TOPIC 1.1.: DEFINITION OF ARTIFICIAL INTELLIGENCE To figure out what artificial intelligence is, this concept should be considered in a close connection with our understanding of what is intelligence in general. In this topic, you will find out what we understand by intelligence, how artificial intelligence is defined, and what is a fundamental approach to assessment of intelligence of a computer system. Nowadays, it is believed that humans, animals and even plants possess intelligence. Human intelligence is studied in-depth in many fields of science, such as philosophy, linguistics, neuroscience, etc. However, this is psychology that has offered a set of theories about the nature of human intelligence and has defined many characteristics of human intelligence. Moreover, this field of science has also developed a set of tests allowing assessment of the level of human intellectual development. At the same time, psychology has to some extent witnessed a crisis in the quest for what human intelligence is, because (Legg and Hutter, 2007): 1) there is still no a unified viewpoint if intelligence is a single ability or a set of abilities and this fact is manifested in drastically different definitions of intelligence, such as “Human intelligence is most parsimoniously conceived of as an emergent property of multiple specialized brain systems, each of which has its own capacity [...] that contribute to an individual’s ability to perform across a broad range of cognitive tasks” (Hampshire et al., 2012) “Intelligence is the aggregate or global capacity of the individual to act purposefully, to think rationally and to deal effectively with his environment. It is global because it characterizes the Dr.sc.ing., Dr.paed., assoc. professor Alla Anohina-Naumeca Department of Artificial Intelligence and Systems Engineering Faculty of Computer Science and Information Technology Riga Technical University E-mail: alla.anohina-naumeca@rtu.lv Address: Daugavgrivas street 2- 545, Riga, Latvia, LV-1048 Phone: (+371) 67089595 individual's behavior as a whole; it is an aggregate because it is composed of elements or abilities which, though not entirely independent, are qualitatively differentiable” (Wechsler, 1939) 2) it is difficult to measure intelligence due to the diversity of people, which is determined by, for example, linguistic, cultural, physical condition and other differences. As a result, the question arises what kind of human intelligence can be adopted as a standard in the development of intelligent computer systems: average intelligence for humanity, intelligence of some genius individual, or intelligence of a child with autism spectrum disorder? 3) many tests that are developed in psychology for assessing human intelligence are static and do not measure individual's ability to learn and adapt to his/her environment over time. Much less we know about animal intelligence. It is mostly related to learning ability of such animals as monkeys, dolphins, elephant, rat, or crows. However it is even more difficult to measure animal intelligence than human intelligence, because animal diversity is even greater taking into account their cognitive and perceptual capacities, available sensory organs and the ways how they explore the environment in which they live. Moreover, it is not possible to directly explain to the animal what its goal is. Additionally, studies on animal intelligence may be biased, expecting that animal intelligence is based on the same principles as human intelligence, although in reality it may not be so. (Legg and Hutter, 2007) Knowledge about plant intelligence is even more limited. It is currently known that plant intelligence is not based on a brain and nervous system. At the same time, plants possess a number of forms of intellectual performance, such as: Dr.sc.ing., Dr.paed., assoc. professor Alla Anohina-Naumeca Department of Artificial Intelligence and Systems Engineering Faculty of Computer Science and Information Technology Riga Technical University E-mail: alla.anohina-naumeca@rtu.lv Address: Daugavgrivas street 2- 545, Riga, Latvia, LV-1048 Phone: (+371) 67089595 • they learn based on repeated practice; • they have memory; • they are able to percept a number of aspects (shape, colour, smell and sound) at the same time; • they have a “sense of place” and an awareness of the neighbourhood. (Gagliano, 2015) “A plant continually gathers and updates diverse information about its environment, integrates this with information on its present internal state, and then makes decisions that reconcile its well-being with its environment” (Trewavas, 2016) Thus, in order to cover the diversity of the existent manifestations of intelligent performance, a general definition of intelligence should (Legg and Hutter, 2007): • encompass the essence of the different types of intelligence; • not be limited to any particular set of senses, environments, goals, and any specific kind of intelligence implementation structures; • be based on principles which are fundamental