Artificial intelligence (Al) is defined generically as the demonstration of intelligence by computers or

machines; that is, making machines do things that are usually done by minds. According to
the Longman Dictionary of Psychology and Psychiatry, for an entity to display intelligence, it requires
a “general mental ability, especially the ability to make flexible use of memory, reasoning, judgment,
and information in learning and dealing with new situations and problems.” Intelligence in this sense
includes the ability to think, see, remember, learn, understand, and, in the long run, use common
sense. This is a useful working definition of intelligence although some AI researchers differ on how it
is to be applied to their work. Proponents of AI such as Kurzweil and Dyson argue that some of
today's computers and software (circa 2000) indeed exhibit intelligence by this general definition.
Kurzweil cites in support of his position the fact that on May 11,1997, IBM's chess playing computer,
“Deep Blue” (a RISC system/6000 scalable power parallel systems high-performance computer
programmed in C language), beat then-reigning human chess champion Gary Kasparov in a six-game
match. Time reported in its May 26, 1997, issue that Kasparov reflected on his experience: “The
decisive game of the match was Game 2, which left a scar on my memory and prevented me from
achieving my usual total concentration in the following games. In Deep Blue's Game 2 we saw
something that went well beyond our wildest expectations of how well a computer would be able to
foresee the long-term positional consequences of its decisions. The machine refused to move to a
position that had a decisive short-term advantage, showing a very human sense of danger.”

I.B. Information Processing of Symbolic Model and Artificial Neural Networks: Two Main Theories
of AI

John McCarthy first used the term artificial intelligence at the field's founding conference held at
Dartmouth College in 1956. The underlying assumption of the conference was that “[e]very 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.” This assumption led initially to the use of symbolic logic, patterned after
the theories of George Boole, as a model of mind. Research based on this assumption resulted in an
information processing or “cognitivist” theory of intelligence, one that emphasizes the functions of
discriminating, acquiring, recording, analyzing, and remembering information and the role these
functions play in decision making. Approaches based on this assumption are
called symbolic or symbol-processing AI. These systems tend to be designed and programmed from
the top down and proceed deductively rather than by means of training the computer and evolving
concepts inductively. Symbolic AI is structured in levels. The top is the knowledge level at which basic
rules and other knowledge are specified. This flows down to the symbol level at which the knowledge
is represented by symbolic structures. Finally, at the base, is the implementation level at which the
symbol-processing operations are actually carried out. Most AI systems of this type employ logical
reasoning methods to deduce actions to be taken and conclusions to be drawn.

Expert systems (ESs) are one popular manifestation of symbolic AI. Arguably the first AI program was
of this type. Dubbed the “Logic Theorist” it was written in the fall of 1955 by Allen Newell, Herbert
Simon, and J. C. Shaw to prove theorems in geometry. This paradigm reigned until the early 1970s
when, among others, Terry Winograd, a student of AI pioneer Seymour Papert, began to question the
strong representational assumptions being made by Newell and Simon. Winograd predicted that
future research would not uncover a unified set of principles of intelligence adequate to form the
knowledge level. Rather, he envisioned a variety of mechanisms and phenomena. Some would be
consistent with the symbolic AI paradigm; some, however, would be quite different. Roger Penrose
in The Emperor's New Mind observed that once a noncomputational element was introduced into
intelligence then computers could never do what humans could. During this time, neural net theory
emerged and a debate commenced.
The second major type of AI is called artificial neural networks (ANNs). ANNs are modeled on the
structure of the brain and nervous system, based in part on theories originally developed by John
von Neumann, Warren McCulloch, Walter Pitts, and Donald Hebb. Sometimes also
called subsymbolic AI systems, ANNs proceed from the bottom up, beginning with primitive signals.
As they move up the hierarchy signals are converted into symbols. Proponents of this approach, such
as MIT professor Rodney Brooks, believe that AI programs should be coupled closely with the
physical world and made to evolve much as human intelligence has evolved over hundreds of
thousands of years. The strategy they propose is called subsumption architecture. It begins
inductively by simulating lower level animal functions and proceeds by adding competencies at
higher levels. Whereas symbolic AI assumes that symbolic representation is necessary and
fundamental to general intellegence, ANN downplays the importance of representation or, as in the
case of Brooks, denies that it is needed at all.

