D
Deleted member 1791944
Guest
Me and my friends have come up with something, and I hope you could help criticize our idea before we waste months worth of effort!
TCwHCC :
Thinking
Computer
with
Human-like
Capabilities and
Characteristics
TCwHCC or Kwack in short, would be a computer that emulates the way an average human thinks and learns.
These are the components we've come up in order to make this work:
made using logomakr
Memories would be separated into 3 categories:
1. Object - this is something that could be held and felt.
2. Concept - an idea that is thought.
3. Action - something that could be done.
*more categories might be added or changed by Kwack if it actually learns
Kwack's Learning Algorithm goes like this (rough summary):
1. Obtain / Gather the input.
2. Compare input with knowledge.
3.1 If nothing is similar with the input, learn by communicating with humans/looking it up on provided internet sources/simply crawling through the internet.
3.2 If something similar is found, create a relationship between the two and form an assumption on how they are related.
4. Repeat
Here is a sample situation:
1. I wave to Kwack's camera.
2. Kwack uses computer vision to identify the person and their action (inverts image and turns it into a sketch).
3. It uses reverse image search and gets a word from (say google) its source or maybe it compares the sketchy image with past knowledge.
4. Learns more about the wave action, seperates the word and the actual action, but keeps the relationship between the two.
5. Turns it into knowledge.
The 2 big obstacles we believe are:
1. Hardware - because as I am typing this thread, I am currently running a cheap pc bought from a closed internet cafe (though Cloud Computing is possible).
2. Efficiency - since we didn't do our research and only thought of algorithms ourselves, we aren't exactly sure whether they're efficient.
Yes we will put few things before actually running it. (Knowledge on basic human concepts, objects and the english language).
No we will not be using pre-made libraries, everything will be from scratch.
TCwHCC :
Thinking
Computer
with
Human-like
Capabilities and
Characteristics
TCwHCC or Kwack in short, would be a computer that emulates the way an average human thinks and learns.
These are the components we've come up in order to make this work:
made using logomakr
Memories would be separated into 3 categories:
1. Object - this is something that could be held and felt.
2. Concept - an idea that is thought.
3. Action - something that could be done.
*more categories might be added or changed by Kwack if it actually learns
Kwack's Learning Algorithm goes like this (rough summary):
1. Obtain / Gather the input.
2. Compare input with knowledge.
3.1 If nothing is similar with the input, learn by communicating with humans/looking it up on provided internet sources/simply crawling through the internet.
3.2 If something similar is found, create a relationship between the two and form an assumption on how they are related.
4. Repeat
Here is a sample situation:
1. I wave to Kwack's camera.
2. Kwack uses computer vision to identify the person and their action (inverts image and turns it into a sketch).
3. It uses reverse image search and gets a word from (say google) its source or maybe it compares the sketchy image with past knowledge.
4. Learns more about the wave action, seperates the word and the actual action, but keeps the relationship between the two.
5. Turns it into knowledge.
The 2 big obstacles we believe are:
1. Hardware - because as I am typing this thread, I am currently running a cheap pc bought from a closed internet cafe (though Cloud Computing is possible).
2. Efficiency - since we didn't do our research and only thought of algorithms ourselves, we aren't exactly sure whether they're efficient.
Yes we will put few things before actually running it. (Knowledge on basic human concepts, objects and the english language).
No we will not be using pre-made libraries, everything will be from scratch.
Attachments
-
You do not have permission to view the full content of this post. Log in or register now.
Last edited by a moderator: