article770.asp.htm

电脑游戏中的人工智能制作的文章收集。 搞游戏设计和编程的人都可以参考一下

<p><b>IA compared to intuition. </b>Consider the steps followed by IA. It stores details of all diseases, their characteristics and the relationships between them in memory. It receives inputs concerning symptoms through &quot;Yes/No&quot; answers. It simulates parallel processing to globally evaluate the current input. It is encoded negatively to use all inputs to eliminate unrelated diseases. If an input indicates any unique symptom, it achieves instant recognition by eliminating all except the related disease. It follows an algorithm which results in instant identification. Compare IA to the recognition process of the mind. When a face is familiar, there is instant recognition. Let us call it intuition. Such recognition, of thousands of such objects, is repeated by people world-wide millions of times every day. Like most other events in nature, such a process must follow an orderly set of instructions to achieve results in a finite number of steps. In essence, intuition must also follow an algorithm.
<p><b>Memory and relationships. </b>A comparison of IA with the current knowledge of the mind, reveals some similarities and several unexplained enigmas. This essay attempts to fill in the gaps to create a composite view of the mind. Firstly IA stores the names of all diseases in memory. It is logical to assume that the mind stores data on all known faces in memory. But the mechanics of memory remains unknown. This essay suggests a sound possibility. Secondly, IA infers that certain symptoms are present, or absent, based on simple &quot;Yes/No&quot; answers to queries. There is considerable evidence that the mind isolates thousands of characteristics of any seen object. Obviously, the mind must perceive the characteristics of faces to be present, or absent. Thirdly, IA stores the relationships between symptoms and diseases. In recognising a face, the mind establishes its identity. Identification demands a link between a face and its known characteristics. One must know that the face is oval, or round. It is reasonable to presume that the mind must have such links. But how the mind stores such links remains a mystery. This essay suggests how nerve cells can establish and store such relationships.
<p><b>Nerve cells eliminate alternative possibilities. </b>Fourthly, IA encodes a negative relationship between diseases and their symptoms. It is deliberately coded to eliminate. Deliberate elimination of alternatives is a well documented feature of the nervous system. (3) Nerve cells have a powerful system of parallel inhibition of surrounding neurons when a particular group of neurons start to send information. This inhibition is strongest for those immediately adjacent to the excited neurons. Throughout the nervous system there are neural circuits which switch off other circuits when their own areas are energised. There is evidence that the mind carries such systematic elimination beyond logic. This is illustrated in the popular vision experiment, where a drawing can be interpreted as a vase, or two faces facing each other. The mind eliminates one interpretation to recognise the other - a vase, or two faces. Evidently each recognition path acts powerfully to inhibit the other. Recognition is firmed up by eliminating even logical alternative solutions.
<p><b>The coding of elimination by nerve cells. </b>The mind is known to have specialised networks which perform unique functions. There is a network to identify the edges of a seen object. Another to detect the beginning and end of movements by muscles. This essay gives some examples of how such intelligence can be achieved through recognition based on the memory codes of neurons. In fact, the key theme of this essay is that such recognition can give intelligence to a network. Such a tool can give neural networks the capability of achieving a variety of intelligent tasks. It is assumed that neurons may be suitably coded, to facilitate elimination of less viable alternatives. This essay does not suggest any probable process the mind may use to determine such elimination. But, elimination, as a neural process, remains a well documented and practically experienced event.
<p><b>Parallel links for speed. </b>Definitive research suggests that the brain simultaneously isolates every incoming sensory image into myriad characteristics. (4) The visual image alone is divided into several hundred million separate characteristics of light, shade, colour, outline and movement. We do not know how all this information gets organised and processed. But, each nerve cell in the system is known to have a hundred to a quarter of a million links with other cells. (5) The average nerve cell is known to respond within about 5 milliseconds of receiving a message. Since all cells work in parallel, any message received by any cell can reach any other cell in the system within just five or six steps - in just one fiftieth of a second. Currently, science does not know how such a process can rapidly transfer information in the system. Recognition may be provide the pivotal link. It can link every cell to the system. If so, every cell in the network can recognise and respond to every flash of incoming information. If we assume a recognition role for the nerve cell, global interpretation of incoming information and instant response becomes feasible for the system.
