Ai is everywhere, it has incorporated into every aspect of our life, unknowingly. It changed the way we live by simplifying things we do in our routine, like shopping, traveling, man-machine interaction. AI almost gained control of our actions. It decides what we shop for, by showing ads and recommendations while you are shopping, AI trip advisors suggest you a travel destination and the best vacation packages for your budget.
AI helping Businesses and financial institutions to serve their customers better with automated question and answer chatbots. AI also defines our social media feeds, how many of your Facebook friends have not been showing up on your wall, even they active in social media? do you have any idea? Because AI knows what and who you are interested in.
Have you ever think what AI is and how does it work? Then this ‘A.I. insider’ is for you. Also, check out the recent breakthrough inventions in AI that surpassed human capabilities.
Table Of Content:
1. What is AI?
Artificial intelligence is a division of computer science that makes computers and machines replicate the natural intelligence performed by humans, such as visual perception, understanding, reasoning, problem-solving, decision making, and translating. AI also sometimes termed machine intelligence. As it enables machines and computers to think and learn from experiences. AI works around the use of algorithms in the approach of solving problems.
In simple words, “Artificial Intelligence makes computers intelligent and conscious by enabling devices to perceive its environment and take necessary actions according in a way to successfully achieve its goals.”
What will comes to your mind when you think of AI?
It’s not your problem…
From Terminator to Ex-Machina most of the science-fictions portrayed AI as a super-robot with human-like characteristics. As AI is a virtual object, it can be disguised into your mobile, camera, a self-driving car, Google’s search algorithm, robot, or it could be a deadly autonomous weapon, don’t worry I just mean to say that AI could be like anything in shape and size that gives human-like thinking properties.
2. How Does AI Works?
Artificial Intelligence uses machine learning techniques like deep learning and artificial neural networks to learn from given data and use those learnings to achieve its goals. These algorithms capable of enhancing themselves by learning new techniques from the actions that worked well in the past. And, some algorithms are capable of update/write other algorithms of their own from previous experiences and learnings.
For an instance, AI enables computers to perceive their surroundings by collecting data from various sensory inputs, and it analyses the scene into objects, their features, and relationships. These learnings/knowledge helps AI to take action/responding to the situations by the set of rules programmed by humans or maybe sometimes machines.
AI is an umbrella term, it works accompanied by its subset technologies like machine learning, artificial neural networks (ANN), natural language processing (NLP), computer vision and speech recognition, etc. Using these technologies, computers can be trained to accomplish specific tasks by processing large amounts of data by recognizing patterns and relationships between the data sets. AI works very similarly to natural (human) intelligence, it performs various stages of operations of intelligence are as follows:
2.1 Machine Perception (getting aware of situations to act wisely):
Mostly, Machine perception deals with the hardware side of the AI, like cameras, sensors, and microphones to get aware of the surroundings/situations. As it inspired by human intelligence, it also works very similarly to us to deduce its environment. The consciousness of AI is possible by scanning the surrounding by means of various sensors and cameras.
Machine vision (computer vision) uses actual cameras for facial, object, and pattern recognition. Speech recognition requires microphones and robots can aware of their geo-location using GPS sensors. Artificial perception will help in building self-controlled cars, security systems, and robots, etc. It’ll help autonomous cars to read and understand the traffic, lane, traffic signals, and signboards on the road and drive safely accordingly.
Robotics is another major field, where development in AI can bring more innovations and possibilities. Intelligent robots are capable of getting aware of their surroundings and responding accordingly. In terms, it paved the way to build advanced security robots that can identify thieves, most wanted criminals, scan the license plate, and help needed, and you can find more at robots helping humans.
2.2 Machine Learning (learning from previous experiences to solve future problems):
Learning evolves with computer algorithms. AI is capable of learning from experience without being explicitly programmed by analyzing and processing the information. Learning will allows AI to identify the patterns and frequencies in data so that algorithms acquire skills. For instance, a simple program of a puzzle game can try random moves until it gets to succeed and remembers the moves in simple mathematical values. So, when a similar problem given to the computer, it can solve the problem immediately by memorizing the previous experiences.
In simple words, learning is simply adding new facts to the existing knowledge base, then analyze, infer, and update the existing knowledge base to avoid redundancy and faster retrieval of facts. Thus, the AI can teach itself how to play games and solve real-life problems. So, it can help businesses and consumers by making better predictions and recommendations. These learning models will adjust through constant training and added data when their prediction went wrong.
2.2.1 Supervised Learning:
This is the basic form method of machine learning, where the defined sets of data act’s as a supervisor to train the machine. In supervised learning given datasets are well labeled and already tagged with the correct answers will teach computers to make correct decisions and future predictions. So, computers can easily generate the correct answer when provided with new examples of datasets. For example, If the shape of an object is rounded and has a pentagon-shaped pattern in white and black color then it will be labeled as – Football.
