Songs without Words

A further refinement of the idea-net concept; moving the theory into actual code.

Proposal 1.  Better than node-counts:

In order to determine the maximum “radius” of search, node-counting is easiest to execute, but can also be misleading.  Thus, the “searching” function should be constructed in the following way:

ltm.searchNet([list of words] (or ‘string’), maximum distance traversed, maximum nodes traversed (default 0), maximum search results (default 3), *options)

Examples of options include:

1.  Contradiction handling.  Suppose that word A leads to B, and C leads to D, but while A and B agree and C and D agree, B and D disagree.  Then, are B and D, which are very close but contradictory, acceptable results?  yes/no

2.  Exclusion of bad routes.  If the maximum distance allowed is 20, but single connections of distance 16 are to be excluded due to excessive length, then this setting should be activated.

By setting a maximum distance, one is able to gather all ideas that are reasonably related.  As an analogy, one is technically very closely related to all of one’s relatives.  But in practice, not all relatives are that close, and friends of friends may be closer than siblings of parents.  As a result, it would be more accurate to poll those who are close, not those who are necessarily directly connected.

Proposal 2.  The binary distance system

The distance system assigns a default “high” value to a “fresh” connection between ideas.  After repeated exposure to the connection, this value is whittled down.

However, there are two problems with a linear system:

1.  It eventually could reach 0, which is not useful at all.

2.  It does not reflect the actual state of things: the second exposure is key; by third exposure, we are willing to believe something.

As such, the proposal is this:

if connection is True:

connection.halveDistance();

Basically, you have a default distance of, say, 16.  This distance is then halved to 8, then to 4.  By the time it reaches 4, it is within a reasonable distance to be confident.

Proposal 3.  Separation of confidence vs. nature

The “distance” concept must be treated independently of the nature of the connection.  It is easy to get confused between the *type of relationship between the nodes* and the *confidence with which one senses the connection*.  As an example, let us suppose that Caroline has just learned that one might refer to a language as a “tongue.”  She establishes a connection (‘language’,’tongue’,16,0).  The 0 is the relation: it represents the conception that language and tongue are synonyms, or true equals.  But the 16 is the distance: she is not yet sure if she has picked up the right connection, so she is not confident in this connection.  As another example, suppose that Caroline learns about love and hate.  (‘love’,’hate’,0.5,-3).  The distance, 0.5, means that Caroline intimately knows the relationship between love and hate.  But this relationship is highly polarized: -3 indicates the strongest repelling between ideas, and so Caroline understands that as concepts, they are as far apart as possible.  They are only close in “distance” because they have a strong connection with one another.

Proposal 4.  Deletion of weak connections.

A lot of connections that will be made will be flat-out wrong.  People might lie or say things incorrectly, or use incorrect diction; or, a connection might be made between two words erroneously linked.  So, to counter the amassing of false information, it is prudent to weed out weak links at the end of the day, the same way that people do.

Basically, the algorithm might work like this: weed out 10 to 20% of 16-distance connections and perhaps 1 to 2% of 8-distance connections at the end of each day.  This might seem counterproductive: what if something useful is lost?  And yet this is a necessary measure.

Why do students studying for a test seem to spontaneously forget things?  It’s because of this same mechanism: the brain discards what it thinks are poor connections in order to allow for revision.  This is also the same reason why we are flexible – why we change our opinions or correct old misconceptions.