West Virginia university What is Artificial Intelligence? A comprehensive introduction of the meaning of machine learning Daniel Thomson Daniel Thomson Page |2 Abstract This document serves to outline and discuss the meaning of artificial intelligence and its applications to modern day technologies. We will also delve into what the actual meaning of intelligence is. This document will discuss various aspects of machine intelligence, problem solving strategies, machine knowledge, reasoning, planning, and learning used by today's machines. We will explore examples of what we consider actual machine intelligence as well as determine what categories and application domains these machines fall into. Introduction To understand what artificial intelligence is, first we must ask what is intelligence? When one thinks of intelligence one tends to relate to how smart someone is, or what their IQ score is. Contrary to that, intelligence can be measured in many different ways. Alfred Binet, a French psychologist in the late 19th century, defined intelligence as judgment, otherwise known as "good sense", the faculty of adapting one's self to circumstances.[1] However, is intelligence only related to ones intuition? Cyril Burt a British psychologist claims that intelligence is just innate general cognitive ability.[1] So anything that seems to display an act of knowledge, attention and working memory. David Wechsler an American psychologist says that Intelligence is the aggregate or global capacity of the individual to act purposefully, to think rationally, and to deal effectively with his environment.[1] This comes closer to what one would generally think intelligence actually represents in humans. Psychological Intelligence Generally speaking, today's psychologists group intelligence into many different categories. This makes intelligence a compartmental trait meaning that an entity can display intelligence in some areas and none in others. These areas include, but are not limited to: logic, thought, understanding, communication, learning, memory and problem solving.[1] As an example, someone can display excellent problem solving skills yet lack the ability to communicate intelligently. I myself believe that I have a poor memory, I can understand and learn core concepts in a course, but when it comes to execution on a test, I Daniel Thomson Page |3 seem to just draw a blank. So unless I study a lot to retain this information, I will most likely either forget or just lose grasp of a concept. It has been said that based on this definition of intelligence, the definition of competence is very similar if not the same. Competence can be defined as a sufficiency of means for the necessities and conveniences of life. It can also be defined as having the basic knowledge to communicate and understand a language.[1] Can we say this is the same as intelligence? Well, we did define a component of intelligence as communication, so there is one similarity. However, we can't really say they are the same thing. Intelligence is so much more extensive and covers so many more specific areas of behavior rather than just being alive. We as humans can recognize a form of intelligence when we see one. For example, let's say you teach a dog a new trick and it remembers it. That right there is an example of intelligence where the dog learned and retained information. Animals are often used as a benchmark to observe intelligence. Japanese scientists, for example, used an octopus to see if it could escape from a jar and even open one from the outside to obtain food. They observed that the octopus was able to understand the logic behind the screw on lid of the jar and ultimately was able to escape and enter.[7] Another more well known example is Koko the gorilla. Koko was taught to use American Sign Language and knows signs for over two thousand English words.[8] In the story of the octopus and Koko the gorilla, we can look back to the previous question whether it is possible to be intelligent in some things and not so intelligent in others. As for the octopus, it seems to be very intelligent in problem solving skills; However it is highly doubtful that an octopus can learn American sign language. Whereas for Koko the gorilla, she is very intelligent in learning, memory and communication. Both of these behaviors are indications of intelligence in problem solving and communication as well as learning and retaining information. There is a lot of speculation when some claim that intelligence is related to speed of response. For example, if one person can solve a rubics cube faster than another, but both can get the same solution, is one more intelligent than the other? This is a difficult question because so many factors come into play here. What if one person had been practicing for Daniel Thomson Page |4 years and the other had just seen a rubics cube for the first time, but, the person who had been practicing for years just finished faster by a short time? Would that make this persons intelligence seem less impressive? If both come up with the same solution, I believe the problem solving skills are more important in measuring intelligence rather than by how fast they are produced. Machine Intelligence So how do we relate all of these factors of human intelligence to machine intelligence, and more importantly define machine intelligence? Using these definitions we can define what artificial intelligence actually needs to be. Based on what we have seen so far in human intelligence, we can say that machine intelligence is a machine that can learn and retain information as well as use the information and knowledge to perceive its environment and takes actions to maximize its chances of success.