So, the next time when the computer sees a round-shaped object with the above-mentioned properties, it will conclude it as a football.
2.2.2 Unsupervised Learning:
Unlike supervised learning, Unsupervised is capable of learning on its own without a supervisor/teacher. In unsupervised learning, data inputs will be raw and unclassified. Though, the learning algorithms will figure things out by observing the data structures and relationships between datasets without any provided training. For example, suppose it is provided with an image of apples and bananas, the machine has no idea how the apples and bananas look like, it won’t label them by their names but it categorizes the objects by their color, shape, and patterns.
With unsupervised learning, the machine can identify the similarities, differences, and patterns and able to categorize the objects from the given inputs.
2.2.3 Reinforcement Learning: My life my rules (hit & trial)
Reinforcement learning is a different approach to get machine learned. Unlike supervised learning, it presented with a set of goals, also known as ‘goal-oriented learning’. It works based on the hit and trial concept. Learning agents will find the best possible path or behavior in a way to solve problems according to the given situation. Then the learning agents will be rewarded or penalized for their correct and incorrect moves respectively and the machine will learn itself based on positive rewards.
Reinforcement learning is the ability of agents to interact with the environment and performing an action or several actions until it finds the best possible outcome. Reinforcement learning enables computers to learn from their own experiences without any supervisor or training datasets. The best examples of reinforcement learning are video gaming and puzzles. If suppose the computer has given a puzzle (see the below image) to find the path, where the goal is to find the path from ‘IN’ to ‘OUT’.
Then the computer performs all the possible moves and rewards the correct moves and penalizes the wrong ones. It can find the best and shortest possible path to ‘OUT’ with the most rewarded moves.
2.3 Reasoning: (Make machines think in a way to gain human-like consciousness)
Automated reasoning in artificial intelligence is the ability to make the machine think rationally. It is the general process of intelligence to draw inferences by evidence and logical thinking used to extract meaningful information from large sets of structured and unstructured data.
To get a conclusion AI requires reasoning. The logical inference is approached in different methods, deductive and inductive:
2.3.1 Deductive reasoning:
Is a logical process in which a conclusion is based on two or more general statements or hypotheses. It starts with general premises and theories to draw a logically true conclusion. If the existing premises are true then the Inferences will be true. In deductive logic, we draw a conclusion or observation on the basis of general premises.
Deductive inference conclusions are logical and true, even if the generalization is not true.
Alice may or may not be a doctor, but the conclusion is logically true.
2.3.2 Inductive reasoning:
Is a process of reasoning from a specific observation to broad generalization. It’s quite an inverse process to deductive reasoning. In inductive logic, existing premises only provide some validity and support to the conclusion. There is no necessity for the conclusion to be true, even if the premises are true.
In this scenario, AI will conclude it probably happened due to a weak password, but it might not be true.
As we’ve seen significant development in AI, but drawing relevant inferences to real-life situations is still a major confront.
2.4 Natural Language Understanding (to direct interaction with humans)
The actual success of AI is achieved when a machine can work with humans by blending organically into our lives. We’ve been much advanced in making intelligent machines, but still human and machine direct interaction is a major hurdle to overcome in AI. Natural language understanding is the most crucial subset of natural language processing (NLP) and AI. It enables direct human-computer interaction, by understanding commands in natural human languages without any computer programs and formalized syntax. And, also communicating back to humans in their own language.
A typically programmed computer can use language without understanding it, but NLU helps computers to better interpret what the text/speech actually means and the intention behind the text/speech.
The actual intelligence in AI is strongly related to solving problems, and it’s a process of searching through a range of possible actions in a way to achieve goals and solutions for the specific context. Problem-solving divided into two methods, special purpose and general purpose.
2.5.1 Special-purpose problem-solving methods are built with very limited features to solve problems of specific situations. These methods mostly used to achieve specific goals like playing chess and facial recognition.
2.5.2 General-purpose problem-solving methods are generally used in a wide variety of situations. Robotics is the perfect example of general-purpose methods, AI-powered robots are capable of performing multiple tasks at a time – mostly navigating, object recognition, and manipulating physical objects.
3 Classification of AI:
Artificial intelligence is classified into three types based on its capabilities, as follows:
3.1 Weak AI:
Also known as ‘Narrow AI’, is based on Artificial Narrow Intelligence (ANI) that built and trained to perform very specific purposes, so they cannot be applied to another situation. In Narrow AI, human decides on what purpose should AI be trained and what data should be given for training. So, Weak AI is as smart as the data provided at the training. Most of the AI we see today is based on Narrow AI, some of the examples are virtual assistants and video games.
3.2 Strong AI
Strong AI is based on Artificial General Intelligence (AGI), which is an AI system that represents human-like thinking and cognitive abilities. General AI can perceive and learn any intellectual task as perfect as humans, such as movements, planning, problem-solving, and conversation.
Currently, there is no such system as general intelligence. Worldwide scientists and researchers are focusing on developing machines with general AI, so we could see them in the near future.
