Introduction to Artificial Intelligence AI What is AI

Introduction to Artificial Intelligence (AI)

What is AI? Artificial intelligence (AI) refers to the simulation(mimicry) of human intelligence in machines that are programmed to think like humans and mimic their actions. The term may also be applied to any machine that exhibits traits associated with a human mind such as learning and problem-solving, in short cognition.

What is AI? AI is an attempt of reproduction of human reasoning and intelligent behavior by computational methods.

What is AI? AI is about thinking, perception, and action in a broad and connected manner. “Algorithms enabled by Constraints exposed by Representations that support Models targeted at thinking, perception and action. ”

What is AI? There are many reasons that the modern era should start adapting AI. Some of the reasons are stated below: ● ● ● AI automates repetitive learning and discovery through data AI adds intelligence AI adapts through progressive learning algorithms AI analyzes more and deeper data AI achieves incredible accuracy AI gets the most out of data

Who invented AI? John Mc. Carthy, who is the Father of Artificial Intelligence, was a pioneer in the fields of AI. He not only is credited to be the founder of AI, but also one who coined the term Artificial Intelligence. In 1955, John Mc. Carthy coined the term Artificial Intelligence, which he proposed in the famous Dartmouth conference in 1956. This conference attended by 10 -computer scientists, saw Mc. Carthy explore ways in which machines can learn and reason like humans.

When does AI started? ● 1943 - Mc. Culloch & Pitts: Boolean circuit model of brain ● 1950 - Turing's "Computing Machinery and Intelligence" ● 1956 - Dartmouth meeting: "Artificial Intelligence" adopted ● 1952 - 69 Look, Ma, no hands! ● 1950 s- Early AI programs, including Samuel's checkers program, Newell & Simon's Logic Theorist, Gelernter's Geometry Engine ● 1965 - Robinson's complete algorithm for logical reasoning ● 1966— 73 - AI discovers computational complexity Neural network research almost disappears ● 1969— 79 - Early development of knowledge-based systems ● 1980 - AI becomes an industry ● 1986 - Neural networks return to popularity ● 1987 - AI becomes a science ● 1995 - The emergence of intelligent age nts

Turing Test Turing (1950) "Computing machinery and intelligence":

Turing Test Can machines think? "Can machines behave intelligently? " Operational test for intelligent behavior: The Imitation Game

How is AI different from other ways of solving a problem? Artificial Intelligence is beneficial for solving complex problems due to its efficient methods of solving. AI uses different algorithms to complete its tasks which are not used in other problem solving methods. Following are some of the standard problem-solving techniques used in AI: Heuristics: The heuristic method helps comprehend a problem and devises a solution based purely on experiments and trial and error methods. However, these heuristics do not often provide the best optimal solution to a specific problem Searching Algorithms: Informed Search and Uninformed Search

Evolutionary Computation : This problem-solving method utilizes the well-known evolution concept. The theory of evolution works on the principle of survival of the fittest. It states that the organism which can cope well with their environment in a challenging or changing environment and reproduce, their future generations gradually inherit the coping mechanism, generating the diversity in new child organisms Genetic Algorithms: The evolution theory is the basis of genetic algorithms. These algorithms use the direct random search method. The developers commonly use genetic algorithms to generate a high-level solution to optimization and search problems by relying on bio-inspired operations such as mutation, crossover, and selection.

How is AI going to affect our lives? Is AI really needed in human society? It depends. If human opts for a faster and effective way to complete their work and to work constantly without taking a break, yes, it is. Artificial Intelligence (AI) truly a revolutionary feat of computer science, set to become a core component of all modern software over the coming years and decades. This presents a threat but also an opportunity.

There are some great examples of how AI has helped in the daily activities : ● Cybersecurity - AI systems can help recognise and fight cyberattacks and other cyber threats based on the continuous input of data, recognising patterns and backtracking the attacks. ● AI against Covid-19 - AI was used for thermography in airports and elsewhere in the case of Covid 19. In medicine, computerized tomography scans can help to detect infections of the lung. It is also used to provide information on the disease's spread. ● Web search - Search engines learn from the vast input of data, provided by their users to provide relevant search results.

How does AI “Think”? They perceive their environments, but they are not aware of them. Computers are furnished with memory, just like conscious beings are, and modern AI systems can anticipate or predict based on informational input.

