There is one day when researcher was watching a movie titled I-Robot. The film shows that the technology of artificial intelligence has grown to the point that the robots are considered as intelligent and become part of the human society. Being a fan of science fictions, the researcher compares this film to the reality of the current world. Currently, the researcher assumes that Artificial Intelligence has recently been applied in almost every aspects of our society. Its application can be seen from the products used every day, such as smart phones, washing machines, and even elevator. Furthermore, with the introduction of human-like robots, such as ASIMO on EXPO 2005 in Japan, it seems that the term “Intelligent Systems” is not far away.
However, depending on the perspective of the users, they might not think that the applications they using are intelligent, such as elevator. On the other hand, people might think that the applications they using are intelligent, such as robots, and some of them might even going so far as treated them as one of their kind. This leads to the question whether the research of the artificial intelligence has reach to the point that people can say that this is an intelligent system or not at all. To provide answers to this question, the researcher will first elaborate the definition of artificial intelligences and the vague term of “Intelligence” itself.
The term “Intelligence” is actually a vague term, as it exhibits many meanings depending on the breadth of concept (Pfeifer & Bongard, 2007). For example, in psychology, intelligence is define as the ability to carry on abstract thinking, while in biology, it is define as the ability to learn and adapt to the new situation or environment. Although the term “Intelligence” can exhibits a lot of definition depending on the perspective of the people, there are two particular characteristics of Intelligence, which are compliance and diversity (Pfeifer & Bongard, 2007). In other words, intelligence always involves action and reactions towards or from an event or events. For example, people tend to follow or comply with the given rules or situation and exploit these rules to produce diverse behavior. Another example can be seen from ASIMO robot, where it is able to recognize there is an obstacle in front of it, hence ASIMO will try to avoid the obstacle or simply walk over it. However, these two characteristics still does not explain how system are said to be “Intelligent”. To address this issue, the researcher will explain the term “Intelligent” in two views. Additionally, before covering these two views, consider the following illustration to give an overview of this chapter. Intelligence can be illustrates in the form of the Tiger and the Spear (Hunt, 2011). In this illustration, it is fair to state that the Tiger is stronger than the human with spears in terms of biology. However, the human with spears can also be considered stronger than the tiger in terms of ecological. In this case, the spear bearer is able to shape the environment to suit his need compared to the tiger. The same thing can be refer to intelligence, in which if intelligence refers to biological capacity of information processing, then it is fair to state that machine is more intelligent than human as it is able to process lots of data. However, if intelligence refers to capacity to solve problems, then it is fair to state that human is more intelligent than machine as machine has limited capability of solving problems.
Classical View of Intelligence
The first one is Classical views of intelligence, where intelligence is based on cognition, in which intelligence is gained through reasoning. In other words, intelligent is though as capable of processing and manipulating rules or condition (Stillings et al., 1998). For example, solving mathematical problem requires the problem solver to manipulate the equations. Another example to further illustrates classical views can be seen from Alan Turing Imitation games or Turing Test. Turing Test is used to determine whether a machine is considered as intelligence (Thagard, 2005). In this case, the test consist of two rooms with the interrogator is outside of the room. The machine and the human are placed in the two rooms separately. The test is conducted by the interrogator who will ask question to either the human or machine to determine which one of them is human, in which the human attempts to help the interrogator choose the correct answer. In this case, the machine is considered have intelligence if it can successfully convince the interrogator that it is a human. However, there are many researchers who critics this view as it does not explain why the machine do what they are doing, thereby classifying them as symbol manipulator (Searle, 1980). To further illustrate this issue consider the following Chinese Room example on the next sub-chapter.
Embodiment View of Intelligence
The Chinese Room is a though experiments where it involves a human which does not speak or know Chinese language at all. The human is placed in a room that contains the rules or instructions on how to processes Chinese characters. Subsequently, assumes that a Chinese human comes and gives questions into the room. The human then process the questions using the rules or instructions and return the answers to the Chinese human. In this case, the Chinese human is convinced that the human inside the room understand Chinese. However, it is fair to state that the human are merely reading the instructions without understanding what the questions means. This lack of understanding is an evident that if the human in the room are to be replace by machine, the machine is also does not understand what it is doing, thereby the machine are not considered to be intelligent (Searle, 1980). This example leads to the second view of the intelligent, which is Embodiment. Embodiment views of intelligence differ from classical views, where intelligence is based on whether the machine is able to exploits the condition or rules and able to interacts with the rules or conditions. For example, when a human wants to lift a cup, he/she first sees where the cup is and then lifts the cup. In other words, this view states that a machine is considered as intelligent if it is able to sense and exploits its environment as well as understands the environment for its purposes.
Artificial Intelligence & the Reality
Artificial Intelligence refers to the science of developing any systems or machines which have intelligent capabilities (Stillings et al., 1998). Hence, artificial intelligence application can be range from the simplest application such as basic search engine to the most complex application such as robots. In this case, artificial intelligence can be divided into two types, Weak AI and Strong AI. Weak AI states that machines are able to simulate mental abilities (Ruiter, 2006). In other words, weak AI machines can only simulate intelligence, even though they can appear to be intelligent. Although the application of artificial intelligence is numerous, however current technologies or systems have not yet reach to the point where it can be considered as intelligent. To further explain this statement, the researcher will review the comparison between the current Artificial Intelligence applications with the reality or the impact of these applications in different perspectives.