ANNs emulate the processing capabilities of the human brain by simulating a distributed network
consisting of tens of billions of cells that form arrays of millions of neurons (approximately 1011) that
are connected by a dendritic tree and excited by as many as 1014 synapses. (The number of synaptic
connections in the human brain is likely greater than the number of atoms in the universe.)
These learning systems are trained rather than programmed. The ANN approach is also
called connectionism. Others have labeled it the animat approach. The symbolic and ANN
approaches differ in their understanding of the nature of cognition and how it works and also in how
one knows if cognition is functioning adequately. Two researchers, Rosenschein and Kaelbling,
propose an intermediate approach in which desired behavior is programmed at a high level
(symbolic) and a compiler is employed to create behavior evoking circuitry at a lower level (ANN).
Although the underlying theories of AI and their implementing technologies are quite different and
the competing theorists continue to debate their relative merits, their moral implications are similar.

Among the computer-based technologies that have been used as a means toward achieving AI's
goals are the following:

Expert systems: Computer programs that support workers in complex situations in which experts
outperform nonexperts.

Neural networks: As described above, these ANN computer programs identify objects or recognize
patterns after having been trained on a group of examples.

Natural language processors: Computer programs that translate or interpret language as it is spoken
by normal people.

Visual processing: Computer programs that emulate the human retina by means of algorithms. Some
of these programs can recognize objects and shadows and identify small changes from one image to
another.

•
Robots: Computer-based programmable machines that have physical manipulators and sensors.
Robots often use visual processing and draw on symbolic reasoning or neural networks for
intelligence. They may perform tasks normally done by humans with greater speed, strength,
consistency, and precision.

Fuzzy logic: Reasoning based on imprecise or incomplete information in terms of a range of values
rather than point estimates.

Case-based reasoning: A computer program that searches for previous cases that are similar to the
one in which a decision must be made.

Intelligence software agents: Small computer programs constructed on AI principles that performs
electronic tasks for their masters. Agents or bots may be launched into a system or network to work
in the background while other processing is going on in the foreground.

I.C. AI's Goals and Criteria for Achieving Them

Two criteria are generally used to determine whether a machine or computer program has achieved
the goals of AI research: (1) It displays general human-type intelligence with all of its subtleties and
ramifications, or (2) it performs specific tasks in a manner indistinguishable from characteristic
human efforts. To date AI technologies have been applied primarily toward projects, such as chess
playing, directed at the second criterion—task performance. Although developments have lulled
from time to time, since its inception AI technology has grown in productivity and use, especially
since the advent of agents on the Internet. AI software and the computers that execute it have
improved in performance—speed, scope, cost, etc. As a result, their use and extent of application
have increased, especially as parts of other computer-based applications.

Show more

View chapterExplore book

Modeling of tacit knowledge in industry: Simulations on the variables of industrial processes

Cláudio Roberto do Rosário, ... Bruna Bueno Mariani, in Expert Systems with Applications, 2015

2.5 Artificial Intelligence (AI)

In this section we discuss the concept of Artificial Intelligence and its articulations with the expert
system.

For Coelho (1995), Fernandes (2005) and Carneiro, Grimoni, and Udaeta (2007) all study of human
mental ability happens through the use of computational models. Basically, Artificial Intelligence can
be defined as an alternative of making a machine to act like a human being when making decisions.
This involves a computer system designed with the same mechanisms that a human being uses for
his/her everyday actions or more complex decisions.

For this study, a “how to do” concept will be utilized, starting from the forms of knowledge
representation from the semantic network type and production rules. For Russell et al. (2003),
Artificial Intelligence is a new science which represents the specification of knowledge, including
activities such as problem solving in decision making, having as a general objective to reproduce, by
means of machines, human activities qualified as intelligent ones.

Russell et al. (2003), affirms that Artificial Intelligence renders intellectual tasks systematic and
automatic and, therefore, it is potentially relevant in every sphere of human intellectual activity.

AI has three different approaches according to Russell et al. (2003) and Bittencourt (2006):

Symbolic: it is based on the development of systems such as cognitive models of rationality, thinking,
and learning, among others.

Connectionist: it is based on the development of systems as structure model and the

human brain operation.

Distributed: it is based on the development of systems as models of human behavior (interaction

process).

In this study, it was used a symbolic approach with the use of the expert systems technique.

