If only it were that simple! However, in the true world of artificial intelligence, even an easy job like fetching a couple of dry socks from the hold of a space shuttle is filled with unintentional risk. Because beyond the stage of a space program, real robots have trouble with the types of principles, even a two year old would find very simple to understand. The appearance of artificial intelligence is that it is not truly intelligent as we see it.
A person’s brain is not only clever-it is clever in an extremely particular way. Even though a computer can add millions of calculations per second, a person’s brain can use logic to solve problems that no computer can. The reason for this is that true intelligence is created to make leaps of relation and intuition that are very hard to note into the types of algorithms a computer can comprehend. One of the best methods to get this is to examine an easy problem, such as R3 getting Luke Skywalker a pair of dry socks.
The Intelligent Agent
One of the biggest aspects of an artificial intelligence is the capability to be an intelligent agent. What an intelligent agent is is an artificial intelligence that is able to take in its surroundings and can decide what it must do within that surrounding to accomplish its goal.
This is not very easy at first. For instance, the action of being aware of the surroundings is much harder for a computer than it is for a living being. Even though the only sense used to perceive this is sight.
Now R3D4 must find the socks in Skywacker’s space shuttle. The first issue is that its cameras do not work in the same way as Skywacker’s eyes. A PC’s “eyes” – are also called a machine vision system, need:
1) Cameras (digital or analog), with the capability to keep images. One camera will suffice -if you add two for binocular sight, it will make the situation more complex.
2) Exceptional light sources. The reason for this is that PC networks are very selective about sensing light in particular ways. If, for instance, soft light makes a red sock appear black, a machine vision system might experience a breakdown immediately.
3) Programs in which to handle pertinent features of the perceived item. This is far more difficult than it sounds; a computer must keep a matching image of the item being regarded, or it will not recognize it, because computer neural networks cannot conclude an object’s class from just generally view it.
All is good so far. However let us pause a while and discuss necessity number three. Where Dook’s eyes look at an entire image and instantly understand it as an object, SOCK, R3’s camera views the sock as a bunch of pixels. It will have to put those pixels together into one image, and then transpose that combination of pixels into an image it recognized as being a sock. Sound easy? Not likely.
R3 might have a photo of a sock in its memory storage. However, it is just that single photo. What if the sock in its memory is red and not blue? R3 will not see it because it does not match the photo. What if the sock in the memory is not folded and Dook’s sock has been nicely rolled up by faithful Princess Cinabuns? Again, R3 will not see the sock because it does not match the photo.
How is Dook able to recognize the sock no matter what it may look like? It is because Dook is able to make assumptions about the sock. He has a mental image of what “sockness” means in his mind. All he has to do is search for defining signs such as L-shapes, tubular shapes, made of cloth, one side open, and he can also designate his rolled up red socks to the “sock” column.
“Pattern recognition is one of the largest walls to building true artificially intelligent robots.”
This is termed, as “pattern recognition” Pattern recognition is one of the largest walls to building true artificially intelligent robots. Pattern recognition is the capability to put together that mental image of an item or surrounding from several cues that the item gives.
Pattern recognition is a very difficult program because every solitary pattern has several elements that can alter the pattern’s cues. To the exact mind of a computer, a sock being in shadow can completely alter the nature of the sock’s pattern.
However, to a human mind, the distinction of shading easily becomes reassigned to the column of “shade” with the inherited idea that items in shade are darker than items in light. (As a funny side remark, this type of thinking is known as “fuzzy logic.”)
To enable R3 to locate Dook’s sock will take an exceptional type of programming, a type of neutral net that is known as cognitive architecture. A subsidiary of the concept of intelligent agents, cognitive architecture is developed by situating many types of “thinking” programs into one bigger control program. For instance, one section of R3’s cognitive architecture must include the concept of visual states, such as the concept that light can influence the visual pattern of an item.
This program would view the general light given off in an area and would be able to tell whether a shadow was there. Another section of the CA program must recognize what theoretical characteristics determine what is “sockness.”
cognitive architecture is developed by situating many types of “thinking” programs into one bigger control program.”
This section of the CA would search for characteristics of texture, tubes, open ends, and common L shapes. If the item being looked at has enough of those characteristics, R3’s neural net would be able to declare, “Hmmm. This matches the criteria of what determines a sock,” it can then categorize that item as a sock.
Finally, R3’s cognitive architecture must know how to test that so called “sock” against the data that it has kept on socks. For instance, the same elementary characteristics of sockness (rough texture, tube with L shape, open mouth) can as well be used to describe a snake.
Think of Dook’s shock when R3 comes with a ten-foot python back to the puddle. As a human, Dook is aware of methods to test his hypothesis; if the “sock” slithers and hisses, it is most likely not a sock.
As well, R3 will require a section of his intelligent agent program that can test items, gaining further information that can clarify whether it carrying a boa constrictor, or an argyle.
Similar to the way a baby shoves stuff into its mouth, R3 will require ways of experimenting with items in its surroundings, and gaining knowledge from experience that any “sock” that has sharp teeth and hisses is likely not a good selection.
Will Dook don a cozy pair of pythons? Not likely, however he might want his trusty laser sword nearby when R3 comes back with the dry pair has asked for. Nevertheless, as we have shown here, even an easy job like getting a pair of socks can be a very trying task when artificial intelligence and robotics mix.
For the near future, it appears as though the service of our handy robot friend will be restricted to the factory area and the laboratory, where the elements of machine vision, cognitive architecture, and pattern recognition are better harnessed.
Eventually perhaps, R3 will be clever enough to tell a pair of socks from a snake, however for the moment, Dook Skywacker will have to march to the ship and fetch his own dry socks.