Introduction
Remember when we used to day dream about flying cars, pens that’ll do our homework without us even stirring a finger or robots that’ll clean our rooms for us? Well, I think we are not too far from making those ideas a reality. Because from autonomous vehicles revolutionising transportation to AI diagnostics and more, artificial intelligence is reshaping the way we look and function throughout our lives.
I’m not going to lie I myself have become slightly more dependent on AI despite promising myself I’d never try Chat GPT.
And as the world goes wild and rightfully so, with this new technological advancement that has the potential to rewrite how the world works, no one really talks about how AI works and its ethical implications.
To me it just seems to exist.
Until a few weeks ago when I started researching it.
So, in this two-part blog I am taking this as an opportunity to research and educate myself and also share what I have learned so far.
Understanding AI
IBM.com describes AI as “technology that enables computers and machines to simulate human learning, comprehension, problem solving, decision making, creativity and autonomy.”
So basically, machines that are programmed to learn and make decisions like humans but a lot faster.
And there are two parent categories of AI that we need to focus on:
Narrow AI
Note: Narrow Artificial Intelligence is alternatively known as “weak AI”.
So, what exactly does any of those words mean?
Narrow AI refers to an AI design that performs specific task or a set of closely related tasks. In other words, it specializes in a specific field. It operates under predefined parameters and does not possess general intelligence or self-awareness.
And not so shockingly, a version of this design is being used all around us. From auto-generated Netflix recommendations, to virtual assistants like Siri or Alexa (which by the way are more annoying than useful) and conveniently spam filters in email systems; we use them all the time.
General AI
Now onto the muscular type.
Just kidding, I mean the general artificial intelligence or the strong one.
Did I just imply that muscular people are strong? Perhaps, but both of us know that, it’s not always true, yet I am not entirely wrong either.
Anyways…
General AI refers to a design of AI capable of performing any intellectual task a human can do. These systems would have a broader scope of understanding and adaptability enabling them to learn and apply knowledge across diverse domains
Let’s just say, it’s a jack of all traits.
But a master of none, because it’s a theoretical model still and the more research is being carried out.
Super-Intelligent AI
Lastly Superintelligent AI is still a future concept involving AI that surpasses the human intelligence across all fields, including creativity, problem-solving and decision making.
The hint’s in the name. It will be super intelligent.
But you know what, I strongly believe that machines can never be smarter than humans because yes, they can be fast and efficient, they still lack emotional intelligence. Without a doubt, we need logic and brains to run our lives but half our decisions are based on emotional queues which they can’t have because they have nothing to lose, no positive or negative consequences that’ll derive their actions.
Anyhow, I assume you’d want to know some of the common application in our lives…
So, either shockingly or not so shockingly, AI is being used by streaming services like Netflix and Spotify and online shopping platforms like Amazon.
Another common use is the popularity of chatbots for customer support (which I experienced first time round on WhatsApp which is quite an experience, because isn’t WhatsApp made to make human to human interaction easier?)
AI diagnostic tools, machine translation and even content creation is being swarmed by AI tools which has its pros and cons simultaneously.
The Building Blocks of AI
Now, hopefully most of us do know that AI isn’t some magic because it doesn’t appear out of the blue.
It is in fact very much based on logic and code hence thankfully it very explainable.
Umm… well sort of anyways.
AI is built on three key pillars: data, algorithms and computing power. These three core elements allow AI systems to perform complex task and mimic human intelligence.
Data
In order to run anything and keep something alive we need fuel to provide energy right?
Well then, you’ll be pleased to learn that AI is no different. Data is often called the “fuel” of AI. Without data, AI system will metaphorically and literally die because it can’t learn, make decisions or adapt to new situations.
And mind you, AI is a greedy monster, it requires large data sets because they allow the AI models to recognise patterns and work on improving accuracy. Similar to learning a language where you’d need perhaps more than a few pages of vocab and learn common language patterns, AI requires large datasets to perform better or become more fluent in “real world situations”.
Types of Data Sets
Luckily, AI isn’t too picky about what it eats or doesn’t, so, datasets can be text data, image data or sensor data.
Here’s a brief about each of them:
- Text Data: Enables chatbots, translation tools, and content recommendations (e.g., Netflix or YouTube).
- Image Data: Powers facial recognition, medical image analysis, and object detection in self-driving cars.
- Sensor Data: Collected from IoT devices and used in smart homes, environmental monitoring, and predictive maintenance.
Algorithms
If data is the fuel, then algorithms are the recipes that transform data into meaningful insights. In other words, algorithms are sets of instructions that allow machines to learn from data and make decisions.
Our AI monster happens to be a chef as well.
But how do algorithms work?
- A decision tree sorts data based on a series of yes/no questions to arrive at a conclusion.
- Then Neural networks, inspired by the human brain, use interconnected layers of nodes to process complex data like speech or images.
