Close

Goal Finder (interact, process, share)

A project log for Hackaday Prize Entry: AVATIFY

Desktop Buddy 4 Time Management AI (Web App) to get out of your Procrastination-Trap

kennyindustriesKenny.Industries 05/29/2016 at 13:380 Comments

A fact: A person needs 3 things:

1) Goal

2) Plan

3) Todo

But it's rare to have them all, because these are logical and thus requiring focus and conscious thinking. This is for many people very difficult. Therefore, the avatify-programme does that for them.
In this log, i will discuss the 1) Goal part.

PART A (Eliza)


To make this possible, we need interaction. This interaction will be based on Eliza-System by Weizbaum, but with different questions. The 4 core questions will be:
- What do you want to achieve?
- Why do you want to achieve it?
- How important is it to you?

- Does it have a deadline?

The Following is less scientific, but I assume it will work that way, on which I will be working on the next few months.


PART B (NeuronPool)

All these input will be stored as a corpus and will be fed to the "NeuronTranslator".
It connects the many artificial feature-neurons like Part-of-Speech-Neurons and ontology-knowledge-base-neurons and wordnet-neurons. According the connection and their strength of bond, the inputted information will be stored implicitly. This means, at the time, we connect the translated new neuron-chunk to the existing local neuron-knowledge-base, there will be a "spike-wave" through the local neuron-knowledge-base. At some point it will stop propagating. The "border" where it stops is the meaning/interpretation/semantics of the input.
So if we pushed artificially the border +1, there will be a semantic prediction or suggestion. By doing this, the input to the system is kept at minimal and through suggesting, getting the significant idea will be easy as pie for the user.

PART C (Crowd Sourcing)

In part B I mentioned "local neuron-knowledge-base" meaning the user-specific storage of the information. By correlating the inputs and suggestions of other users (anonymously) there can be extracted many relations specific to the domain of planning. But this requires an already annotated corpus, which I do not possess. First experiment will be a new years resolution, but I assume that it will not be sufficient.

Therefore I will introduce "help-points" into my system. Basically, there is a global "job-hunting" list, which will be ranked by individual skills of each user. If there is a task, most suitable and the helping user is willing to help, he can earn such a point. The one, whom is helped, pays one or more help-point.

Such a task could be to do the "border + 1" of Part B, if the system fails to predict. This is at is essence very simple and clear, which is the obligational part of crowd sourcing or micro-tasking.

In a sum, the extracted relations and the humans supporting the system to improve and to extend, a "GLOBAL neuron-knowledge-base" can be created, which is the main goal of this software-module.

Discussions