and thus unlikely to alter over time; • ideally be formally expressed, objective, and practically realisable as an effective intelligence test. Unfortunately, the general definition of intelligence does not exist because we, people, still have sufficiently limited knowledge of the world in which we live. As a result, if we cannot define what intelligence is, then we cannot also define what artificial intelligence is, and it also means that we do not have a standard that could be used when assessing if a computer system possesses or does not possess intelligence. However, historically, it has been established that artificial intelligence is considered in a strong connection with human intelligence. This is also justified by Dr.sc.ing., Dr.paed., assoc. professor Alla Anohina-Naumeca Department of Artificial Intelligence and Systems Engineering Faculty of Computer Science and Information Technology Riga Technical University E-mail: alla.anohina-naumeca@rtu.lv Address: Daugavgrivas street 2- 545, Riga, Latvia, LV-1048 Phone: (+371) 67089595 a goal defining why intelligent computer systems are developed. People in the real world perform or know how to perform different functions and activities. Intelligent computer systems are designed to support people in the completion of these functions and activities. Consequently, in this context, it would be logical to compare artificial intelligence with human intelligence (Fig. 1). Fig.1. Human intelligence as a standard for intelligent computer systems The first definition of artificial intelligence was offered by John McCarthy in 1956. At that time, this researcher organized a summer school with the participation of scientists from the United States who were interested in automata theory, neural networks, studies of intelligence and language simulation. The term “artificial intelligence” was chosen as neutral enough to avoid highlighting a particular topic of the summer school. The definition proposed by McCarthy defines artificial intelligence by emphasizing its similarity to human intelligence, and so it is the following: Dr.sc.ing., Dr.paed., assoc. professor Alla Anohina-Naumeca Department of Artificial Intelligence and Systems Engineering Faculty of Computer Science and Information Technology Riga Technical University E-mail: alla.anohina-naumeca@rtu.lv Address: Daugavgrivas street 2- 545, Riga, Latvia, LV-1048 Phone: (+371) 67089595 “The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it. An attempt will be made to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves... " (McCarthy et al., 1955) Meanwhile, Stuart Russell and Peter Norvig (2010) have grouped definitions of artificial intelligence into four categories. Human intelligence is highlighted in two of them. These categories are shown in Figure 2. Fig.2. Categories of definitions of artificial intelligence The previously mentioned categories of definitions focus on different aspects of intelligent performance (Russell and Norvig, 2010): categories at the top include definitions related to thought processes and reasoning; categories at the bottom encompass definitions related to behaviour (Fig. 3). Dr.sc.ing., Dr.paed., assoc. professor Alla Anohina-Naumeca Department of Artificial Intelligence and Systems Engineering Faculty of Computer Science and Information Technology Riga Technical University E-mail: alla.anohina-naumeca@rtu.lv Address: Daugavgrivas street 2- 545, Riga, Latvia, LV-1048 Phone: (+371) 67089595 Fig 3. Aspects of intelligent performance highlighted in the categories of definitions of artificial intelligence Moreover, the categories use different standards of intelligent performance (Russell and Norvig, 2010): categories on the left include definitions that adopt human intelligence as a standard; categories on the right encompass definitions that adopt an ideal intelligence called rationality as a standard (Fig.4). Fig.4. Standarts of intelligence adopted in the categories of definitions of artificial intelligence Each category of definitions of artificial intelligence is based on a specific approach used to develop intelligent computer systems (Russell and Norvig, 2010): Dr.sc.ing., Dr.paed., assoc. professor Alla Anohina-Naumeca Department of Artificial Intelligence and Systems Engineering Faculty of Computer Science and Information Technology Riga Technical University E-mail: alla.anohina-naumeca@rtu.lv Address: Daugavgrivas street 2- 545, Riga, Latvia, LV-1048 Phone: (+371) 67089595 • A cognitive modelling approach underlies systems that think humanly. It considers human being to be an information processing system. Therefore, research works focus on how information is represented, processed, transformed, and stored in the human mind and in computers. • Systems that think rationally are based on logics. As a result, knowledge is represented as logical rules, for example, if we know that “All people have a head” and “Alex is one of people”, than we can conclude (to infer new knowledge) that “Alex has a head”. • The approach of the