What is Artificial Intelligence and What are its Subsets

Artificial Intelligence:

Artificial intelligence (AI) is advancing at a fast pace. While science fiction frequently depicts AI as humanoid robots, AI may refer to anything from Google's search engines to IBM's Watson to autonomous weaponry. It is the capacity of digitized computers or computer-controlled machines to do tasks often associated with intelligent humans, and it is a broad field of computer science focused on developing clever robots capable of completing tasks that normally need human intellect. The phrase is commonly referred to as the aim of building systems with human-like cognitive processes, including the reasoning ability, discover meaning, generalize, or learn from prior experience.

Artificial intelligence algorithms are programmed to make choices based on real-time data. They differ from passive machines, which can only respond in mechanical or programmed ways. They integrate information from many sources using sensors, digital data, or remote inputs, quickly evaluate the material, and act on the insights produced from that data. As such, they are purposefully developed by people and reach conclusions based on their immediate analysis.

 

                                              [Image is taken from unsplash.com]

Artificial intelligence is currently appropriately characterized as narrow AI (or weak AI) since it is meant to do a certain job, such as merely facial recognition, internet searches, or driving a car. However, many scientists' long-term objectives are to develop generic AI (AGI or strong AI). While narrow AI may beat humans in a specialized skill, such as chess or problem solving, AGI would surpass humans at almost every cognitive endeavor.

The objective of limiting AI's negative influence on society inspires study in a wide range of fields, from economics and law to technological problems like verification, validity, security, and control. Since the invention of the digital computer in the 1940s, it has been proven that computers can be trained to do extremely complicated tasks with high accuracy. Nonetheless, despite ongoing advancements in computer processing speed and memory capacity, there are no programs that can match human flexibility across broader areas or in activities requiring a high level of everyday knowledge. In the long run, a crucial concern is what would happen if the search for strong AI is successful and an AI system outperforms humans in all cognitive activities.

 

                                                    [Image is taken from unsplash.com]

Although some programs have achieved the competence levels of human specialists and professionals in executing certain tasks, artificial intelligence in this restricted meaning may be found in applications ranging as medical diagnosis, computer search engines, and voice or handwriting recognition. Some argue that strong AI will never be developed, while others argue that the development of superintelligent AI would always be good. 

History of Artificial Intelligence:

Alan Mathison Turing, a British logician, and computer pioneer did the first significant work in the subject of artificial intelligence in the mid-twentieth century. Turing developed an abstract computing machine in 1935, consisting of an infinite memory and a scanner that travels symbol by symbol across the memory, reading what it finds and recording new symbols. The scanner's operations are regulated by an instruction program that is likewise stored in memory in the form of symbols. This is Turing's stored-program notion, and it implies that the machine can operate on, and so change or improve, its program. Turing's notion is currently referred to as the universal Turing machine. In essence, all contemporary computers are universal Turing machines.

 

                                                    [Image is taken from unsplash.com]

Turing was a key cryptanalyst at the Government Code and Cypher School in Bletchley Park, Buckinghamshire, England, during WWII. Turing could not begin work on the idea of creating a stored-program digital computing machine until the end of World War II in 1945. Nonetheless, he gave much attention to the subject of machine intelligence throughout the conflict. Donald Michie, one of Turing's colleagues at Bletchley Park, subsequently remembered that Turing frequently described how computers might learn from experience as well as solve new issues using guiding principles, a technique now known as heuristic problem-solving.

Turing presented what was arguably the first public presentation on computer intelligence in London, 1947, noting, What we want to see is a system that can learn from experience, and also that the potential outcome of letting the machine modify its instructions offers the mechanism for all of this. In a paper titled "Intelligent Machinery," he proposed many of the fundamental principles of AI in 1948. Turing, however, did not publish this work, and many of his concepts were eventually recreated by others. One of Turing's early ideas, for example, was to train a network of artificial neurons to do certain tasks.

