Researches are going on across the world to create machines which will surpass the human-level intelligence. These machines, known as super-intelligent machines, will employ the new technologies, obtains super intelligence and knowledge and hopefully serve the humans better. Scientific community is divided as to whether the super-intelligent machines will remain a dream or will become a reality. This article explains expectations from super-intelligent machines.
What are Super-intelligent Machines?
Intelligence is the ability to learn and to solve problems. Super-intelligence is the superior ability to solve problems and even to recognize problems. To qualify as super-intelligent, an intellect must be much smarter than the best human brains particularly in every field including scientific creativity, general wisdom and social skills. Super-intelligent machines are the combined productions of neuroscience and computer science. These machines can think, decide, see, feel, react, and communicate with other similar machines and with humans. They will be in continuous touch with central command and control. Intelligent machines will learn much quicker than humans. Once a small number of intelligent machines are produced, they will produce the similar machines on their own.
Speed or the processing power is the most important capability a super-intelligent machine should have. Enormous computing power is required to search the huge knowledge base, to process the images, to solve complex and real world problems and to react in time to the changes in the environment. The human brain has 100 billion neurons, each has over 1,000 connections to other neurons and each connection is capable of performing about 200 calculations per second. So far, no one could calculate the processing power of the human brain accurately. In one estimation, it is approximately, 100 teraflops. The processing power of Blue Gene/L supercomputer already crossed 280.6 teraflops – that is 280.6 trillion calculations per second . So, as per this estimation, today’s machines already crossed the brain power.
A machine which is similar in computing power to the human brain requires ready access to sufficient memory. That is, it should match with that of human memory.
The memory capacity of the human memory is calculated as,
Memory capacity of Human Brain = 100 Billion (10^11) neurons * 1000 (10^3) Connections/Neuron * 10 bytes (information about connection strength and address of output neuron, type of synapse) = 10^15 bytes = 1 PB = 1000 TB.
So, the size of the human memory is estimated to be about 1,000 trillion bits.
Problem solving ability
Compared to humans, computers are faster and accurate in doing complex mathematical calculations. But that is not enough since the nature of the problems humans need to solve are not always mathematical problems but real life problems. Logic does not always solve the problems. Emotional intelligence also plays a major role in solving many real life problems. By developing complex and efficient programs and by using machine learning techniques it is possible to make the machines solve almost all the problems.
The sizes the super computers are so big that they occupy a large room. With this big size the dream of super-intelligent machines cannot be realized. The circuits made of silicon have some limitations. The density grows and so the heat and amount of interference. Sooner it will reach a state where no more compression will be possible. Many technologies are used for making the circuits smaller in size. There is a new class of components called single-electron transistors and quantum dots. These require the interference of electron waves in order to cause disruptions. As they shrink in size, the single electron transistors work even better. It is expected that silicon-germanium may displace silicon for sometime. There is another technology, circuit-making via etching, uses light to etch the circuit pattern onto silicon chips. It promises to produce tinier circuits. But the drawbacks are that high problem error rates and the need to avoid even the smallest dust contamination. Silicon semiconductors can be etched with circuit lines about one micron in width and over the period of time it can be etched with even one-tenth to one micron in diameter. Interestingly, neurons are about 20 microns across and the axons that connect them are one-tenth to one micron in diameter. Scientists are looking for elements that are superior to that of silicon. Carbon, in the form of manufactured diamond is one option. It can tolerate and conduct heat very well and is also suited for making circuits smaller than neurons. Buckytubes is another form of carbon which may provide molecular-scale wires much thinner than axons. Carbon in the form of organic materials is another option. Gates and switches made of hinge-jointed protein molecules will be of the size of molecules. Light-sensitive solutions of synthetic DNA may be used to store huge databases.
The electron spin computer (ESC) can exploit quantum mechanics to change the electrons into on-off switches. Interestingly, as the ESCs shrink they work better. ESCs can be made of conductive metals and the elements will be approximately one-hundredth of a micron across and that will result in one trillion transistors packed into a single chip. The information densities of quantum computers will be much higher than those of brains.