Turing test underlies systems that act humanly. It assumes the development of such intelligent computer systems that are able to reach human-level performance in all cognitive tasks. • Systems acting rationally are based on the approach of rational agents. In this context, an agent is any entity which is capable to perceive its environment and to act within it. A rational agent is one that acts to achieve the best outcome or, when there is uncertainty, the best expected outcome. To develop such agents, it is necessary to use approaches mentioned in all previous categories of definitions of artificial intelligence. The Turing test underlies systems that act humanly and it is also one of the fundamental tests in the field of artificial intelligence intended to assess if a computer system possesses intelligence. It has the following idea. There are 3 rooms. In the first one there is an artificial intelligence, in the second room - a human being or natural intelligence, and in the third room there is a tester (Fig.5). The tester does not see nor artificial intelligence, nor human being and asks questions to both intelligences using a text-based input tool. If the tester cannot differentiate which answers were given by the human being and which ones by the artificial intelligence, than the artificial intelligence is intelligent. Dr.sc.ing., Dr.paed., assoc. professor Alla Anohina-Naumeca Department of Artificial Intelligence and Systems Engineering Faculty of Computer Science and Information Technology Riga Technical University E-mail: alla.anohina-naumeca@rtu.lv Address: Daugavgrivas street 2- 545, Riga, Latvia, LV-1048 Phone: (+371) 67089595 Fig.5. The Turing test To pass the Turing test, the computer system must have the following capabilities (Russell and Norvig, 2010): • natural language processing for maintaining successful communication with the tester as the human being does; • knowledge representation, in order to store information acquired before and during the test. The human being uses for this purpose long-term memory to store all previously acquired knowledge and short-term memory - to store knowledge processed at a specific time; • automated reasoning, in order to use the stored information for answering questions and drawing conclusions; • machine learning, in order to be able to adapt to a new situation, as well as to acquire and generalize patterns of different situations (for example, to a new topic or intent of the tester, etc.). The regular Turing test excludes direct physical interaction between the computer and the tester, because it is not necessary to simulate a human being to make conclusions about intelligence. However, the so called the Total Turing test brings forward two more requirements: Dr.sc.ing., Dr.paed., assoc. professor Alla Anohina-Naumeca Department of Artificial Intelligence and Systems Engineering Faculty of Computer Science and Information Technology Riga Technical University E-mail: alla.anohina-naumeca@rtu.lv Address: Daugavgrivas street 2- 545, Riga, Latvia, LV-1048 Phone: (+371) 67089595 • the computer must have computer vision to perceive objects, and • robotics to manipulate objects and move about. (Russell and Norvig, 2010) So mastering this topic you found out that there is a richness of manifestations of intelligence performance in our world, and, as a result, it is still difficult to define what intelligence is. On the one hand, these difficulties also prevent us to define precisely what artificial intelligence is. On the other hand, the diversity of interpretations of the concept of artificial intelligence permits a variety of intelligent computer systems and research works to be developed. This enables the industry of artificial intelligence to develop and offer a variety of solutions – gaming software, self-driving cars, industrial robots and many others. Information sources Gagliano M. (2015). In a Green Frame of Mind: Perspectives on the Behavioural Ecology and Cognitive Nature of Plants. AoB Plants, (7). DOI 10.1093/aobpla/plu075 Hampshire A., Highfield R.R., Parkin B.L., Owen A.M. (2012). Fractionating Human Intelligence. Neuron, 76(6), pp. 1225-1237. Legg S., Hutter M. (2007). Universal Intelligence: A Definition of Machine Intelligence. Minds & Machines, 17(4), pp. 391-444. McCarthy J., Minsky M. L., Rochester N., Shannon C.E. (1955). A Proposal For The Dartmouth Summer Research Project On Artificial Intelligence. Russell S., Norvig P. (2010). Artificial Intelligence: A Modern Approach. Pearson. Trewavas T. (2016). Plant Intelligence: An Overview. BioScience, 66(7), pp. 542-551. DOI 10.1093/biosci/biw048 Wechsler D. (1939). The measurement of adult intelligence. Baltimore: Williams & Wilkins. Dr.sc.ing., Dr.paed., assoc. professor Alla Anohina-Naumeca Department of Artificial Intelligence and Systems Engineering Faculty of Computer Science and Information Technology Riga Technical University E-mail: alla.anohina-naumeca@rtu.lv Address: Daugavgrivas street 2- 545, Riga, Latvia, LV-1048 Phone: (+371) 67089595