top of page
Ali Hasany

ARTIFICIAL INTELLIGENCE - All you need to know in this One Super Guide

INTRODUCTION TO ARTIFICIAL INTELLIGENCE


A robot standing with AGI Computer in 3D voxel Art.
Artificial Intelligence

What is Artificial Intelligence

Humans have different types of intelligence, like being good at languages or art, but not necessarily both.


Computers are similar.


They excel in specific tasks, where rules and patterns are clear.


For instance, computers have outperformed humans in games like chess and Go, but they don't understand these games; they just follow rules and match patterns well.


This makes it tricky to define what makes a computer truly intelligent.


AI works best in areas with set rules, like games or online shopping. When thinking about using AI, focus on tasks where pattern recognition and following rules are important.


General Problem Solver

In 1956, Allen Newell and Herbert A. Simon developed a program called the general problem solver, based on the idea that computers could be smart by understanding symbols like stop signs or letters.


But not everyone agreed.


Philosopher John Searle in 1980 argued that this doesn't mean true intelligence. He used the "Chinese room" example, where a person responds to Chinese symbols using a guidebook without understanding the language. It's like asking Siri how it feels – it responds but doesn't understand.


This symbol-matching approach was key in AI for a while, but it was too complex and couldn't lead to real intelligence.


Strong vs Weak AI

When talking about smart computers, we can think about them in two ways, like the philosopher John Searle said. There's "strong AI" which is like robots in movies who have feelings and can think creatively.


This kind of AI is more like science fiction right now.


Then there's "weak AI", which is like Siri on your phone. It does specific tasks, like turning your words into text or sorting pictures.


It listens to you and uses its database to answer, but it's not really thinking or understanding like a person.


In the past, there were expert systems in AI. These were like smart lists made by experts to solve problems.


For example, in medicine, a computer might help diagnose a cough by following a list of symptoms and solutions.


But it was just matching patterns, not really thinking. These systems had limits because there were too many combinations to handle.


But they were important in the early days of AI and are still used today, sometimes called "Good old-fashioned AI."


3 APPROACHES OF ARTIFICIAL INTELLIGENCE


A- Machine Learning

Imagine a computer that learns on its own, just like you learn from watching the world.


Early AI recognized symbols to seem smart, but this method was limited.


Then, in 1959, Arthur Samuel created a checkers program that improved by playing itself, learning from patterns, not from direct programming.


This was the start of machine learning.


As the internet grew in the 1990s, so did the data for these computers to learn from.


More data meant better learning.


Today, machine learning is a big part of AI, constantly growing as it finds patterns in vast amounts of information, helping us understand and use this data effectively.


B-Deep Learning / ANN (Artificial Neural Network)

Machine learning has grown a lot thanks to artificial neural networks, which are just like the human brain.


Imagine a game where you guess if something is an animal, vegetable, or mineral by asking questions.


Artificial neural networks do something similar, but instead of questions, they use many number dials to make guesses, like saying there's a 64% chance it's a cat.


They have layers of 'neurons':

  • input,

  • hidden, and

  • output.


The network sees a picture, like of a dog, and tries to guess what it is. It learns by adjusting these dials based on many pictures. It doesn't see dogs like we do, just as patterns of dots.


These networks need lots of data, like thousands of dog pictures, to learn and recognize patterns.


This self-training ability is what makes artificial neural networks so useful in machine learning.


Three types of Neural Network


1. CNN (Convolutional Nural Network)

CNNs are a type of ANN especially good at handling images.


Imagine you have a puzzle. CNNs look at small pieces of the puzzle (like edges, colors, and shapes) and then put these pieces together to understand the whole picture.


They're really good at tasks like recognizing faces in photos or identifying objects in images.


2. RNN (Recurrent Neural Network)

RNNs are another kind of ANN.


They are like a memory wizard for computers.


They remember what they've seen before and use that memory to make decisions.


This is super useful for things that have a sequence, like understanding spoken language (like Siri or Alexa),or predicting the next word in a sentence.


3. GAN (Generative Adversarial Network)

Imagine two artists in a competition: one creates paintings, and the other judges them.