View article

A review on digitalization trends in patent information databases and interrogation tools

Lothar Walter, ... Jan Kebel, in World Patent Information, 2022

4.4 Artificial intelligence

Artificial intelligence is currently one of the fastest growing technology fields and already generates
enormous revenues worldwide [24,56]. Hence, this domain significantly influences modern patent
search and analysis as well. Although artificial intelligence has been discussed and significantly
advanced since the 1950s, there is no universally accepted definition of it yet [24,57–59]. The
Cambridge Advanced Learner’s Dictionary [60] defines artificial intelligence as:

“the study of how to produce machines that have some of the qualities that the human mind has,
such as the ability to understand language, recognize pictures, solve problems, and learn.”

This definition, like many others, reveals that artificial intelligence is compared with human
intelligence on a regular basis. Accordingly, intelligent machines and computers are supposed to
perform tasks that require human intelligence, as characterized by high levels of adaptivity and
learning ability. This kind of capability is attributed to machine learning, which has established itself
as an important research field within artificial intelligence [6,61]. Consequently, it is no surprise that
machine learning has also been identified as a digitalization trend in the field of patent search and
analysis [62].

Machine learning focuses on ‘artificial’ knowledge creation from experience, i.e. an artificial system
learns to generalize distinct patterns from many examples. For this purpose, machine learning
algorithms build statistical models based on training data [63]. As an almost unlimited amount of
patent data is available worldwide, machine learning is an appropriate means for patent search and
analysis.
Machine learning is subdivided into reinforcement learning, supervised learning, and unsupervised
learning. Reinforcement learning is the least prominent mode in the area of patent search and
analysis [7] and refers to the learning of “what to do – how to map situations to actions – so as to
maximize a numerical reward signal” [64]. For example, reinforcement learning is used to cluster or
classify patents without reliance on manual effort [65].

Supervised learning refers to predicting a known outcome based on a number of input measures
[66]. For example, supervised learning is leveraged in patent analysis to predict the value or potential
of technologies [67], and in patent search to identify relevant patents for interdisciplinary
technologies fields such as FinTech (financial technology) [68] or quantum computing [69], or to
identify relevant prior art [70].

Finally, in unsupervised learning the outcome is unknown and the algorithms try to detect patterns in
data sets [66]. Patent search and analysis benefit from unsupervised learning primarily due to its
exploratory nature. A comprehensive understanding of a patent data set or a focused technology is
not required. For example, unsupervised topic modeling assists patent search in refining the
selection of patents [71] or patent analysis in identifying emerging trends or novelty in patents
[72,73].

Apart from machine learning, natural language processing has also been identified as a trend in
artificial intelligence. Natural language processing has become significant for patent searches and
analyses, as it enables the processing and analysis of large amounts of natural language data. Thus,
the subject of natural language processing is human language and its properties. The ability of
humans to speak and use language in written form counts as one of their most fundamental
characteristics and is central to communication among the species [24]. Natural language processing
takes over human performance in language processing and tries to interpret contained meaning [74].
This represents a special challenge for a computer in comparison to the traditional analysis or
computation of simple data. After all, understanding language is not only about the evaluation and
syntax of words, but more specifically about semantic relationships in the context of interaction
between computers and human language. Another characteristic of natural language is its flexibility
to convey the same message by means of different sentences and phrases, so that an algorithm
cannot exactly match a message to the sentence that renders it [75].

In a nutshell, natural language processing – as a branch of linguistics – deals with the meaning
analysis of linguistic expression. In particular, it has laid a theoretical foundation for the analysis
of semantic structures, such as those given by subject-action-object structures or n-grams, and finds
much application in patent data analysis [76–78].

In addition to machine learning and natural language processing, machine translation, i.e. the
automatic translation of patent texts into other languages, is a further trend in patent search and
analysis [79–81]. Especially Asian patents necessitate an above-average number of translations into
English, for example due to the language barrier to Western nations. Understanding different
languages is therefore important for artificial intelligence applications. Both the extraction of
information and the translation of a text presuppose that the message’s content is understood and
interpreted correctly in its original language. For both considered areas of natural language
processing, information extraction and machine translation, the concept of automatics is crucial.
Souili et al. [82], for example, acknowledge patents as an important source of technical knowledge
while criticizing that this knowledge is not yet being exploited and used sufficiently. There is an
urgent need to automate the process of extracting information from patents. This should not be
limited to data mining of structured data, but also applies specifically to text mining for the analysis
of unstructured patent data [82]. In this way, in addition to automatic translation (which is often
already available as a sophisticated search feature), it becomes possible to analyze patent claims,
classify texts or summarize entire patent specifications [83,84].