Think of algorithm as a step-by-step process for solving a problem.
Real-World Examples of Algorithms:
- Decision Trees: Used in credit card fraud detection by analysing transaction patterns.
- Neural Networks: The backbone of voice assistants like Siri and Alexa, which understand and respond to speech.
Computing Power: The Engine Driving AI
Well monsters aren’t always the most independent creatures because of their wild and scary nature. Therefore, in order to make them co-orporate and keep them in line with expectations as to prevent them from scaring people and messing things up, advances in computing power are crucial.
High performance hardware enables the training of massive AI models, often involving a ton of math and code.
Advanced Hardware: GPUs and Beyond
Graphics Processing Units (GPUs) are especially important in AI. Unlike traditional CPUs, GPUs can process multiple tasks simultaneously, making them ideal for training complex models like deep neural networks.
Cloud Computing and AI Scalability
Cloud computing platforms, like AWS, Google Cloud, and Microsoft Azure, provide on-demand access to powerful hardware. This allows even small organisations to develop AI solutions without needing to invest in expensive infrastructure.
3 CORE CONCEPTS
Machine Learning
Machine learning occurs through a similar process to sitting in a class room as a student. Because in this case the machines are students, learning from prior data and algorithims rather than being programmed in the first place. It is used by many AI systems but the three main categories include:
Supervised Learning:
This model uses labeled data sets to predict outcomes and recognise patterns.
“What are labeled data sets” you may ask?
Valid question because it took me a while to understand it as well.
It is a collection of data where each piece of information has assigned tags or labels. These aid in assigning outputs to specific outputs.
This is why spam detection in emails and predictive analytics which can forecast sales, use supervised learning.
Unsupervised Learning:
Well everything has its equal and opposite ( I am sure AI was not one of the things Einstein was thinking of when he came up with this).
In unsupervised learning, the student is a bit more independent and works with unlabeled data.
Its like being given a maths problem. In supervised learning, the student was told how to carry out BIDMAS . It has set rules, and steps.
In unsupervised learning, the student AKA the machine doesn’t have a template because the date set is unlabeled. Therefore the student has to seek patterns and group data on its own.
This is often used to identify anomalies in system monitoring.
Reinforcement Learning:
This type of model learns through trial error. Like humans, it is guided through rewards and penalties making it useful for scenarios incolving decision making.
Deep Learning
This subset of machine learning uses, neural networks modeled after the structure of the human brain. And just like humans, these networks can process vast amount of information.
Neural networks consist of interconnected nodes or neurons in a structure similar to human brain. These models excel at facial recognition and voice assistant such as Alexa and Siri.
And last core component is:
Natural Language Processing (NLP)
This enables machines to understand, interprent and produce human language. Enabling extensive understanding of computers; allowing chatbots, translation apps and content creation to be more extensively used.
Training An AI Model
AI has truly transformed how we approach problems and has provided us with efficient solutions.
But all that data, algorithm and code work isn’t enough.
Training an AI model involves multiple steps:
Collecting and Cleaning Data:
Data is the backbone of AI models and is immensely crucial.
Data can come from various sources therefore, will contain inconsistencies and missing values. Hence cleaning it is important. Cleaning data involves standardiSing formats, handling missing data and removing irrelevant entries
Training the Model On Data
Right now, we have all the essentials to train our machine, what do we do next, I hear you ask.
Well, we feed the cleaned data into a machine. A machine that learns the patterns and relationship.
But choose a model based on the problem because neural networks work well for image recognition, while decision trees are suitable for classification tasks. During the training, the model adapts its internal parameters to minimise errors.
It’s basically like any job, you gotta perfect and adjust over time!
Evaluating Performance
Evaluation is basically like a testing a student, but instead you are making sure that the machine is reliable and meaningful.
Making Predictions
This is when a trained model becomes a tool for making decisions or predictions in real-world scenarios.
Example Applications:
Here are some example applications…
- Weather Forecasting: Predict temperature, precipitation, and other conditions using historical weather data.
- Fraud Detection: Analyse transaction patterns to identify anomalies and flag potential fraud.
- Deployment: Models are integrated into applications or platforms where they function in real time. For instance, a fraud detection model may monitor transactions on a banking app
Feedback Loops
And the last part of training a model is to ensure the model maintains relevance and accuracy.
This is where models are retrained to adapt to learn changing conditions. It’s basically incorporating feedback and monitoring model performance over time to detect changes in accuracy.
Outro
And that’s it. We have a trained model, for now anyways.
This explanation is really dumbed down, and there are a lot more complex steps that are crucial in making AI models and I’d highly suggest looking deeper into it in order to get a more solid grasp of how AI works.
We’ll discuss the ethical challenges of AI in a few weeks, so look out for that and many more.
Let me know how you found this post!
Raniya Abrar