<p><b>IA imitates intuition. </b>IA has classic simplicity and power in its logic. The elimination process is logical. It is discrete and does not leave a fuzzy answer. Yet it has the ability to evaluate possibilities with vague qualities. If a face is known to occasionally wear spectacles, all faces which never wear spectacles can be eliminated. A vague characteristic is productive for IA. As opposed to this, a search and match algorithm finds the &quot;occasional use&quot; type of information futile. IA logic is holistic, since it evaluates its entire database, with each input. Every answer updates its perspective, by eliminating all elements that fail the search criteria. Every answer narrows its focus. It creates in IA the equivalent of &quot;global awareness&quot; of the mind. As against this, a search and match algorithm ambles about in the vast search space without a clue as to the global picture and appears stupid. Finally, IA instantly identifies a pattern, if it indicates even a single unique quality, through simultaneous elimination. In conclusion, IA is logical. It imitates intuition in being holistic, avoiding &quot;stupid questions&quot;, handling uncertainty and in providing instant recognition.
<A NAME="4"><p align="center"><H1>The Nerve Cell and Recognition</H1>
<p><b>A nerve cell has many inputs and a single output. </b>A cell is the basic unit of all living tissue. In the human body, there are specialised cells called neurons, which transfer information rapidly from one part of the body to another through electrical nerve impulses. Each of the one hundred billion or so nerve cells has many inputs and a single output. (6) A typical neuron has thousands of minute threadlike growths called &quot;dendrites&quot; which conduct impulses towards the cell body. A central &quot;cable&quot; called an &quot;axon&quot;, conducts impulses away from the cell body. The output of every cell in the entire nervous system is an &quot;all, or nothing&quot; impulse, called an action potential, despatched through its axon. A neuron receives many inputs and dispatches a single output.
<p><b>Neuron believed to be a computational device. </b>Current research views this output of the cell as a computational message. (7) The voltage of a neuron at any given moment, is presumed to reflect all the summation activities of a thousand inputs. As the inputs arrive, they are supposed to be rapidly added to or subtracted from the total neuron voltage. It is presumed that if the stimulus is strong enough to breach a critical threshold level, an action potential is fired. Other neural network theories assume complex calculations, giving weightages across neurons. Current scientific theory assumes that nerve cells use some form of computation, meaning mathematical, especially numeric methods.
<p>"><b>Nerve cells may not compute. They may recognise. </b>IA points to intuition as a process, which acts through elimination based on simultaneous recognition of millions of separate characteristics. It has been reasoned that, at the seminal level, recognition may be accomplished by a nerve cell. There are many supporting arguments for this thesis. &quot;Recognise&quot; means &quot;to establish an identity&quot;. Mathematical computational ability does not focus on the identity of a node. Weightages may give greater identity, but fail to give a node a singular quality, which can be recognised by millions of other nodes. Yet, there is experimental evidence that a single nerve cell may inhibit the actions of millions of other cells. If addition or subtraction is the principle, it is hard to justify the idea that the firing of a single nerve cell among thousands of others can add up to trigger an action potential in an axon. You cannot add &quot;1&quot; to &quot;-1000&quot; and get &quot;+1&quot;. If recognition is the key, even a single microscopically small input from a single cell can trigger recognition and inhibition of a whole battery of cells.
<p><b>The nerve cell may operate a form of Boolean Logic. </b>Each nerve cell may be functionally competent to recognise a single event. It may fire a volley of impulses when the event is recognised. The all or nothing response of the nerve cell may be a form of Boolean logic. In Boolean algebra, all objects are divided into separate classes, each with a given property. Each class may be described in terms of the presence or absence of the same property. An electrical circuit, for example, is either on or off. Boolean algebra has been applied in the design of binary computer circuits and telephone switching equipment. These devices make use of Boole's two-valued (presence or absence of a property) system. Firing by each neuron may represent the presence, or absence of a distinct property. The entire nervous system may recognise an input from a cell as a perception of the presence of a property. Alternatively, the system may recognise firing by a cell and respond with a specific activity, such as a muscle movement.
<p><b>Recognition at the input level. </b>For sensory inputs, the firing of a nerve cell is known to indicate recognition. The entire in formation input into the human nervous system is through cells called receptors which convert sensory information into nerve impulses. (8) Chemoreceptors in the nose and tongue report on molecules which provide information on taste and smell. Other receptors are massed together to form sense organs such as the eye and the ear. There are receptors which report on pressure, touch, pulling and stretching. Nociceptors report on cutaneous pain. Peripheral nerves connect these sensory receptors to the central nervous system. At the entire input level, nerve impulses indicate recognition of the occurrence of millions of isolated events. The whole system recognises the firing by each one of these cells as the perception of a single microscopic event. At the input level, the firing of a cell indicates an act of recognition and not one of computation.
