Anthropomorphic Robotics Project/Brainstorming/Artificial intelligence and cognitive systems
This is a rough draft. |
Suppose you wanted to design a robot to farm. Suppose you wanted to have an organic farm, so you build yourself a robot with the millions of dollars and hours of time you have to do such a thing.
You build this robots mechanical systems and now you need to program it.
You tell the robot, "Dig a circular hole four feet deep and 10 feet in diameter on the north east corner of our property no closer than 10 feet to property boundary. Put the dirt in trucks that we will bring in as needed." (you want to make a pond to house coy but you are too busy building more robots to actually dig the hole yourself)
So we contract with a group to monitor the robot and bring big trucks in as need to haul away the dirt.
The robot will then have to decode and execute the instructions. This presumes that the robot can walk and knows how to use its body and has somewhat complex motor control.
The robot tells itself:
- Okay. Act...
- Act when? I only act during the day when there is light to act.
- Could I act now in the dark? What would I need to act now in the dark?
- I would need light. Do I have a way of providing light? Yes. I have infrared vision. I can see in the dark.
- "Dig a circular hole four feet deep and 10 feet in diameter on the north east corner of our property no closer than 10 feet to property boundary. Put the dirt in trucks that we will bring in as needed."
- Dig. What is to dig? I will look in my knowledge database....
- So the robot checks its Wiktionary database first. It either does this through a wireless connection it has to a computer in the barn (or house) (or where ever), or checks its local copy of the data base and finds after parsing the page:
Verb
edit- To move hard-packed earth out of the way, especially downward to make a hole with a shovel. Or to drill etc. through rocks, roads, etc.
- They dug an eight foot deep ditch along the side of the road.
- In the wintertime, heavy truck tires dig into the road, forming potholes.
- If the plane can't pull out of the dive it is in, it'll dig a hole in the ground.
- "dig into" - To research a particular subject.
- She is going to dig into Egyptian basket-weaving this semester.
- as slang - To appreciate, or like.
- Baby, I dig you.
- as slang - To understand or show interest in.
- You dig?
So using modeling it asks itself: Is what my owner meant by "dig" most likely that he wants me "to like a hole"? Looks up like... "like" is positivly correlated with "positive emotional valance" but positive emotional valance is very weakly associated with "farm" and "dig"... this is a farm... I do not see "positive emotional valance" correlated with "farm" and "dig".. I do see that "farm" is correlated more heavily with "hole", "pond" and with "plant" which is correlated with "dig" No... he wants me to move hard packed earth.
What do I need? My database tells me that shovels, and heavy equipment are useful for digging. What do I have access to? I have access to a shovel in the barn. What does a shovel look like? According to the database a shovel has the following features associated with it. 4-5 feet long, wood, handle, metal piece on end .5-.15 feet squared in flat area attached to handle. I must walk to the barn and find such an object....
The robot walks to the barn... goes to where tools are known to be kept... looks for shovel... finds object most likely to be a shovel.
The robot looks up word handle - grab handle hold... handle. The robot grabs the handle half way down the handle and is careful to avoid object collision as it walks out the the north east corner of the property. Usings its GPS model it creates a top down map of the property in its on board computer...
It orients itself on the map.
It creates the environment in 3D including itself. It knows where its hands are. It places the shovel on the ground.
"Dig a circular hole four feet deep and 10 feet in diameter on the north east corner of our property no closer than 10 feet to property boundary. Put the dirt in trucks that we will bring in as needed."
Am I ready to start digg? The robot surveys the map of the area and checks it against his vision. The land is "flat", "barren" "covered" - "with grass" - the condition of the soil is "probably soft". Metal can penetrate. There is a truck placed near by. There is also a wheel barrow. The robot decides it is time to dig after assessing no threats to safety of humans or self.
Robot sees "shovel" - knows is highly correlated with "dig".
Examines shovel and earth and runs through simulatory analysis engine to see how to to best utilize the highly correlated "shovel" with "dig".
The physics engine simulates what the final hole will look like.
It works backwards to see the slope of the whole created in such a way to not trap the robot in the whole.
The shape of the metal part of the shovel is correlated with "sharp", and "peirce", and "dig" and "carry", "soil", bang, hit.
The robot examines "dig" and "property" together they are correlated with "soft" "soil" "earth" "soil" "ground" "rights" "ownership" "excavate" "soft"...
The robot runs more analysis...
The robot simulates the movements it would take for it to pick up the shovel. It sees the shovel on the ground next to itself. It keeps this model of it next to the shovel. It see itself bending its knees and lowering its body to a point it can reach the shovel....
To be continued...
Initial thoughts from this brain storming session are that for a robot to have "intelligent" cognitive systems it would need a rather large database of precorrelated terms to other terms. So every term would have all possible correlates arranged by probability, so when "reasoning" a robot would have to be able to effeciently grab terms correlated to objects it its environment and auditory/cognitive/memory systems.
At least for inputs... have one device to listen... one to transcribe... one ...
Reading
edit- Natural language processing at Wikipedia