The Turing Test:

Turing dodged the usual argument over the concept of intelligence in 1950 by proposing the Turing test, a practical measure for computer intelligence that is now known simply as the Turing test. A computer, a human interrogator, and a human foil take part in the Turing test. The interrogator seeks to establish which of the two individuals is the computer by asking questions of the other two participants. The keyboard and display screen is used for all communication. The interrogator is free to ask as many probing and broad-ranging questions as he or she wants, and the computer is free to do whatever it takes to compel a false identification.

The foil must assist the interrogator in making an accurate identification. A variety of persons perform the roles of interrogator and foil, and if a significant proportion of the interrogators are unable to identify the computer from the human being, therefore the computer is deemed an intelligent, thinking being. Hugh Loebner, an American philanthropist, established the annual Loebner Prize competition in 1991, offering a $100,000 prize to the very first computer to complete the Turing test and paying $2,000 each year to the finest effort. No AI software, however, has gotten close to passing an unmodified Turing test.

 

                                                           [Image is taken from unsplash.com]

Subsets of Artificial Intelligence:

Now that we have talked about the history and what is Artificial Intelligence, we are now going to mention below its subsets or the types of Artificial Intelligence. Following are the most common subsets of AI:

● Machine Learning

● Deep Learning

● Natural Language processing

● Expert System

● Robotics

● Machine Vision

● Speech Recognition

Machine Learning:

Machine learning is a field of research in which a computer learns without being explicitly taught using accessible data/historical data. Today, machine learning (ML) is perhaps the most relevant subset of AI to the ordinary organization. As stated in the Executive's Manual for Real-World AI, a recent study paper conducted by Harvard Business Review Analytic Services, machine learning (ML) is a mature concept that has been around for quite some time. In machine learning, we will not explicitly write the code for each sort of problem, as the computer will try to find out how to tackle the problem. In machine learning, we will take a different approach; instead of giving direct instructions, a specific algorithm will identify patterns and anticipate the best potential output based on those patterns.

 

                                                          [Image is taken from unsplash.com]

Machine learning is further classified into three types:

1. Supervised Learning:

Supervised learning is a form of machine learning in which the computer learns from known datasets (sets of training instances) and then predicts the output. A supervised learning agent must determine the function that best matches a provided sample set.

Supervised learning is further subdivided into two types of algorithms:

➢ Classifications

➢ Regression

2. Reinforcement learning: 

Reinforcement learning is a kind of learning in which an AI agent is educated by issuing orders and receiving a reward as feedback for each action. The agent enhances its performance by utilizing this feedbacks.

Reward feedback can be either positive or negative, which implies that for each successful activity, the agent receives a positive reward, while for each incorrect action, it receives a negative reward.

There are two kinds of reinforcement learning:

➢ Positive Reinforcement

➢ Negative Reinforcement

3. Unsupervised Learning:

Learning without supervision or training is referred to as unsupervised learning. Unsupervised learning involves training algorithms on data that has not been tagged or categorized. Unsupervised learning requires the agent to learn from patterns that do not have matching output values.

Unsupervised learning algorithms are divided into two types:

➢ Clustering

➢ Association

● Deep Learning:

Machine learning, as we all know, is a subset of Artificial Intelligence. Machine learning is a subset of deep learning. The biggest distinction in the deep learning model is that it improves with experience without even any particular instruction. In deep learning, a model gets rewarded and disciplined depending on feedback, and it then adjusts weights for the input variables accordingly. According to one explanation supplied by deep AI, deep learning employs claimed neural networks that learn from processing the labeled data supplied during training and uses this response key to understand what qualities of the information are required to construct the appropriate output.

To construct models, deep learning needs a vast quantity of data. Neural networks serve as the foundation for deep learning models. Deep learning algorithms employ neural networks that are copies of neurons in the human brain. In the human brain, neurons interact with one another to build a deep neural network; similarly, in deep learning, artificial neurons interact with one another to form a deep neural network, which is why this type of learning is known as deep learning. Amazon and Netflix use deep learning to make product and content recommendations. It runs in the background alongside Google's speech and picture recognition engines. Deep learning is suited for supercharging preventative maintenance frameworks due to its capacity to break down a large amount of high-dimensional information.

In AI, a perceptron is a copy of a human neuron that is linked together to build deep neural networks. A perceptron has input nodes that are similar to dendrites in the human brain, an actuation function to make a tiny choice, and output nodes that are similar to axons in the human brain.