The researchers at the Cornell University have created a machine that can replicate itself. The technique used in self-replication can be used to develop robots that could self-replicate or repair themselves in space or hazardous environments. The robots are made up of a set of modular cubes called molecubes. Each cube contains machinery and a program for replication. The cubes selectively attach to and detach from one another using electromagnets on their faces. Obviously a complete robot is nothing but a collection of cubes linked together. The replication process starts with the stack of cubes that bends over and sets its top cube on the table. Then the stack bends to one side or another and picks up a new cube and places it on the top of the first. This process repeats till a complete robot made up of a stack of cubes is created. This experiment has demonstrated that self-replication of machines is possible just like biological replication.
Consciousness means having an awareness of one’s environment and one’s own existence, sensations, and thoughts. Stan Franklin has developed software called IDA (Intelligent Distribution Agent) that interacts with Navy databases and communicates with the sailors using natural language email dialog. It implements a theory called GWT (Global Workspace Theory) in order to obtain functional consciousness. Pentti Haikonen proposes a cognitive architecture to reproduce the processes of perception, inner imagery, inner speech, pain, pleasure, emotions and the related cognitive functions. This architecture would produce higher-level functions by the power of the elementary processing units namely, the artificial neurons, without any programs. Haikonen believes that, when implemented with sufficient complexity, this architecture will develop consciousness.
Super-intelligent machines will be able to communicate with many people across the world. There are facilities for the super-intelligent machines to have a touch with the people: Internet is one of them. The ability of the super-intelligent machines to have intimate social relationships with billions of people will lead to understanding the thoughts of the people and so develop the consciousness. When machines are able to reason about themselves, introspect and interact in a meaningful manner with humans, they will meet the basic conditions for machine consciousness. When the biological consciousness is better understood, then it will be easy to achieve the equivalent machine consciousness.
Super-intelligent machines should love the human community. It has to understand the feeling of the people it has to communicate. So it should have emotions. It has to recognize happiness and sorrow in human facial expressions, human voices and human body language. It should have the innate emotional values that should be positively reinforced for expressing happiness and negatively reinforced for sorrow.
The mobility of the machines has found expression in the form of robots. The robots are in various shapes: snakes, spiders, humans and so on. Some robots, ASIMO for example, can walk like a human being, some robots, UFR-II for example, can fly like insects or birds and some robots, Souryu for example, can crawl like a snake. In fact, the shapes and the types of mobility they can perform depends on the factors like environment they are going to work, the problems they are required to solve, and so on. Still the robots are not completely autonomous. In order to become autonomous they need to be more intelligent. The super-intelligent machine should act on its own and it should not dependent on any controller.
Recent developments in machine vision have made us believe that the machines will get the human-level vision. In order to see the world around them, more advanced robots use stereo vision. The two cameras allow these robots to have depth perception, and image-recognition software gives them the ability to locate and identify various objects. Robots can also use microphones and smell sensors to analyze the environment.
So many researches are going one across the world to develop machines which can understand and generate the natural language. Even though no perfect system is created as yet, significant achievements in this field have been achieved.
Learning and Knowledge Acquisition
Learning is central to intelligence. It is essential for acquiring both skill and knowledge. Learning is a continuous process and so the modification of knowledge. Intelligence requires knowledge. Without knowledge the intelligence cannot be put into use. The machine can learn using the machine learning techniques and also by imitating the humans. A humanoid robot known as DB learns by imitating human actions. Knowledge can be acquired from sources such as Internet using the data mining techniques.
Creativity is the ability to derive novel ideas. Creativity requires knowledge and imagination. Creativity is a mysterious process. Freud believes that we are getting new ideas unconsciously. One way of generating new ideas is by combining or correlating the information stored in the knowledge base and also by modifying the information. Genetic algorithms and artificial neural networks also play a major role in developing machine creativity. In genetic algorithms, mathematically simulated ‘genes’ combine randomly and mutate to produce new potential offspring which in turn represent new concepts or ideas. But the drawback is that they fail to produce expected results when the dimensionality of the problem increases. Creativity machines that employ neural architecture can overcome this problem. They are self-organizing systems and work faster.