Demystifying AI Fundamentals: A Beginner’s Guide

Let’s cut the formalities and keep it real; we are all so confused!

When it comes to AI, many of us are completely lost right now. The amount of jargon being thrown around these days is more than enough to make your head spin. Terms like AI, ML, and LLM are nearly inescapable, and it feels like every week, there’s more to remember! I am sure at one point you’ve asked yourself, “What does it all mean?!”. Look no further, this article will break down the most fundamental terminology needed to navigate today’s AI landscape. So, without further ado, let’s get to it!

Here’s the high level:

• Artificial Intelligence (AI) is the idea that software can be built to mimic human intelligence.  AI has two major categories, Traditional Rules Based AI and Machine Learning AI.
• Traditional Rules Based AI follows a script and provides a consistent experience.
• Machine Learning AI (ML) is designed to learn and improve over time, able to give dynamic and contextual results. 
• Generative AI (GenAI) is a type of ML that specializes in creating new and unique text, images, or sounds.
• Large Language Models (LLMs) is a type of GenAI that focuses on understanding human language, making it easier for humans and technology to have a dialogue.

Welp, there you have it 🙂, a short and sweet summary of the most fundamental terminology needed to navigate today’s AI landscape.

For those that are looking for a deeper walk through, the rest of this article is for you so buckle in for an in-depth overview of these concepts!

Let’s start out with Artificial Intelligence (AI). AI is the idea that software can mimic human intelligence. AI can be quite simple, or extremely complex, and happens to already be widely used in a range of ways in our day to day.

Have you ever called to pay your bill, and interacted with an automated voice instead of a human operator? Have you ever started to type a search or a message, and autocomplete started to give you suggestions? Have you ever taken a picture with your smartphone, and the camera automatically began to focus the shot perfectly, and even blurred the background noise? Have you ever used Google Maps for directions, instead of listening to that well-meaning backseat driver? 

These technologies all use the magic of Artificial Intelligence!

AI can be broken down into two major categories, Traditional Rules Based AI and Machine Learning Based AI.

Traditional Rules Based AI follows a human predefined template. Domain experts, data scientists, and engineers collaborate to create the template for the system, like a team of writers drafting a script. Once everything is ready, the system is released to users, and every interaction is intended to be quite structured, like an actor following a script. Let’s take the example of the automated voice operator. Using Traditional Rules Based AI, the interaction might go something like this. “Thank you for calling. For bill pay, press 1. For technical support, press 2. For new accounts, press 3. All other inquiries, stay on the line.” This generic template is well defined and consistent for all customers. The interaction replaces the need for a human operator in many cases, benefiting the company and reducing wait times for customers. 

Machine Learning AI (ML) follows a human defined “approach to thinking” known as a model. Researchers, data scientists, and engineers meticulously work to create the model in ways that will be able to guide the system when faced with unknowns, much like a teacher teaching a student core learning concepts such as recognizing patterns, using context clues, pneumonic devices, or fundamental arithmetic like addition and subtraction. Once the model is defined, the system is given training data, to learn from and practice on, just like a teacher would provide a study guide of reading materials and practice exams. ML is designed to improve overtime, given a better model or better training data, are both ways to improve ML systems before releasing them. 

ML powered software gives dynamic outputs that don’t follow a predefined script. Let’s take the example of Google Maps. Imagine you are driving to a restaurant for the first time. The restaurant is located on the other side of town, and as you make your way over, traffic conditions start to change unexpectedly. Google Maps is well ahead of the curve, learning in real time that there is an accident about 6 miles from your current location and predicting a 40-minute slowdown if you don’t change course! Nearly simultaneously, Google Maps is able to reassess the city map and traffic patterns in seconds, providing you three alternative routes to pick from, each with an arrival estimate for you to consider bringing into consideration speed limits, road closures, and even how other drivers are changing their routes! This is an ML AI system at work. ML AI systems are designed in ways that can adapt, learn, get better, and often can provide a very personal experience. 

Now, let’s go one step further and dive into the concepts that have really brought AI and ML centerstage in 2023: GenAI and LLMs.

Generative AI (GenAI) is a type of ML that can create (generate) new and unique content such as text, images, or audio. The range of what GenAI can produce is nearly as broad as the human imagination goes; from beautiful logos to business plans, from trip itineraries to working code samples, from music samples to research papers, GenAI has been able to create high quality outputs in ways that are nearly indistinguishable from humans across every domain. This feat is even more impressive as we look at how GenAI solutions are being made available to people of all ages and backgrounds, requiring zero technical know-how, thanks to the power of LLMs.

Large Language Models (LLMs) are a type of Generative AI that is designed to understand and interpret human language. LLMs, in many ways, make it truly possible for an unbounded dialogue between human and machine. The impact of this cannot be overstated. Have you heard of Jarvis from Marvel’s Iron Man, C-3PO from Star Wars, or Marvin from The Hitch Hiker’s Guide to the Galaxy? These are no longer far-off fantasy characters; these are much more within reach due to the innovations of LLMs!

Let’s break down how LLMs work. One fundamental aspect of LLMs is their approach to training is quite intensive. Imagine the difference in skill between an actor that has taken a few improv classes before getting on stage, as opposed to one that has studied every film, play, or animation ever made, across all genres and languages to hone their craft! Or the difference between a student that has completed an assigned summer reading list as opposed to a student that has completed the necessary studies required to earn PhDs across all disciplines underlying STEM and the Humanities! LLMs are overachievers in every regard, and teams have trained these models using everything on the internet; this broad knowledge allows LLMs to be the ‘know it all’ solutions of today. We humans know that information overload isn’t always the best thing and that everything you read online isn’t true, so although LLMs are quite impressive, researchers are continuously refining the underlying model and training sets to help these systems improve to be more effective, more inclusive, and safer for everyone.

You did it! Now you have an understanding of the core concepts underlying the next big technological innovation of humanity.

I recommend you explore for yourself today’s AI landscape using one of these free products that bring these technologies to life:

• Bard by Google and Google DeepMind
• ChatGPT or DALL-E by OpenAI
• Copilot by Microsoft and OpenAI

Using any of these you can bring to life a reality where AI assistants are helping YOU daily!

#artificialintelligence #machinelearning #generativeai

#ai #ml #genai #llm