<p><b>Motor events at the output level. </b>At the output level, individual nerve impulses control motor outputs. There are motor areas in the cortex, the wrinkled surface layer of the cerebral hemispheres of the human brain. (9) Careful electrical stimulation of these areas send nerve impulses which invoke flexion or extension at a single finger joint, twitching at the corners of the mouth, elevation of the palate, protrusion of the tongue and even involuntary cries or exclamations. The nerve fibres carrying inputs to and outputs from the cortex pass through the thalamus, a major neural junction in the brain. This junction plays a key role in this explanation of the activities of the mind. The nerve impulses passing through follow a form of Boolean logic. They report the presence or absence of individual events, or activate or are quiescent to isolated motor functions. Each action potential indicates, at the input and output levels, the perception or the triggering of a property - a distinctive event.
<p><b>Nerve cells cannot add apples to pears. </b>At the input and output levels, the firing of a nerve cell indicates an event. Current theory admits the Boolean function at these levels. But scientists imagine computation by nerve cells at subsequent levels, where these messages are interpreted and transmitted further. While it has a single &quot;all or nothing&quot; output, a typical neuron receives thousands of inputs from other nerve cells. Numeric computation (adding, subtracting, dividing, or multiplying) of widely varying inputs is quite improbable. The inputs are distinctly different events such as sound, light, pressure, or smell. The outputs are complex muscle movements. It is wildly chaotic to include all this into an integrated computation. It is like adding apples to pears, or subtracting the sense of touch from the sense of pain. It is more realistic to assume that a pain cell recognises touch and reacts by despatching or inhibiting a pain message. Recognition can evaluate varied inputs and trigger an appropriate output. Recognition may provide the key to understanding intelligence.
<p><b>Recognition the first step to intelligence. </b>Throughout the nervous system there are networks of cells, which appear to act intelligently. These events have been assumed to be some form of network intelligence - a mysterious mental capability. But such intelligence can be explained if we assume that nerve cells recognise incoming information and respond with action potentials through their axons. A typical unexplained act of intelligence is the baffling capability of the mind to modify the sensation of pain on its route to the cortex. The sensation of pain is known to be reported, enhanced or suppressed, under varying conditions. Consider the following explanation. A neuron which reports cutaneous pain may receive inputs from its primary pain sensory neuron (P), along with other dendritic inputs from neighbouring (sympathetic) pain (SP) and touch sensory (T) cells. The cell may report pain and sympathetic pain. It may ignore the sense of touch to report pain. It may also inhibit sympathetic pain giving priority to the sense of touch. In such a context, the cell responses to the listed inputs may be as follows:
<p>P - Fire. Reports pain.
<p>SP - Fire. Reports sympathetic pain.
<p>P+T - Fire. Ignores touch and reports pain.
<p>SP+T - Inhibit. Suppresses sympathetic pain to highlight touch.
<p>In reporting, or suppressing sympathetic pain, the cell may be selectively responding to combinations of nerve impulses received at different dendritic inputs. It may be recognising unique combinations to trigger its own interpretation of a single event.
<p><b>An executive attention centre. </b>The recognition model can also illuminate the puzzling process of paying attention. (10) William James, in one of the best writings on the mind, suggested that attention is &quot;the taking possession by the mind, in clear and vivid form, of one out of what seem several simultaneous objects or trains of thought. Focalisation, concentration of consciousness are its essence&quot;. The focus of attention is believed to be the key in trying to understand the concept of consciousness. Research has revealed some facts concerning attention. (11) PET scans create images of brain activity by detecting the presence of glucose in blood flow to nerve cells in the brain. When particular cells are more active, there is more glucose in the local blood flow. The scans detect increased presence of glucose to construct a three dimensional model of the brain on a computer screen showing greater activity with brighter colours. Recent research using PET scans have revealed activity in an executive attention centre (EAC) in the cortex, when people focus attention. This area of the cortex lights up when a person pays attention to a sensory input. Mystery remains as to how activity in this region can enable the system to pay attention.