These applications are often enhanced by further artificial intelligence techniques. Particularly
machine learning, often in the form of so-called deep learning based on artificial neural networks,
plays a significant role in making applications adaptive [85,86]. Algorithms can be developed to not
only process natural language, but also automatically increase their own understanding of it, thus
improving their accuracy in interpretation. With regard to the peculiarities of patent language, an
algorithm development that is explicitly tailored to the processing of patents therefore becomes
significant if, for example, patent claims are to be analyzed [87].

Another important digitalization trend in the context of patent information databases and
interrogation tools pertains to image recognition, which is used to identify patterns in images. Apart
from text, patent documents also contain drawings that incorporate valuable information about the
inventions that are described. One objective of image recognition is to classify patents and their
contents on the basis of the drawings they contain. For example, looking at the visual information
comprised in patent applications is of central importance for patent examiners in the context of
novelty assessment [88]. Similarly, there are approaches to searching patent databases by means of
a reference image in order to detect patents that contain a similar or even the same image [89]. This
is especially important with regard to searching biosequences (e.g. fertilizers or gene sequences)
[90–93]. Image recognition is faced with particular challenges in patent documents, as patent
drawings are usually monochromatic, do not contain any information on texture, and due to the
patent’s inventive step mostly describe (at least partially) unknown structures [88,94]. These
challenges call for the development of new image recognition approaches that are specifically
tailored to patent documents and the characteristics of the drawings contained by them. In this
context, Ni et al. [94] identify opportunities to combine text-based methods with the analysis of
drawings. After all, the drawings in patent documents often are crucial for understanding the
content, so that a purely text-based classification would fall short of analysts’ goals and lead to
fuzzier results [94]. Moreover, according to Bhatti et al. [88] and Rusiñol et al. [95], important
features of the invention are often described exclusively in the drawings, which leads to an actual
loss of key information when only considering the patent text.

Finally, it should be mentioned that recent scientific studies largely focus on optical character
recognition for capturing bibliographic and textual information from pre-1975 US patents, as these
patents are only available in the form of image data. By transforming the image files into structured
text, scholars have been analyzing the geographical progress of patents [96] and technological
developments since the mid-19th century [97].

Show more

View article

Improving cloud/edge sustainability through artificial intelligence: A systematic review

Belen Bermejo, Carlos Juiz, in Journal of Parallel and Distributed Computing, 2023

2.3 Artificial intelligence

Artificial intelligence (AI) can be defined as an area of study in the field of computer science
concerning the development of computers able to engage in human-line through processes such as
learning, reasoning, and self-correction. Moreover, artificial intelligence is the study of techniques to
use computers more effectively by improved programming techniques [48].

Many problems in AI can be solved theoretically by intelligently searching through many possible
solutions. Traditionally, reasoning can be reduced to performing a search. Planning algorithms search
through trees of goals and subgoals, attempting to find a path to a target goal, a process called
means-end analysis. Robotics algorithms for grasping objects use local searches in configuration
space.

However, artificial intelligence began to have an application in the area of computing, and especially
in everything related to cloud services when the demand for these services was growing
exponentially. The allocation of computing resources to different users can be treated as a
traditional optimization problem. That is, all the classic techniques can be applied to the current
computing ecosystem. Even these techniques have been improved, giving rise to machine
learning and deep learning [32].

View article

Artificial Intelligence: Search

R.E. Korf, in International Encyclopedia of the Social & Behavioral Sciences, 2001

See also:

Artificial Intelligence: Uncertainty; Artificial Intelligence in Cognitive Science; Decision Support

Systems; Deductive Reasoning Systems; Intelligence: History of the Concept; Scientific Discovery,
Computational Models of

View chapterExplore book

Ethical Issues in Artificial Intelligence

Richard O. Mason, in Encyclopedia of Information Systems, 2003

I.A. Artificial Intelligence Defined

Artificial intelligence (Al) is defined generically as the demonstration of intelligence by computers or

machines; that is, making machines do things that are usually done by minds. According to
the Longman Dictionary of Psychology and Psychiatry, for an entity to display intelligence, it requires
a “general mental ability, especially the ability to mak