Introduction to Artificial Intelligence

By: Koné Mamadou Tadiou, PhD

A.      What is Artificial Intelligence ?

John Mc Carthy, an MIT and Stanford Professor said in 1955 that Artificial Intelligence (AI) is “the science and engineering of making intelligent machines.” AI is indeed about the study and research leading to the simulation of human intelligence. Its main goal is to precisely describe this human intelligence and made it available to machines for processing and creating intelligent agents. The widely known pioneers of AI are John Mc Carthy, Marvin Minsky (MIT), Herbet. A. Simon (Carnegie Mellon U.), Allen Newell (Carnegie Mellon U.) and Stuart Russell (U.C. Berkeley) who first established with much optimism, its foundations in 1956 at a conference at Dartmouth College.

Although a multitude of subfields has appeared over the decades, AI most well known components are Knowledge Representation (KR), Machine Learning (ML), Natural Language (NL) processing, Artificial Neural Networks (ANN), Decision Theory (DT), Genetic Algorithms (GA), Artificial Vision (AV), and Robotics.

B.      History of Artificial Intelligence

The current state of research in AI has much to do with its history and evolution over several decades. From its foundations laid down at the Dartmouth conference in 1956 by its pioneers, the field of AI has gone through a series of rise, fall (AI winters) and renaissance triggered by research results on the one hand and the expectations of its funders on the other hand. The AI adventure began with Alan Turing, a British mathematician and computer scientist, who proposed his famous Turing Test in his landmark paper “Computing Machinery and Intelligence” [1] in 1950. In this contribution, Turing defines an intelligent machine as one that is capable of carrying a conversation with a human through a computer program. In addition, the human subject should not be able to tell whether his/her interlocutor is a human being or a machine.

Then, over almost two decades, the discoveries and achievements of AI enjoyed much admiration [2]  for two main reasons. First, through a clever use of heuristic-based algorithms (“rule of thumbs”), it tackled concrete problems in game playing and theorem solving. Second, research in AI with the aim of giving computers the ability to communicate in natural language (ex. English, French) helped create astonishing programs like Student, a program to solve elementary algebra, and Joseph Weizenbaum’s Eliza [3], the first program to simulate a text-based intelligent human conversation. Nowdays, machines are expected to solve problems and not behave and think like humans.

C.      Fundamentals of Artificial Intelligence

In theory and practice, artificial Intelligence [4], [6] is concerned with reducing complex and intractable problems to manageable ones through the following approaches:

  • Problem solving by search;
  • Knowledge representation;
  • Machine learning;
  • Natural language processing;
  • Artificial neural networks.

C. 1. Aspects of search

An example of a search problem that may seem simple and naive to a human being is “Planning in the bloc world” depicted in Fig.1 below:

Fig. 1: Planning in the bloc world

In AI, a machine needs to solve this problem by means of search which is the investigation of several legal (allowable) sequences (ex. “move bloc C first, move bloc B on top of bloc C, move bloc A on top of others”) in a state space that leads effectively to the goal. Here, the problem is to move from the sate “original situation” to the state “target situation”. In addition, if reaching the goal must be done efficiently (minimal time, resources) then we talked about an optimization problem. A tree with its root (initial state), nodes (states), branches (one or several legal moves) and leaves (terminal state) is used to represent the set of all possible legal moves and can be explored breath first or depth first. For more complex search problems, other techniques that rely on heuristics [7] are used.

C. 2. Knowledge representation

In artificial intelligence, knowledge must be acquired, represented in a special language (ex. first order logic, production rules or LISP) and coded in order to make it possible for machines to manipulate. Selecting the right type of representation for knowledge has always been critical in engineering intelligence. It appears that first order logic is the most common means – not the only one – of representing knowledge because it helps make fairly simple deductions (inferences) from known premises and enrich a given formal system with new facts.

C. 3. Machine learning

Machine Learning (ML) is the branch of AI that aims to give computers the ability to learn new facts with no explicit program and adapt to a new situation. To achieve this goal, ML uses several approaches based on models and algorithms. The main approaches are decision tree learning, association rule learning, artificial neural networks, genetic programming, inductive logic programming, clustering, Bayesian networks, and reinforcement learning. Tom Mitchell [4], an American computer scientist at Carnegie Mellon University gave the first formal definition of ML as follows: “A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E.”