The first artist is the "generator", making new images or data.


The second is the "discriminator", trying to tell if the images are real or fake.


They both improve through this competition.


GANs are used to create very realistic images or videos that didn't exist before, like making new fashion designs or generating new video game levels.


C-Expert System (oldest and most common)

All those systems which are based on symbolic logic, rules engines, expert systems and knowledge graphs. are called expert systems. These systems are out of scope of this article.


HOW ARTIFICIAL INTELLIGENCE WORKS


APPROACH NUMBER ONE - MACHINE LEARNING


A) Learn from Data


1) Labeled vs Unlabeled Data

Machine learning is like teaching a computer to learn.


Think about learning chess.


You could get a chess tutor who teaches you the pieces and how to move them.


Or, you could go to a park, watch others play, and learn by observing.


These methods are like how machines learn.


1. Supervised learning

Supervised learning is a type of machine learning where an algorithm is trained on a labelled dataset by an expert teacher or human.


This means that the data used to teach the model includes both the input data and the correct output.


The goal of supervised learning is for the model to learn a mapping between inputs and outputs, so that when it is given new, unseen input data, it can predict the corresponding output.


Examples are:

  • Image Classification,

  • Spam Detection in Emails

  • Credit Scoring:

  • Medical Diagnosis:


2. Unsupervised learning

Here the machine observes data by itself.


Unsupervised learning is a type of machine learning where the algorithm is trained on a dataset without any labeled responses or outcomes.


The goal is to explore the structure and patterns within the data, often to discover hidden features or groupings.

Examples are:

  • Clustering,

  • Anomaly Detection

  • Association Rule (Market Basket analysis)

  • Natual Language Processing


2) Massive Datasets

If you've used a computer or phone, you've seen how apps work because someone programmed them.


But with artificial intelligence (AI), it's different.


AI can't just follow step-by-step instructions for everything because there are too many possibilities.


So, AI learns from examples instead. Think about a program that finds spam emails. You could tell it to look for words like "lottery" or "winner" and mark those emails as spam.


But for more complicated tasks, AI learns by finding patterns in data.


For example, you give AI lots of emails, some spam and some not. It learns what spam looks like. Then, you test it on even more emails to see if it's good at spotting spam.


AI uses special math (algorithms) to learn and make decisions, like whether an email is spam or not.


This is called machine learning, and it's really useful for lots of things, not just emails. The main point is, AI uses patterns in data to learn and make choices.


B) Identify Patterns


1) Classify Data (Classification)

We often sort things, like documents or contacts.


Businesses do this too, like airlines categorizing frequent flyers.


In machine learning, there's a method called binary classification, which sorts things into two categories, such as spam or not spam emails.


This uses supervised learning, where the system learns from examples, like emails labeled as spam.


However, teaching the machine can be tough.


For instance, a credit card company might need thousands of fraud examples to train it accurately.


This process helps in various areas like detecting fraud or spam, essentially teaching the machine to categorize things into predefined groups.


2) Cluster Data (Clustering)

Sometimes, machines in technology don't just sort things into groups we already know, like when you put your favorite candies in one pile.


Instead, they find their own groups, like when you get a surprise bag of candy and have to figure out how to sort it without knowing what's inside.


This is called clustering, and it's like when a website suggests things to buy together, like a mouse and a keyboard.


The machine looks at what people buy and makes its own groups.


Big companies like Amazon use this to show you things you might like based on what you and others buy.


It's a smart way for machines to help us find what we might like or need.


3) Reinforcement Learning (RL)

Online music, like on Spotify or Apple Music, is big business.


What sets these services apart is their song recommendation systems.


They use a type of machine learning called Reinforcement Learning.


Here's how it works: When you play a song they suggest, the system gets a digital "reward." The more you listen, the more rewards it earns. This teaches the service what music you like.


It's like the system has a bank account, and every time you enjoy a song, it gets richer.


This way, it gets better at finding new songs you might love, helping you discover music in a creative way.


C) AI Algorithms


1. Regression (Classification)

Regression is like trying to draw the best straight line through a set of points on a graph.


It's used to understand the relationship between things.


For example, it can help predict a student's final exam score based on how much they studied.