[1][3] When you think of artificial intelligence, the first thing that usually pops into your head is robotics. While these traits are fine for a robot, some, such as communication seem left out; However, this is where that idea of compartmental intelligence comes back into play. It's more important that a machine displays intelligence in its primary application domain. Because of this, many instances of what people consider artificial intelligence are actually not, and some consider a machine is not intelligent, actually is according to our definition. The Turing Test Alan Turing, a British computer scientist known as the father of artificial intelligence, wrote a paper in 1950 called the "Imitation game".[2] This paper raised the issue of whether machines can ever be made to think and proposed that we judge intelligence the way we did with humans, by observation. He proposed a test we call today the Turing test. It is a test given to a machine to see if it has the ability to show signs of intelligent behavior similar to a human. The test is done by having an human interrogator have two text-based conversations, one with a human and one with a computer. The interrogator could ask questions, and the respondent is to answer the questions. If the respondent was a computer but judged to be human, this was taken as an indication that the machine was acting human-like or "thinking" and the machine is said to have passed Daniel Thomson Page |5 the test.[1][2] The basic abilities that the machine must have to pass the test are language understanding and communication. The first machine to have actually passed the test is known as Eliza. This machine was first developed around 1966 and was designed to simulate the psychotherapy behavior of a therapist or counselor.[3] People would interact with Eliza via a keyboardterminal, and the way the program would work is by asking questions about the user and their problems. Once the user gave an answer Eliza would then pick key points in the response and re-word the users question into response to dig deeper into the users issue. This created some interesting behaviors on the human user side of the interaction. It was found that Eliza was actually helping users talk through some issues just by almost rewording and constantly asking questions based on the users input. Is this actually artificial intelligence or is it just clever programming? Looking at it from a programming standpoint, it wouldn't be very difficult to program something like this using some algorithm to determine what to reply to the user. Looking at it from our definition of intelligence, this computer had natural language understanding as well as an ability to formulate a response based on reasoning. I believe that Eliza has some form of artificial intelligence, and for 1966 it was very impressive. Many of the arguments against Alan Turing in his article "Imitation Game" are about how machines shouldn't be able to think or cant because of God, instead of actually proving how they cannot think from a computer science point of view.[2] One argument says that unless the machine is able to produce emotions then it is not intelligent.[2] Another argument states that because the consequences of machines thinking would be too dreadful let's hope they never can. [2] I believe that machines are capable of thinking, this article was written in 1950 and makes some excellent points that today have already been proven because of technological advancements. Given that the article was written in this time period, this explains some of the arguments against Turing since computers weren't a part of everyday life during the fifties. Given some of the amazing inventions that have been created recently such as the self driving car from Tesla and Google, it is more accepted in the 21st century that artificial intelligence must exist. Daniel Thomson Page |6 Components of Machine Intelligence The four major components of artificial intelligence are knowledge, reasoning, language understanding and learning.[1][3] Knowledge; does the machine must have a vast database of information on its specific application domain? Reasoning; should the machine "think" about what it has to do or just do it? Language understanding; can the machine communicate and understand the human interacting with it? Learning; can the machine learn from given input? Intelligence in machines can be specific to a single domain and still be considered intelligent. The KUKA robot is a table tennis robotic arm that was tested against the world champion table tennis player.[5] The robot is intelligent in the fact that it can play table tennis against another human and learn from the opponents style. This machine doesn't need a general intelligence to be considered as artificially intelligent. One of the most accessible and common forms of artificial intelligence in robotics is the Roomba, a robotic vacuum that cleans a floor and stops when the goal is complete.[6] The Roomba uses a pretty specific set of instructions and simply repeats them. First it circles until it hits something then it goes to free roam mode, if it hits something it simply turns and carries on until its dirt bin is full.