Java-code or similar, it means that I use a way to describe the structure of it (.variablename, (brackets), FIXPOINTS important to the processor)

regai_commands are the very foundation of higher AI-scripting - without these it's impossible to start.
Some examples:

• db_find .word
  • returns (index/position to fetch data from)
• parse_next
  • returns (WORD "" || INT 0 || FLOAT 0.0 || …)
• think_about .word
  • returns (.word (NOUN || VERB || ADJECTIVE || …))
• think_from .word to .target
  • returns ( ((.word .target) DISTANCE) 1, ((.word .target) STRENGTH) PROBABLY )
• speak_sentence .word1 .word2 .word3 .word4
  • tries to speak a sentence containing these words
  • will use the CURRENT_CONTEXT of these words
    • includes information about if a word is a NOUN, VERB, …
    • includes information about the connection of these words (DISTANCE, STRENGTH).

regai_commonsense defines:

• sentence structures - examples (VERB, NOUN)
  
• default wisdom - definitions of (is, not, equal, computer. do this, computer. do that, …)
• …

regai_parser to structure logic.

• word word word - separated by spaces.
• word, word, word - a list of words
• word, word and word - a list of (word, word) becomes a list of (word && word && word)
• word, word or word - results in (word || word || word).
  
• …

regai_processor speed booster - reduces the display block count required to define the AI's function by providing higher logic and compressing the length of expressions - examples:

regai_context or regai_current or regai_cc - (haven't decided jet) but it should handle the thing we call short-term-memory.

• CC <> CURRENT_CONTEXT - use an alias for CC (both ways).
  
• CC .word .otherword - .word was recently used with .otherword
• CC .otherword . word - don't forget to link otherword backward!
• CC .word NOUN - .word can now be used as a noun in sentences.
• CC NOUN noun - .word is a NOUN is a noun, DISTANCE=2 (over NOUN).

​

regai_speak - is responsible finding good answers.

Do you notice how I used "DISTANCE=2" the in regai_processor example?

If the question is "Do you like to have a cookie", "Yes I like to have a cookie" is closer than "I want a cookie" because that answer contains different words which are further away.

However, you may still get the answer "yes, give me cookies!" from a human because humans have a personality and prefer some words over others.

Humans sometimes forget intermediate steps and think simply: "cookie yes/no? → I want cookies"
then "yes, cookie, (give me)" before they form a sentence "yes, give me (a) cookie (please)" which turns to a simplified form or a formal form depending on the mindset.

• answer_quick - statement "yes" or formal "sir, yes sir!"
  
• answer_short - telling a possible flaw "I guess … but this is magic for me".
• answer_medium - "If we speak about the same, … . But this means … . And there are 41 more questions about the meaning of life"
  
• answer_assumptions - "this is true if …" or "if … then this is true".
• answer_questionFalse - "I think yes, is it the red one?"
  
• answer_long - For example the contents the regai_speak paragraph as answer.

Finding the right answer depends on multiple variables, but mostly on answer-length, severity of the topic and the type of the current interaction (forum post, military call / appeal).​

Everything you add will make the AI more complex and bigger.
But a certain minimum is required (short-term memory, etc) to have an efficient assistant, other things are for luxury only (showing personality).
How human-like do you want your AI?

• have personality
• tell "emotions", have "global emotions" or just local opinions
• be self-aware (self-similarity) or too dumb to see the self-similarity
  
• have favourite words or phrases or states (put/sulk if getting told "you can not")

… to be … continued …
… continue … to be …​

 

[NeonSturm]

Update:

regai_search modifier brainstorming:

I said, AI could be a path-finding algorithm, so how does search work?

You just reduce the distance for your search term:

in a folder like:
music:downloads / creator:Mind.in.a.Box / album: / track:Supermassive Gravity

Whenever you encounter {music, creator, album, track} and it matches {downloads, Mind.in.a.Box, ANY, Supermassive or Gravity}, you may disable use of that entry again and reduce distance by 2.
distance = 1
encounter music:downloads
new list {NONE, creator, ablum, track} & new distance = -1​

Effect:

0 / downloads / Mind.in.a.Box / albumX / Supermassive Gravity / 5
0 / regai_search(args) / downloads / Mind.in.a.Box / albumX / Supermassive Gravity / -1

a negative or 0-value will bind the search-term to the target.
​

Update:

regai_preload brainstorming:

Included Tags
Excluded Tags

Database:

Chunk : 4546.movie « 6135.startrek « 6843.starwars
Chunk : 8615.group « 4546.movie « 3612.game
Numbers are absolute indexes similar to the record-time of a security-camera video.

All entries are double-linked for when they have to be moved.

Double-linking can take advantage of absolute-chunks and relative-addresses.
Chunks and formatting-thresholds reduce overhead and formatting flux as well as load times and required display blocks and an entry can be present in multiple preload-chunks.
Database tag-entries might be fragmented with topic-irrelevant neighbours, but the preload-chunks serve as cache for specific topics.
​

An efficient implementation computes heavy-loaded entries with

1. first timestamp or an index referring a timestamp
   
2. recent timestamp & recent request counter for activity
   
3. previous-recent timestamp & request counter for activity threshold tracking.
   
4. These heavy-loaded entries are then becoming partners of entry-chunks for benchmarking purposes (grabbing new-lonely entries and ditching least-used ones from chunks)

Geometry is the Formal Structure of Thought.

 

[NeonSturm]

I have found a good video telling about the complexity of human thought.
Maybe it is useful to see what AI-programmers are required to archive as a final goal.
At youtube

And now, I have found more information about how brain cells work.
If they are sensitive to the time of input and analog, simulations of multi-neuron networks could be replaced by a simulations of single more complex neurons. (( Our brain is ten times more active than previously thought ))