The line shows the trend: usually, the more you study, the better you score.


2. K-nearest Neighbour (Classification)

Imagine you have a bunch of different fruits mixed together and you want to sort them. K-nearest neighbor is a method where you take one fruit and find the 'k' closest fruits to it (like the 3 nearest).


If those 3 fruits are mostly apples, you guess the fruit in question is also an apple.


It's a simple way for computers to classify things based on similarity to other things.


3. Naive Bayes (Classification)

This is a method for making educated guesses.


It's based on probability - like guessing the likelihood of something happening based on what has happened before.


Imagine you're trying to figure out if it will rain today.


Naive Bayes would look at things like how often it rains when there are clouds, and use this information to make a guess.


It's often used in spam filters to predict if an email is spam or not.


4. K-means clustering (Clustering)

This is a way to group similar things together.


Let’s say you have a bunch of points on a graph. K-means clustering finds the center points (called centroids) of 'k' groups. Each point is then assigned to the nearest centroid.


This helps in organizing data into groups that share similar features, like separating different types of customers based on their shopping habits.


APPROACH NUMBER TWO - ARTIFICIAL NEURAL NETWORKS


1) Build a neural network

Artificial neural networks (ANNs) are a type of machine learning that processes large data sets like the human brain.


Instead of traditional machine learning algorithms, ANNs break data into smaller pieces.


They consist of neurons arranged in layers: input, hidden, and output layers.


In deep learning ANNs, there are many hidden layers, helping to identify complex patterns.


For example, to determine if an image contains a dog, the image is broken into pixels, each fed into the input layer.


Neurons in each layer decide whether to pass data to the next layer.


In the end, the output layer gives a probability score for the binary choice (dog or not dog).


ANNs, often used in supervised learning, tune themselves to improve accuracy.


2) Weighing the connections

In simple terms, when we see blurry objects in different scenes, like a grassy field or a desert, we guess what they might be based on where they are.


This is like adding 'weights' to our thoughts.


Artificial neural networks in computers do something similar.


They adjust 'weights' between neurons to make better guesses.


For example, if there are 100 neurons, each one connects to 100 others, making lots of connections.


Each connection has a weight, like W1, W2, up to W100.


These networks learn from data, starting with random weights and adjusting them to make more accurate predictions.


Just like tuning a guitar, they get better over time at guessing correctly.


3) Activation Bias

An artificial neural network is like a musical instrument that tunes itself to match a perfect note.


It's a type of machine learning that adjusts its own settings to make accurate predictions.


The network does this by adding weights to connections and bias to neurons.


This process is like trying to throw darts tightly around a bullseye.


Adding weights corrects variance (how spread out the darts are), while adding bias adjusts the direction they're thrown.


This balancing act is known as the bias-variance trade-off.


However, neural networks often overfit data, making it hard to find the right balance between bias and variance, like driving on an icy road with risks of sliding too far in one direction.


Bias is assigned to neurons, not connections, helping the network adjust after seeing the effects of variance changes.


4) Improve Accuracy

4.1 Learning from mistakes

We have learned how artificial neural networks use mistakes to improve.


Unlike humans who often see things as right or wrong, neural networks measure how correct or incorrect they are using a cost function.


This function assigns a number to indicate how far off the network's answer is from the correct one.


If the network's prediction is very wrong, like mistaking a snow-covered mountain for a dog, it faces a higher cost.

Neural networks adjust themselves through a process called gradient descent, which helps them make smaller or larger changes based on how far off their predictions are.


They constantly adjust their weights and biases to get closer to the correct answer, a method known as backpropagation of errors.


This back-and-forth process helps the network learn from its mistakes and improve its accuracy.


4.2 Step through the network

To build an AI system, data scientists first decide what they want from the data.


For example, in a project to identify dogs in images, they use a binary classification system: dog or not dog.


This is part of supervised machine learning, where the AI is trained with labeled images.


The scientists might use an artificial neural network, which has layers and nodes for processing.


They start by setting random weights and zero biases, and then train the network with labeled images.


The network makes guesses, adjusts its weights and biases to improve accuracy, and is tested with unlabeled images.