[6] The machine here doesn't learn, adapt to its environment, doesn't understand, it just follows instructions, it's a very static machine. However, many consider it artificial intelligence, why? The machine does adapt to its environment, for example, if it hits something the sensor will tell the machine to turn until it is no longer hitting something. As well as stopping when the dirt bin is full it will stop cleaning. While many may argue against this, is does display intelligence in its application domain, it will have to determine what to do given the input from its sensors. While this is artificial intelligence at a quite low level, it is still intelligent given our definition. A long with the general machine intelligence components, there is emotional intelligence. This is the ability to tell the difference between emotions of people and name them. Using this information we can guide thinking and behavior. A machine can have emotional intelligence, this is used a lot in facial recognition systems where based on body language and the current status of your facial muscles. For a simple example if you're Daniel Thomson Page |7 frowning the emotion to relay is sad and if you're smiling the emotion to relay is happy, obviously it can get much more complicated than that for example is someone is crying tears of joy. Based on body language, tone of voice and actions computers could theoretically tell the difference. Boston Dynamics created a dynamically stable quadruped robot in 2005. This robot was named BigDog (since it looks and walks like a dog) and is one of the most famous machines to display "thinking" before an action happens. This machine has the ability to "catch" itself if it begins to fall when walking from asphalt to and icy surface. BigDog is also able to react to getting pushed or kicked where it determines what it must do to stabilize itself.[9] Some intelligent machines don't need to adapt to situations as their domain is so specific it isn't required, an example of this are voice recognition systems such as Siri and Cortana do not require this trait. Google and Tesla Motors have started to push Artificial Intelligence to its limits by developing self driving cars. These vehicles would use sensors to determine unpredictable objects in its test environment such as pedestrians crossing the road as well as red lights among many other obstructions. These cars would also be able to use global positioning systems to determine the speed limit as well as traffic conditions and changes in speed due to work zones.[10] This is the very top level of Artificial intelligence as of today. These vehicles display all four categories of machine intelligence. Knowledge; the car keeps an extensive database on roads and speed limits. Reasoning; the car can re-route based on conditions and determine what to do given certain situations. Language understanding; given voice input from the user the car would be able to route and drive to the given location. Learning; the car can learn from its environment and adapt to events such as red lights and other cars. Since there are four states in the United States that allow self driving cars to be driven on the road, Google has logged over 700,000 miles on an autonomous vehicle with only two accidents on record, both of which caused by human interaction.[11] Conclusion Artificial Intelligence is a field of study that is constantly changing due to advances in technology. It ties computer science and psychology closely together to define what Daniel Thomson Page |8 constitutes as a intelligent agent and whether or not that is possible. Living examples such as Boston Dynamics' BigDog, KUKAs robotic arm and iRobots' Roomba series created in the early 2000's have paved the way for the future of Artificial Intelligence and have really unlocked the true potential of robotics and machine intelligence. Today we have corporations such as Google and Tesla Motors creating the future of travel with self driving cars. We have shown today in the 21st century that artificial intelligence is truly possible and that machines can display acts of intelligence through knowledge, reasoning, language understanding and learning. Daniel Thomson Page |9 References [1] Russell, S., & Norvig, P. (2010). Artificial intelligence: A modern approach (3rd ed.). Upper Saddle River, NJ: Prentice Hall. [2] Turing, A. (n.d.). I.—Computing Machinery And Intelligence. Mind, 433-460. [3] McCorduck, P. (2004). Machines who think: A personal inquiry into the history and prospects of artificial intelligence (25th anniversary update. ed.). Natick, Mass.: A.K. Peters. [4] Hsu, F. (2002). Behind deep blue: Building the computer that defeated the world chess champion. Princeton: Princeton University Press. [5] KUKA KR AGILUS VS. Timo Boll. (n.d.). Retrieved April 1, 2015, from http://www.kukatimoboll.com/ [6] Sandler, A. (n.d.). How to Program Roomba - Introduction. Retrieved April 1, 2015, from http://www.robotappstore.com/Knowledge-Base/1-Introduction-to-RoombaProgramming/15.html [7] Winter, L. (2014, May 9). Octopus Vs. Jar. Retrieved April 1, 2015, from http://www.iflscience.com/plants-and-animals/octopus-vs-jar [8] Luce, J., & Miles, H. (1983). Apes and Language: The Search for Communicative Competence. In Language in Primates Perspectives and Implications (pp. 43-61). New York, NY: Springer New York. [9] BigDog - The Most Advanced Rough-Terrain Robot on Earth. (n.d.). Retrieved April 1, 2015, from http://www.bostondynamics.com/robot_bigdog.html [10] Inside Google's Quest To Popularize Self-Driving Cars. (n.d.). Retrieved April 2, 2015, from http://www.popsci.com/cars/article/2013-09/google-self-driving-car [11] Google’s self-driving car passes 700,000 accident-free miles, can now avoid cyclists, stop at railroad crossings | ExtremeTech. (n.d.). Retrieved April 2, 2015, from http://www.extremetech.com/extreme/181508-googles-self-driving-car-passes-700000accident-free-miles-can-now-avoid-cyclists-stop-for-trains