C. 4.  Natural language processing

Natural Language Processing (NLP) is a subfield of AI that deals with human-computer interaction through the understanding, processing and elicitation of human language. One of the most notable early success stories of NLP is Joseph Weizenbaum‘s ELIZA program that simulates a therapist. Modern NLP heavily uses machine learning algorithms and has several areas of interest. These areas are automatic summarization, discourse analysis, machine translation, optical character recognition (OCR), sentiment analysis, speech processing (speech recognitiontext-to-speech), word sense disambiguation, information retrieval and extraction.

C. 5.  Artificial neural networks

Originally, an artificial neural network (ANN) [8] is an imitation of the natural human brain with its biological neurons and synapses. Its goal is to reproduce the computing power of the human brain. An artificial neural network is composed of a set of artificial neurons that process information and weighted links connecting these neurons. Every neuron receives some type of stimuli as input, processes it and sends through its related links an output to neighboring neurons as depicted in Fig.2. There are several kinds of ANN structures according to their topologies and search algorithms.

introduction-to-artificial-intelligence-neural-networks

Fig. 2: Basic structure of an Artificial Neural Network (ANN)
Adapted from two pictures on Wikipedia

D.      Applications of Artificial Intelligence

Nowadays, there are scores of hidden, apparent or explicit applications of AI in our life. The most notable are in medical science (disease detection, artificial muscles, robotic surgery assistance), law, finance industry (international stock trade), intelligent transportation (driverless trains in Japan and UAE), space exploration (NASA vehicle on Mars), engineering, artificial life, data mining, semantic Web, computer vision, pattern recognition (optical character recognition), handwriting recognition, intelligent agents (smart home security systems), robotics, super computing (Japanese K computer, American IBM Sequoia).

E.      Future Research and Perspectives in Artificial Intelligence

Thanks to the numerous applications [9], [10], the future of AI remains bright. The lack of common sense knowledge in machines, the scarcity of resources and funds do not seem to be major obstacles for this field on its way to realizing its full potential. The Association for the Advancement of AI [9] suggests that:

  • AI software will be used in the future to dramatically improve weather predictions for the benefit of forestry and agriculture;
  • AI devices and software will soon be used to clean up the environment and help identify patterns in our ecosystem and protect it;
  • AI-enabled robots will be heavily used as substitutes in repetitive tasks in factories, in environments too dangerous for humans like radioactive materials, in land mines, deep sea exploration, and firefighting;
  • Transhumanism as the ultimate application of AI to our lives, promises to improve our human experience by expanding the limits of our body and mind.

F.      References

[1] Turing, Alan (October 1950), Computing Machinery and Intelligence, Mind vol. 49, pp. 433–460;
[2] Weizenbaum, Joseph (January 1966), “ELIZA A Computer Program For the Study of Natural Language Communication Between Man And Machine”, Communications of the ACM, vol. 9, no. 1, URL= http://www.cse.buffalo.edu/~rapaport/572/S02/ weizenbaum. eliza. 1966.pdf ;
[3] McCorduck, Pamela (March 2004), “Machines Who Think: A Personal Inquiry into the History and Prospects of Artificial Intelligence”, ISBN 978-1568812052;
[4] Russell Stuart. and Norvig Peter. (December 2009), Artificial Intelligence: A Modern Approach, Prentice Hall, ISBN , 1,132 p. - ISBN 0136042597, URL= http://aima.cs.berkeley.edu/;
[5] Mitchell, Tom (1997), . Machine Learning, McGraw Hill. ISBN 0-07-042807-7;
[6] Nilsson, Nils J. (1980 ), Principles of Artificial Intelligence, San Francisco, Morgan Kaufmann ;
[7] Pearl, Judea (1984), Heuristics: intelligent search strategies for computer problem solving; Addison-Wesley Publishing Co., ISBN 0201055945 ;
[8] Rojas, R. (1996). Neural Networks: A Systematic Introduction. Springer, Berlin;
[9] Association for the Advancement of Artificial Intelligence. “Banking, Finance and Investment.” Oct. 30, 2010. (Nov. 26, 2010), URL = http://www.aaai.org/aitopics/pmwiki/pmwiki.php/AITopics/Banking;
[10] The AAAI conferences on Artificial Intelligence (1980-2013), URL = http://www.aaai.org/Conferences/AAAI/aaai.php