A deep neural network is divided into three sections:

The input layer is the initial layer in the network, and it receives raw information and processes it before passing it on to the next layer of neurons. Hidden Layers are the intermediate layers, and the number of layers varies from one to hundreds depending on the intricacy of the issue. Each layer of neurons processes the information provided from the input layer before passing it on to the next. The final layer of the neuron provides output to the user.

 

                                                     [Image is taken from unsplash.com]

● Natural Processing Language:

Natural language processing is a computer science and artificial intelligence subfield. NLP allows a computer system to comprehend and process human language, such as English. Natural Language Programming is a branch of linguistics concerned with the interaction of a computer and a person. There are two ways to doing NLP: rule-based NLP and statistical NLP. The rules in Rule-based are described based on the grammatical rules of a language. In statistical NLP, a vast amount of data termed Corpora of real-world communication samples is fed into the system.

The algorithms then attempt to learn from the data, and a model is created that can interpret and make sense of any command based on prior experience. Statistical NLP is superior because humans do not adhere to grammatical rules in general communication, and speaking styles vary from person to person. In rule-based NLP, defining rules for everyone becomes extremely tough, and some rules oppose others, making it much more difficult.

 

                                                         [Image is taken from unsplash.com]

● Robotics:

This has emerged as a very hot field in artificial intelligence. An interesting sector of inventive work revolves mostly around the design and development of robots. Robotics is a subfield of artificial intelligence and engineering that focuses on the design and manufacture of robots. It is also an interdisciplinary subject of science and engineering that includes mechanical engineering, electrical engineering, computer science, and a variety of other disciplines. It makes decisions on the design, manufacture, operation, and application of robots. It controls computer systems to govern them, provide intelligent outcomes, and modify data.

Robots are pre-programmed devices that can carry out a sequence of tasks automatically or semi-automatically. Robots are used daily to direct tasks that are difficult for humans to accomplish consistently.

AI may be used in robotics to create intelligent robots capable of performing tasks using their intellect. To enable a robot to execute increasingly complicated jobs, AI algorithms are required. Scientists in artificial intelligence are also developing robots that use machine learning to establish interaction levels at social levels. AI and machine learning are now being used on robots to create intelligent robots that can engage socially like people.

 

                                                    [Image is taken from unsplash.com]

● Expert Systems:

An expert system is a type of artificial intelligence application. Expert systems are computer programs in artificial intelligence that rely on acquiring information from human experts and putting that knowledge into a system. Expert systems are designed to mimic the decision-making abilities of human experts. Instead of traditional procedural code, these systems are meant to handle complicated problems using bodies of knowledge. A Suggestion for a spelling problem when entering text in the Google search box is one example of an expert system.

● Machine Vision:

Machine vision is a computer vision application that allows a machine to recognize an item. Machine vision uses one or more recording devices, analog-to-digital conversion, and digital signal processing to gather and analyze visual data. Machine vision systems are configured to do certain tasks such as counting objects, reading serial numbers, and so on. Computer systems cannot see in the similar way that human eyes do, but they are not limited by human constraints such as being able to see through a wall. An AI agent can see past walls with the assistance of machine learning and machine vision.

● Speech Recognition:

Speech recognition is a technique that allows a machine to interpret spoken language and convert it into a machine-readable format. It is also known as computer voice recognition and automated speech recognition. It is a method of communicating with a computer, and a computer may do a certain job according to the instruction. Some voice recognition software has a restricted vocabulary of words and phrases. To comprehend and accomplish a given activity, this software requires unambiguous spoken language. Today, there are a variety of applications and gadgets that use voice recognition technologies, such as Cortana, Google virtual assistant, Apple Siri, and others.

 

                                                      [Image is taken from unsplash.com]

Artificial Intelligence is ever-growing. These are the few of its subsets that are common in technology as compared to others. AI has numerous subsets, which consist of mathematics, science, and so on. We have discussed some important subsets like machine learning, deep learning, Natural language processing, etc. 

Relative blog:

How to Solve your Business Problems using AI as a Service