The body of a robot is sometimes humanoid but frequently not at all. The aim of constructing a robot has primarily been to create a functional "complement" to humans and their needs, and that usually necessitates a human-like structure with appendages that grasp or optical sensors that can "see". But fundamentally a robot needs only the proper structure to handle the tasks it is built for, and while miniaturization techniques continue to improve, robot builders are developing newer, smaller, and more versatile designs.
Unlike computers which simply follow a set of directions, robots are expected to sense their environment, gather information about it, and to change their behavior in accordance with the information they process. Many different kinds of sensors have been developed. The first type of sensors are ones that robots share with humans, or the five senses (hearing, sight, smell, taste, touch). Aural sensors can detect the faintest sounds, and light sensors are also highly developed. Some specialized robots can detect the presence of water and other liquids, some have thermovisual scanners, and still others are equipped with ultrasonic senses. Touching sensors can determine the difference between hard and soft, and gripping mechanisms are used in a variety of areas, from toxic sites to the space shuttle.
The other type of sensors are those that measure sensations beyond human recognition. Some robots can measure higher and lower pitched hearing, while others being developed may detect x-rays, gamma rays, infrared light, radar, and radio waves.
Perceiving the surrounding environment is a field that artificial intelligence researchers call pattern recognition. The problem is thus: A robot can sense its environment reasonably well, recognizing where objects are and how to avoid them. But when a robot is asked to perceive what it is seeing and if it has seen this object before, the task becomes much more difficult. The main idea of perception is the filtering process. For example, 600 trillion light rays strike the human eye every second. Some of these rays clump together to form a chair or a lamp, and we perceive these objects as independent of the surrounding environment. Robots have difficulty filtering these light rays into separate objects, and even when they are successful in this respect, perceiving them is more of a problem. How can a robot know that a bean-bag chair and a bar-stool, two visually dissimilar objects, are both used for sitting?
Memory today is reasonably cheap and efficient. Robots need a good quantity of memory to store their directions and to access previously collected information. In 1986, Derek Kelly hypothesized that "a really general-purpose robot is probably going to need at least 3000 megabytes of memory to be able to talk, understand, see, hear, perceive, and act intelligently." In retrospect, the most advanced robots twelve years ago used only three megabytes of memory. Todayís home computers have 6-12 gigabytes of memory. So memory is probably not a problem. But questions remain: will we store every possible procedure for every function a robot would ever be asked to perform? Or do we develop general procedures and a system for applying them to specific situations? Do we store the entire dictionary, or perhaps a dictionary for every language?
Computers are built to follow rule intelligence like no other. They take rules given to them and follow the rules repeatedly in an organized, direct fashion. Parallel processing has created computers that act more like a human brain when performing computations. But do we want robots to have brains similar to ours, or is a direct approach better suited? Robotís should follow the same rules consistently, but are Asimovís rules the ones to follow?
To make robots more human, or to have a robot follow its own directions (that it was not directly fed), the robot must have creative intelligence, or what one author calls "the ability to break the rules." For a robot to be competent in the area of creative intelligence, it should be able to take data and information of any kind and hypothesize about it, relate it schematically, carry out permutations of the data, create combinations, aggregate the data in different ways, or correlate the data. This capability is far-fetched for a computer to acquire, for it would mean the computer could ponder random thoughts or play with words in the form of puns.
Although in most science fiction realms bad things happen when robots acquire emotions, research is being done on the subject of mechanical emotions and robotic emotional capabilities. We could someday have robots that get agitated, excited, passionate, sensitive, or that have "gut" reactions. The thought of a touchy or agitated robot frightens most, so it is our tendency to claim "robots canít have emotions", but that remains to be seen.
Similar situation, similar consequences. A robot that feels can appreciate the significance of events, but is also susceptible to failures and breakdowns. But what about love?...
To be imaginative is to turn inward for inspiration in oneís actions, to go beyond what is actual and what is possible (from a human perspective). We as humans cannot understand human conscious capacities , so it is unlikely that we can mechanize them.
Intuition is another of humanityís capacities that no one truly understands. But if imagination, feeling, and emotion are feasible to some, intuition must be also.