If it performs well on training data but not on new test data, it might be overfitting. This method of binary classification can be applied to different types of data for insights.


USES OF AI

Using AI

Artificial Intelligence (AI) began with the general problem solver, using symbols for programmed responses.

Soon after, machine learning emerged, where systems learn from data patterns instead of being explicitly programmed.


Today, AI is increasingly used in workplaces for complex and creative problem-solving.


Understanding AI is crucial, especially for business people and entrepreneurs who will likely work on AI projects.


Managing AI effectively involves ensuring accurate, real-world data, patience during its trial-and-error learning, and providing supervision.


AI can process vast data and uncover patterns beyond human capability. The future of AI will augment rather than replace human creativity.


COMMON AI SYSTEM EXAMPLES AND BRANDS

1. Searching for Patterns in Data

In the last 30 years, machine learning, a part of artificial intelligence (AI), has grown a lot. It helps computers learn from lots of data, like photos, to recognize patterns.


For example, by seeing many dog pictures, a computer can learn to identify dogs. This learning is easier now because we can easily get lots of digital information.


Websites use this to understand what we like by watching our video habits. Big companies like Google use AI to improve their products.


But sometimes, we don't fully understand how AI makes decisions, especially in important areas like healthcare. AI is smart, but different from human thinking!


2. Robotics

Robotics, a part of AI, lets machines do real-world tasks.


This includes everything from factory work to driving cars. Earlier, robots did specific jobs like welding, but now, they can learn and adapt thanks to machine learning.


For example, self-driving cars use sensors and AI to understand roads and react to things like animals and people.


They're still learning, so humans often supervise them.


However, simpler robots like Roombas use basic AI to avoid obstacles.


For tasks with big safety risks, like dispensing medication, robots are programmed more cautiously. This way, they safely and effectively assist in our physical world.


3. Natural Language Processing (NLP)

We're teaching machines to communicate like us.


Machines share information accurately and quickly, but humans often struggle to convey their thoughts fully.


This is where natural language processing (NLP) comes in.


It helps machines understand and respond to us in our own language.


For example, when you ask a smart device for a waffle recipe or the meaning of love, NLP helps it understand and find the right answers.


It learns from lots of data to grasp words' meanings and contexts.


NLP is making machines more human-like in their communication, allowing them to understand and interact with us better.


1.Text (ChatGPT, Clause, Gemini, Bart, Bing)

These are computer programs that understand and generate human-like text.


They can chat, answer questions, write stories, and more.


They learn from reading a huge amount of text, which helps them understand language patterns and how to respond in a conversation.


Think of them as really smart virtual assistants that can talk about almost anything.


2.Image & Video GPT (DALLE-3, FireFly, Stable Diffusion and Midjourney)

These are like the GPTs for text, but they work with images and videos.


They can create new, realistic images or even videos from a description.


For example, if you ask for a picture of a cat riding a skateboard, these programs can generate an image that matches that description.


They're used in art, design, and entertainment to create visuals that might be difficult or impossible to make in the real world.


4. Internet of Things

The Internet of Things (IoT) includes devices like smartwatches, thermostats, and doorbells that connect and share data online.


These gadgets collect information about our location, activities, and even health.


For example, a smartwatch could inform your thermostat when you're heading home. IoT devices communicate with each other to automate tasks, like unlocking doors or turning on computers.


This generates a lot of data, which AI analyses to predict our behaviour and needs.


In healthcare, IoT devices like smartwatches provide health data, which AI uses to identify health trends.


IoT bridges our digital and physical worlds, helping companies tailor products to our lifestyles.


CONCLUSION

In this blog about artificial intelligence (AI), you learned about its technology and history, starting from basic concepts like problem-solving and symbolic reasoning to advanced topics in machine learning.


You explored how machine learning evolved, using algorithms to analyze huge data sets and artificial neural networks to find complex patterns.


The blog also taught about back propagation, a method networks use to learn from errors and improve.


Additionally, it highlighted the importance of considering AI's ethical aspects, like decision-making in health and finance.


There's also a suggestion to learn more about data ethics.

18 views0 comments
bottom of page