Artificial Intelligence VS Robot AI - Programmed to think Social Interaction Learns The BRAIN Robot - Programmed to do Low level interaction Only as smart as program The BODY

Why do we need AI? • Taking over dangerous and repetitive job • Improvement in our daily life • Less time taken for services at any level of job • Every task is done efficiently • Reduce risk

What is intelligence? Intelligence can be defined in many ways: ❏ ❏ ❏ ❏ Logic Understanding Problem-solving Learning Critical thinking Reasoning Planning Creativity

Triarchic theory of intelligence by Robert Sternberg

- Practical Intelligence Being practical means you find solutions that work in your everyday life by applying knowledge based on your experiences. This type of intelligence appears to be separate from traditional understanding of IQ. Comprises the mental processes through which intelligence is expressed. - Analytical Intelligence Closely aligned with academic problem solving and computations. Analytical intelligence is demonstrated by an ability to analyze, evaluate, judge, compare, and contrast. It is bound in a sociocultural milieu and involves adaptation to, selection of, and shaping of the environment to maximize fit in the context. - Creative Intelligence Marked by inventing or imagining a solution to a problem or situation. Creativity in this realm can include finding a novel solution to an unexpected problem or producing a beautiful work of art or a well-developed short story.

Human intelligence ➢ The intellectual power of humans. ➢ It gives humans the cognitive abilities to learn, form concepts, understand, and reason, including the capacities to recognize patterns, innovate, plan, solve problems, and employ language to communicate. ➢ Enables humans to experience and think. ➢ According to the psychologist, there are four bounds of human intelligence : - Intelligence Quotient (IQ) - Emotional Quotient (EQ) - Spiritual Quotient (SQ) - Physical Quotient (PQ)

Symbolic thinking Definition: We use symbolic thinking to understand the world around us, we use mathematical expressions, linguistics and symbols to demonstrate logic and explore the world around us, this has been the reason behind how our intelligence works, we use symbolism to create representations of real life models to solve problems or overcome challenges or to communicate, since the dawn of human race to now.

Symbolic thinking In AI: A symbolic AI system can be realized as a microworld, for example blocks world. The microworld represents the real world in the computer memory. It is described with lists containing symbols, and the intelligent agent uses operators to bring the system into a new state.

How do AIs learn? Artificial Intelligence (AI) already helps to solve complex problems in sectors as varied as medical research and online shopping. Yet, AI has limitations. In unpredictable environments, autonomous agents are dependent on human feedback to determine what is interesting and what is not, and they lack the ability to self-adapt and self-modify. Now, in a paper published in Neural Networks, a team from the University of Southampton in the UK have developed a generic architecture that allows AI to create its own learning strategies and to adapt to changing situations.

How do AIs learn?

How do AIs learn? Matter of fact, though AI have strong and powerful algorithm, still cannot create art works like human, cannot create wonderful music like human. There is a lot of things AI cannot do but human can. What is more, if one day AI can learn by themselves easily just like human or even more powerful, it may cause dangers.

Machine learning is an application of artificial intelligence (AI) that provides systems the ability to automatically learn and improve from experience without being explicitly programmed. Machine learning focuses on the development of computer programs that can access data and use it to learn for themselves. The process of learning begins with observations or data, such as examples, direct experience, or instruction, in order to look for patterns in data and make better decisions in the future based on the examples that we provide. The primary aim is to allow the computers learn automatically without human intervention or assistance and adjust actions accordingly.

Machine learning Methods ● ● Supervised machine learning algorithms can apply what has been learned in the past to new data using labeled examples to predict future events. Starting from the analysis of a known training dataset, the learning algorithm produces an inferred function to make predictions about the output values. In contrast, unsupervised machine learning algorithms are used when the information used to train is neither classified nor labeled. Unsupervised learning studies how systems can infer a function to describe a hidden structure from unlabeled data. Semi-supervised machine learning algorithms fall somewhere in between supervised and unsupervised learning, since they use both labeled and unlabeled data for training – typically a small amount of labeled data and a large amount of unlabeled data. Reinforcement machine learning algorithms is a learning method that interacts with its environment by producing actions and discovers errors or rewards. Trial and error search and delayed reward are the most relevant characteristics of reinforcement learning.

Is machine learning important? Why Is Machine Learning Important? Data is the lifeblood of all business. Machine learning can be the key to unlocking the value of corporate and customer data and enacting decisions that keep a company ahead of the competition.

Difference between human learning and machine learning Human learning is what we also call the one shot learning, meaning we learn something definitive from every encounter or interaction, our learning is based on explanations, unlike machines which we need to train(using what you just learned about in machine learning methods) to process great masses of data and have powerful computers process them, their learning is what we call bulldozer computing using neural nets, boosting and NN(Nearest Neighbor) learning.