Reasoning and Thinking
Both reasoning and thinking are interrelated terms, where it is essential criteria in determining whether machine has intelligence or not. In simple terms, both of them can be thought as a complex information processing system, where the aim is to solve problems. A notable application of reasoning and thinking can be found in expert system. A general expert system is used to give reliable advice in a limited area of expertise (Bethune et al., 2007). This allows the users to gain decision faster as if they were consulting an expert. In reality, the impact of expect system can be seen from the earliest expert system, Eliza by Josef Weizenbaum in 1960s, where the experiment shows that most people are impressed with its capability of “Understanding” (Pfeifer & Bongard, 2007).
Based on the case above, the system does successfully manipulate or trick the people into believing that the system “Understand” them, where the machine is considered intelligent if according to the classical view of intelligent. However, when related to embodiment view, the system cannot be considered as intelligent. This can be seen from its algorithms, where the system employs pattern-matching algorithms which rearrange the input sentences and represent it as question (Pfeifer & Bongard, 2007).
Another indication can be seen on how the expert system infers the situation. Current expert systems normally employ logic-like inference rules, which able to give the same solutions for all abstract problems, hence expert system is 100% reliable (Bethune et al., 2007). However, in reality humans reason better on concrete problems compared to abstract problems as well as other similar problems which the humans are familiar with (Manktelow, 2005). In other words, unlike current machines which are limited to its domain or area, humans are able to perform on unfamiliar situations according to their abilities, albeit the solutions might be bias.
The Affection and Its Implication
Affection refers to emotion where it is also one of the criteria of intelligent. Affection or emotion is considered as responses by a subject to certain sorts of events where the subject concerns (Sousa, 2010). Hence, it influences decision making of the subject. Affection or emotion differs from thinking, where it can interfere with cognition and is thought as the antonym of reasoning (Lehrer, 2007). In this case, most current machine or technology are based on the application of reasoning, which can be seen from toys, such as AIBO, Puppy, Tamagotchi (digital pet) and humanoid robots, such as WF-4, Robovie and so on.
In reality, these robots were able to interact with humans and can even impress the humans to think that they were showing emotion, such as Kismet (Pfeifer & Bongard, 2007). Therefore, in relation to the classical view, these machines are considered to be intelligent. Albeit this leads to the question whether emotions are result of reflex-like behavioral rules compared to complex cognitive processes (Pfeifer & Bongard, 2007), nevertheless it is fair to state that current machines still do not exhibit intelligent. To further illustrate or explained this statement, consider it from the perspective of knowledge representation and the figure A below.
The aim of knowledge representations is to attempt to represent or illustrates the issues or problems as such, that it is easy for computers to infer or interpret the issues (Stillings et al., 1998). One of the most popular knowledge representations is mental models, where humans represent reality or issues in the form of ‘small-scale models’ to which the humans determine the validity of the ‘models’ (Johnson-Laird, 2001). In reality, it is easier to model concrete problems than abstract problems (Manktelow, 2005), which can be seen on the application of expert systems. Therefore, feelings or emotions did not fit into any preferred language of thought (Lehrer, 2007); hence it is almost impossible to represent emotion as such that the computer or machine can understand. Even humans also have difficulty in expressing their feelings, such as love when proposing to someone.
Probability and Possibility of Judgment
Judgment can refer to decision making, in which most decision is made based on probability. As almost everyday people are face with making decision for almost every issue, hence it has become one of the criteria for a machine to be intelligent. Current machines or robots have implemented a lot of decision making techniques, where it can be seen in the form of probability. This is based on the observation that almost every people judge probabilities (Manktelow, 2005). In this case, the most famous judgment probability is the Bayes’ rules. Bayes’ rules are designed to allow multiple information to be taken into account in calculating the probability of an event (Manktelow, 2005). An example of this is weather forecast system, where it is able to forecast weather accurately based on the information given, such as history of the weather, current state of the atmosphere and so on. Therefore, it is fair to state that weather forecast system exhibits some degree of intelligent, hence a system is considered intelligent if it is able to take account of all information given. However, if relate to the reality of how human make decision, this is not the case. To further illustrate this statement, consider the experiment in the following paragraph below.
Assume that two people are meeting on the road and decide whether to greet one another by employing a systematic analysis (Slovic et al., 2004). The experiments shows that in reality most people would not make decision based on multiple information or events, the same goes to Bayes’ rules. Even though having many information would mean that the accuracy of the probability would be more accurate, in real life most people would made decision based on single assessment (Manktelow, 2005). Additionally, in mental model, most people are more likely to construct one model for inference compared to multiple models (Johnson-Laird, 2001). Furthermore, in real life situation, most decision making are actually done based on feelings or emotions, such as determining whether an animal is dangerous or not (Slovic et al., 2004). Therefore, unlike machine which employs multiple information in order to infer a solution or conclusion, humans are able to infer a solution or conclusion based on single assessment. Furthermore, affection or emotion also influences the way human comprehend risk (Slovic et al., 2004). An example for this statement can be seen from playing chess where the cheese master determines whether a move “feels right”. In this case, the machine will have to calculate every possibility of each move before determining a move.