3.3 Super AI or Artificial Super Intelligence (ASI)
Also known as ‘Artificial Superintelligence’, it is when machine intelligence surpasses human intelligence and cognitive abilities. ASI requires hardware with a speed of petaflop clock rate. If you add Quantum computing to the mix, the capabilities of AI become unimaginable. It is what most scientists and tech leaders warn about.
4. Types of Artificial Intelligence:
AI can be categorized into four types based on their functionalities, as follows:
4.1 Reactive machines.
Reactive machines are the very basic type of AI, and it doesn’t have the ability to store memories to use past experiences. It only focuses on present situations/problems and reacts according to the best possible action. IBM’s Deep Blue and Google’s AlphaGo are the perfect examples of reactive machines
IBM’s Deep Blue was the first chess-playing AI to beat the world champion, Garry Kasparov, in 1996. It won’t store any data/previous experiences to predict future moves. Reactive machines don’t rely on past experiences, it sees only what it sees and acting on what it sees and analyzes each and every possible move of its own and its opponents. So, it can find the best strategic move to win the game.
4.2 Limited memory.
As it there in its name, limited memory AI can able to store past experiences, but for a limited period of time. The best example of limited memory AI is self-driving cars. Observations from not-so-distance time periods will inform the actions happening in the future, such as signboards, traffic signals, speed, and distance of nearby car information to navigate the road.
4.3 Theory of mind.
This might sound like a psychological term, but it tells that a machine can understand people’s emotions, thoughts, and beliefs. Theory of Mind type of AI can have their own desires and intentions like us that will reflect in every decision they make and can interact socially like humans. This type of AI doesn’t exist. Researchers are making lots of efforts to developing such AI machines.
This is an extension of type III AI (theory of mind), where computers can go beyond human intelligence. In this category, conscious machines are aware of themselves and have a sense of their own consciousness, sentiments, and feelings.
As this type of AI can go beyond human intelligence, famous scientists and tech leaders are expressed their fear that AI can be a threat to humanity. Self-aware AI may write its own constitution and can take control of humanity or completely wipe out us. This type of AI doesn’t exist in reality. Though this is just a hypothetical concept till now.
5 History of AI
It all begins when Mr. John McCarthy a professor of Dartmouth College coined the word ‘Artificial Intelligence’ at the Dartmouth Conference in 1956. Since the word ‘artificial intelligence’ was introduced, it became much fascinating and enthusiastic for computer science researchers and scientists. In these recent years, AI is developing much faster than ever. Here are a few of the most interesting milestones in the AI timeline.
Year 1943: Before the word ‘artificial intelligence’ was introduced, neuroscientists Warren McCulloch and Walter Pitts worked on Artificial Neurons, which is a base concept for today’s deep learning, was inspired by neurobiology.
Year 1955: Two computer science and cognitive psychology researcher created a first-ever problem-solving program called ‘Logical Theorist’, is also known as the first-ever artificial intelligence program’.
Year 1956: The term ‘artificial intelligence’ was first time introduced by American computer science scientist Mr. John McCarthy at the Dartmouth conference addressing his ideas about machine learning and artificial neural networks (ANN).
Year 1966: The first chatbot, ELIZA was created by Joseph Weizenbaum to find solutions for mathematical problems.
Year 1972: The first AI embedded humanoid robot called WABOT-1 was built in Japan.
Year 1997: For the first time in history, AI defeats humans. IBM’s Deep Blue was the first AI to beat the human
Chess champion, Gary Kasparov.
Year 2002: For the first time AI entered a home, an autonomous vacuum cleaner called ‘Roomba’ was introduced by iRobot.
Year 2011: IBM’s AI-powered supercomputer ‘Watson’ won ‘Jeopardy’, a game of complex questions and riddles. Watson proved that AI can understand and process natural language for the first time.
Year 2012: Tech giant Google introduced a new feature called ‘Google Now’ to its Android-based Google Search app, which can show predictive results on the basis of user search and browsing habits. And, it is doing well.
Year 2016: Deep Mind’s AlphaGo has beat the world Go champion Lee Sedol. Go has over a hundred thousand possible moves, which is much more times complicated than Chess. AlphaGo used artificial neural networks to learn to play Go as it played several times.
Year 2018: Google Duplex was demonstrated by Google CEO Sundar Pichai. It was the major milestone in natural language processing in AI, Duplex can able to make calls and book appointments for their users without user interaction. Google Duplex is capable of generating natural language much like a human, along with sounds like ‘um’ and ‘ah’.
Year 2018: Samsung’s 8K AI upscaling technology turns every into 8K super high-resolution.
Now AI has got its pace and development in AI clear the barriers in most of the innovations and inventions. Major tech companies like Google, Facebook, Amazon, and IBM are constantly making their efforts to develop AI, some of them we’ve already seen and a lot more to see in the near future.