Learning Tree Human learning(constraints ) One shot learning Explanation-based learning Bulldozer computing (regularity) NN learning Boosting Neural Nets

Illustration of the Learning Tree We have two kinds, learning based on observation and regularity(which computers are peculiarly good at) which we will first jump into the regularity branch we has three sub branches, the nearest neighbors learning(pattern recognition) , the neural nets(used in machine learning; mimics biology) and the boosting. Nearest neighbor learning Neural nets Boosting

Illustration of the Learning Tree Nearest Neighbor learning : Imagine you wish to classify an unknown variable inside a pre-determined, detailed environment (illustrated either through a x-y graph or a table) and you wish to specify the unknown variable with the most accuracy for optimal functionality, we use NN to solve this issue. Basically the system creates a representation of the original space as the reference and then uses this reference to identify the newly added, unknown variable. You can use this both as a classifier and a trainer, a classifier? Sure but why a training algorithm? Great question. Imagine you were to program a mechanical arm to be able to perfectly bounce a tennis ball for nearly infinite times, so lets try doing that!

Illustration of the Learning Tree Some may say well we can represent the mechanical arm as a 2 -d model of arms and joints, which is correct, and you might continue on to say that well you can also use newtonian physics to program the arm since it has joints meaning it has angles of rotation and movement and you will have to go through a considerably large amount of equation solving and when you’re done you see it doesn’t work, now why is that? It’s because there are multiple factors to the issue that haven’t been accounted for in the equation like torque and other angles, and thus you need to create a detailed table of references to fill it, but do you really have to? No. The arm will try and fail and each time it will fill more of the table, basically learning and going through its “childhood”. Why does it get better every time? Because when he comes back to a motion he has already done before, he will be better at it and this will continuously go on.

Illustration of the Learning Tree Some notes to consider when using this algorithm 1. Space spread : We can use our high school statistics variance to normalize the data spread to solve the issue. Why? Because the data spread needs to be the same in all the dimensions. 2. Trying to bake a cake without flour : When we try to predict an outcome whilst the data available is the data which is independent of the question.

Illustration of the Learning Tree Neural nets : Neural networks or nets mimic the function of neurons. It is the piece of a computing system designed to simulate the way the human brain analyzes and processes information. It is the foundation of AI and it attempts to solve problems that would prove impossible or difficult by human or statistical standards. We use Neural nets to to classify objects and test inputs, like Jeff Hinton’s neural network which had a model of 60 million parameters and could categorize an image from 1000 categories, like animals, vehicles, objects and so on. Neural networks are one of the fundamental but recently modified branches of AI as even back in 2010 there wasn’t really a decent model of brain neural network that represented the mentioned appropriately.

Illustration of the Learning Tree Boosting : In simple words, boosting is the collective concept of algorithms which we use weak classifiers to create a strong classifier. We define a strong classifier as one that has a near zero error and a weak classifier has an error rate of below 0. 5, a bit better than a coin flip. One of the main and early algorithms of boosting was created by Freund and Schapire which they named adaboost as an acronym for adaptive boosting which uses weak classifiers assigned with weights and multipliers in a step by step algorithm until all samples are correctly classified OR there are no weak classifiers left, this has a fun but intermediate level of calculus for proving the algorithm in the mathematical aspect as well.

What is Logic is the systematic study of valid rules and principles of inference, for instance the relations that lead to the acceptance of one proposition on the basis of a set of other propositions. More broadly, logic is the analysis and appraisal of arguments. There is no universal agreement as to the exact definition or boundaries of logic. We use logic in everything we do to achieve a cognitive optimality or in simpler words, rationality, meaning we function and act based on the templates we have to reach a certain goal. We call this approach the logical approach.

Use of mathematics in AI A great amount of AI is founded on the appropriate use of mathematics to implement logic, we use matrices, multivariate calculus, linear algebra, probabilities, functions, vectors etc as mathematics is an essential requisite towards the path of understanding AI in depth and more importantly to be able to utilize its functionalities for certain purposes. As said before some examples are the use of algebra and matrices in neural nets and image processing, the use of graphs in classifiers, the use of vectors in NN and use of mathematical equation design in basically all aspects of solving a question or problem through AI hence why it’s so crucial.

Reasoning : goal trees and problem solving In AI, we use logic and interpretation of multiple definitions and occurrences to form concepts thus allowing AIs to understand how we think and talk, how we learn and how we see the world around us. This part is a more practical, in-depth detailed demonstration of how to mimic such sort of understanding and intelligence through mathematics and programming.

