Artificial intelligence

Artificial intelligence (AI) is the science and engineering of constructing "intelligent" machines. However, there is not a commonly accepted definition of what it would mean for a machine to be truly "intelligent." More broadly, AI is a field of science and engineering involved with the study, design and manufacture of systems that exhibit behavior that is reactive to conditions within the surrounding environment, adaptive toward the accomplishment of system goals, autonomous to varying degrees, and complex in breadth or scope. As the field progresses the context and expectations for the extent of reactiveness, adaptiveness, autonomy and complexity of the systems has grown.

Artificial Intelligence has been described as a "broad sub-field within computer science" by many authors. Certainly, AI has its strongest roots within computer science, but it also has an interdisciplinary character that both draws from and feeds back to the fields of mathematics, philosophy, language, robotics, and possibly others. More recently, an interdisciplinary approach to cognition and intelligence has developed known as cognitive science. Though many of the methods are similar, cognitive science shifts the focus, by looking at the methods of artificial intelligence as a means to gaining greater insight into the basic mechanisms of perception, memory and thought.

Will Machines Become Intelligent?

While earlier AI systems (circa 1970s and 1980s) were well-described as narrow, brittle, and highly engineered, often relying on expert systems or knowledge bases, progress is being made in many areas.

It is generally recognized that much of human behavior is reactive, adaptive, autonomous and complex as used in this definition and constitutes an instance of a "natural intelligence." Whether an AI system could rival the general cognitive capacity of humans is a matter of some debate that mostly centers on "free will", "self directed goals", "self awareness" and similar highly complex human behaviors. One of the first practical commercial projects did not entail the complex human traits but rather that of animal behavior. In the late 1990's, artist Sorayama was approached by the Sony Corporation AI engineers to design an organic robotic form. It became the famous "AIBO" dog, the first artificial intelligence pet, which received Japan's highest design award and was acquired by the Museum of Modern Art and the Smithsonian Institution for their permanent collections.

AI has spawned many great successes such as advances in problem solving algorithms, search algorithms, knowledge representation, machine learning, object oriented programming, natural language understanding, and expert systems. However if you ask many experts in these areas about their craft, you may be told "What we do is not AI."

Sub-fields

• Semantic networks
• Rule based inference systems
• Artificial neural networks and connectionism
• Genetic algorithms
• Fuzzy logic
• Constraint based programming
• Natural language processing
• Knowledge representation
• Bayesian methods
• Ontologies
• Evolution of artificial intelligence
• Machine Learning

Artificial intelligence by means of evolution

Cognitive scientist Lee Spector (Hampshire College, Amherst, MA) discusses the potential of exploiting evolutionary processes (e.g., natural selection) to create intelligence in machines, and he explores whether evolved artificial intelligence could meet the goals of AI in their broadest ambitions (Spector 2006). He claims a marginalization of the notion of evolved artificial intelligence in most AI texts, with the notable exception of N.J. Nilsson’s book (Nilsson 1998). He recognizes the evolutionary nature of genetic algorithms, but claims AI researchers relegate genetic algorithms to special purposes and not specifically to create evolved artificial intelligence machines.

Spector asserts that inasmuch we appreciate evolved intelligence (e.g., human intelligence) as the predominant form of intelligence in the world, AI workers should in principle take a serious interest in evolved AI.

He notes that the mechanisms involved in biological evolution has advanced enormously since Darwin: “We now know, for example, that biological evolution depends in important ways on genetic representation, on ecological interactions within and among populations, and on genetic control of gene expression, reproductive strategies, development, and learning.” On a positive note, he states:

The problem-solving performance of evolutionary algorithms has advanced significantly in the past decade or so, to the extent that human-competitive results have recently been achieved in several areas of science and engineering; these include evolved designs for antennas [cites (Lohn et al. 2004)], photonic crystals [cites (Preble et al. 2005)], quantum computer algorithms [cites (Spector 2004)], and even search heuristics [cites (Fukunaga 2002)]. Many of these results have been achieved using systems that incorporate at least a few of the insights from recent biological advances—for example, many involve the genetic representation of developmental processes instead of “adult” phenotypes—but there is still a long way to go if the computational work is to catch up with biology. Nonetheless, work in evolutionary computation is moving rapidly forward and it is doing so within an increasingly mature and stable research community. One indication of this progress is the recent establishment, by the Association for Computing Machinery, of a full-fledged Special Interest Group on Genetic and Evolutionary Computation (SIGEVO).

Spector’s article merits a read for his concluding speculations on the future of the interaction of human design and evolved design in the creation of machines with artificial intelligence.

Preserving the departed with AI

Using various AI technologies, we may not need to resort to such, to some, morbid or fantastic techniques as cryogenics to preserve the departed for resurrection by future technologies capable of such revitalizations, and cures of whatever led to the departure. Now projects underway hope to implement “AI archiving” — using computer imaging with the sophisticated graphics of video game technology to create virtual digital people with lifelike appearance in image, voice, body language, and at least with some of the knowledge and memories of the departed individuals. Recently the National Science foundation has funded such projects at the University of Central Florida (Orlando) and the University of Illinois (Chicago). According to science writer Patrick Tucker (Tucker 2007):

Stringing together the diverse technologies of live computer animation with artificial intelligence, speech recognition, speech synthesis, and facial expression recognition will be the biggest challenge of the project. Leigh doubts that the team will be able to create a totally credible and naturalistic avatar within the project's three-year time frame, but he is optimistic that the work will stimulate interest in the field.

Founding parents of AI

(Ordered by alpha)

- John McCarthy -- Professor Emeritus (as of 1/1/2001) of Computer Science at Stanford University. His home page there is at: John McCarthy's Home Page

- Marvin Minsky -- Pioneering author, researcher, professor, and leader of MIT's Artificial Intelligence Lab. His web page is at: Marvin Minsky

- Allen Newell -- Self described as a scientist. A comprehensive biography and bibliography at: National Academies Press

- Herbert A. Simon -- Nobel Laureate (1978, Economics), polymath, scientist, author. His departmental pages and other links are at: Herbert A. Simon

- Alan Turing -- Posited the Turing Test in 1950 in a paper he authored titled, "Computing Machinery and Intelligence" More info is at: The Alan Turing Home Page