Reasoning : goal trees and problem solving *The Integration calculator program of James Slagle which was a very dawn of the time yet pretty smart program is basically a one-lecture course of Artificial intelligence. Imagine an AI is being tasked to solve every integration problem given to it, the approach is to have a program that uses safe and heuristic transformations to turn the question to the simplest form. The AI will have a table of transformations under two categories, safe and heuristic. After that we input the integration problem into the program and it starts performing ordered transformations until it reaches the goal. The way this transitions will be represented is via a goal tree, also known as a problem reduction tree also known as a and/or tree because it had AND and OR operators which perform the function on the nodes.

Reasoning : goal trees and problem solving First we input the problem, AI turns this into the root node and starts the process of simplifying it. We call this process the problem reduction. We will also have a list of safe transformations, these are transformations that are valid for all situations, and always work. We also have a second category of transformations which are often helpful but don’t necessarily work. We call this the heuristic transformation. After forming the tables we start the procedure of simplifying the problem.

List of safe transformations 1. Minus of integral of f(x) is equal to the integral of minus f(x) 1. Taking constants out 1. Sum of integral is equal to integral of sum 1. Integral of p(x)/f(x) → DIVIDE (divide the polynomials if the degree of the numerator is greater than the degree of the denominator to simplify)

List of heuristic transformations 1. Transformation from trigonometric form to polynomial 1. Transformation from polynomial to trigonometric using (We have two transformations in this kind, one when 1 -x^2 which we use x=sin(y) and one when 1+x^2 which we use x=tan(y)

Reasoning : goal trees and rule based expert systems Imagine an AI is tasked with understanding the semantic of a passage of text, like a book(like Macbeth) and we wish to program it so it can interpret the text OR we want the AI to classify an unknown variable based on given data, extract statements and concepts and be able to answer questions about itself and its choices, let’s call this the Selfaware program. The AI will break down the passage and form a tree graph of nodes under templates, and link the nodes based on the actions taken by the passage’s characters, like how Macbeth murders Duncan. It sounds challenging right? In fact it’s quite simple, so much that you can do it with what you have learned from Part 1 of our course! After the tree is formed it will traverse inside the graph and find the predetermined concepts from the passage thus understanding the text, but how will it be able to answer the questions about its actions? Simple, we use a formation called as rule based expert system, which is how we use AI to interpret concepts based on given data and how it classifies the result, like giving characteristics of an animal and having the AI say what that animal is, and now we wish to ask the program about why it has given such an answer.

Reasoning : goal trees and rule based expert systems So the AI has been given some characteristics, first it will turn these inputs into nodes, and then it will start forming the graph and move towards the goal. Forward-chaining Strong Stripes Orange claws AND operator Tiger Sharp teeth Forward-facing eyes carnivore Backwardchaining To get to the goal and the answer of what the animal is we use forward chaining rule based expert systems and to get the answer to why the answer is what it is we use backward-chaining rule based expert systems. To answer the questions with why we use Backward-chaining and to answer questions with how we use forward-chaining.

Illustration of the Learning Tree The constraint branch, also known as the human learning is splitted into two branches of learning, the one shot learning (which means we learn something definitive from every encounter or interaction; important to have a good understanding of this) and the explanation based which means our intelligence is also derived from our ability to tell stories, and this is something extremely important. How you may ask? Everything we do in solving a problem or teaching or communicating is done by telling a story, it’s done in all course problems, especially mathematics, physics and chemistry, the teachers basically tell a story which you learn something definitive from (One shot learning) and understand it through the flow of the story(thus why some teachers are great and some not so much).

Symbolic learning completion So we discussed how our way of thinking is highly dependant on using symbolism as our way of thinking is based on symbolic thinking, and as discussed in the first session, we said Java also inherits this way of thinking from us humans, it creates representations of real life models to solve problems, same as all other sciences, blueprints of machinery, anatomy, and atomic models, we use model representation in every attempt of doing a problem or simply exploring the uncharted through cognition. We also said that we learn through story-telling as it’s how we understand a model and learn from it and hence comes the completion of symbolic thinking : “Humans use story-telling to illustrate a problem that is solved through creating a representation of the problem’s model through symbolic thinking. ”

AI is human too We may thought AI is omniscience, but truth is AI is just like us, the human being. AI need to learn before they know, and improve itself through practice. From time to time, AI learns mistakes and avoid making them again, thus, AI looks like omniscience.

AI learning from mistakes Chess AI Alpha. Go has defeated many top Go players, but hasn’t always ended with 100% victory. The game between Alpha. Go and South korean professional player Lee Sedol ended 4: 1. Alpha. Go won the first three games, and Lee beat Alpha. Go in fourth game. However, Lee didn’t defeat Alpha. Go in fifth game.

How can AI swept the human top players? Scientists are trying to apply what we know about the way humans learn to machine learning. Thus, AI has developed an algorithm which makes it possible for AI to learn from its own mistakes in the same way babies do it. Their new technology, called Hindsight Experience Replay (or simply HER), allows AI to review its previous actions when completing a specific task, and this makes AI become more and more perfect.

AI understanding text A simple project which has been demonstrated in some of the highest ranking AI intro university courses is the AI understanding the Macbeth story, same as a human, as it analyzes the text, segments it and turns it into a connected graph. How does it “Understand” the text? It uses interpretation, meaning it will use classifiers as illustrated before to understand emotions and thoughts, imagine this : How does AI understand Macbeth murdered Duncan ? Well it will first implement the term “murder” with “kill” and it will use the database and definitions we have added for it to give the following statements based on defined concepts : Macbeth Kill Duncan (sequence or action that is happening) Duncan Die (property of Duncan is being dead) Macbeth motive (the reason behind Macbeth killing Duncan, which the AI uses the previous nodes of graph and connections to understand, here because he wanted to become king) consequence

Graph(semantic net) of Macbeth story Macbeth Reifications : They are the arrows that link two other arrows or links. (red arrows) Duncan Murder Kill Template Agent : Macbeth Victim : Duncan Action : kill Combinators : They are the arrows that link the nodes. (black arrows) Property of being dead Localization : The layer of template that we have added on top of the semantic net.

AI playing games What is AI in gaming? AI in gaming refers to responsive and adaptive video game experiences. These AI-powered interactive experiences are usually generated via non-player characters, or NPCs, that act intelligently or creatively, as if controlled by a human gameplayer. AI is the engine that determines an NPC's behavior in the game world.

Games have been used for decades as an important way to test and evaluate the performance of artificial intelligence systems. Real-time strategy games to AI is a great challenge, for example Starcraft. Alpha. Star is an AI player form Starcraft plays with deep neural network and it defeated the best human player 5: 0. And the Go AI - Alpha. Go Zero, the next vision of Alpha. Go. It develop itself by playing with itself, and only 3 days’ self training, Alpha. Go Zero beat its predecessor Alpha. Go(second vision) in 100% win rate, and after 40 days learning, the final vision Alpha. Go - “Master” lose its no. 1 title.

AI in games not just against the player, but also provide player an immersive experience. Red Dead Redemption 2 for example. The game provides a living western world that NPCs in game are not just decorations, player can interact with them, and NPCS will respond to what happened in the environment. Like player don’t take shower for a long time, NPCs will stay away from player, and if player keep following a NPC, it would act like a normal people, be scared and warn the player to stop. NPCs in game also have their schedule, they are not always wandering on the road, they have things to finish.

Summary of session ● Artificial Intelligence refers to the simulation of human intelligence in machines that are programmed to think like humans and mimic their actions. And it is about thinking, perception, and action in a broad and connected manner. ● AI improve the life quality and raise the productivity in our daily life. They uses different algorithms to complete its tasks efficiently and learning to enhance itself to do better. ● Intelligence can be divided into 3 types, Practical/Analytical/Creative. ● four bounds of human intelligence : Intelligence Quotient (IQ)/Emotional Quotient (EQ)/Spiritual Quotient (SQ)/Physical Quotient (PQ) ● A symbolic AI system can be realized as a microworld, for example blocks world. The microworld represents the real world in the computer memory. ● Machine learning is an application of AI that provides systems the ability to automatically learn and improve from experience without being explicitly programmed. Machine learning focuses on the development of computer programs that can access data and use it to learn for themselves.

● Human learning is the one shot learning that learn something definitive from every encounter or interaction, our learning is based on explanations, unlike machines which we need to train to process great masses of data and have powerful computers process them, their learning is what we call bulldozer computing using neural nets, boosting and NN(Nearest Neighbor) learning. ● Logic is the systematic study of valid rules and principles of inference. More broadly, logic is the analysis and appraisal of arguments. ● In AI, we use logic and interpretation of multiple definitions and occurrences to form concepts thus allowing AIs to understand how we think and talk, how we learn and how we see the world around us. ● AI can make mistakes just like human being, and AI learn from mistakes to make itself become more perfect by avoiding make them again. ● AI can understand text by dividing sentences into several parts and analyzing them. ● Game is an important way to test and evaluate the performance of artificial intelligence systems.

Why is AI a growing science? Artificial Intelligence seems to be on the tip of everyone’s tongue these days. It becoming wide-used that even appearing as one of the most indemand areas of experience for job seekers. Data shows that the use of AI in many sectors of business has grown by 270% over the last four years. And people are increasingly relying on electronic devices that promote the development of AI, also the higher productivity and less time cost. As human invent car to replace walk, they will invent AI to replace driver.