What Is AI?

Are you familiar with the Dunning-Kruger effect? You probably are in some way - maybe you've seen it in person. But, can you identify it when you are the one portraying it...?

Dunning-Kruger Effect: 

A Mental Bias where people w/ low skill or small knowledge in a field greatly overrate their own ability because they lack self-awareness to see their own mistakes. Conversely, people w/ high skill or great knowledge in a field greatly underrate their own ability because they know how hard/complex the subject is.

Confidence Level != Competence Level.

Video: Helpful Professor Explains - The Dunning-Kruger Effect (Explained in 3 Minutes)

You don't know what you don't know until you learn what you don't know.

Why do I bring this up? It's been my experience, from presentation after presentation, conference after conference, that no one is really an expert in AI. Now, I'm excluding those with backgrounds in neural networks and who actually work in the AI industry. I'm talking about the average Joe who uses AI. You think you understand the technology, you think you know all you need to know, and you're even really confident about it. And then, you learn something you didn't know/understand before. Maybe it's the environmental impact, being worst (or not as bad) as you thought. Maybe it's the cognitive offloading that happens when you give all your difficult tasks to something else. Maybe its the sheer amount of privacy you give up when you turn your life over to AI. Maybe its the quantity of inferences that AI makes about you in the background that you didn't realize was being made (and stored long term). Whatever it is, there's something that you likely don't know about AI.

You should consider yourself an open book, willing to take in new material related to AI - how it works, why it does what it does, what to watch out for, etc. You may not learn something new on this page, but if you're open to it, you'll likely discover something new... and if you're not open to it... well, a closed book can't get new information added to it.

If someone asks you if you use AI, your initial instinct will probably be to think about whether you use chat tools like ChatGPT, Gemini, Claud, etc. right? It's important to recognize that those tools are specifically Generative AI, or GenAI, for short, but, this is not the only type of AI.

You may have used AI today without even realizing it.

• Have you opened your phone with facial recognition?

• Have you used Facebook, Instagram, Netflix, spotify or any other application that suggests items you might want to see (posts, movies, music, news)?

• Have you selected any of the predictive text suggestions while typing out an email?

• Have you used any spellcheck or grammar improvement tools in Google Docs or Microsoft Word?

All of these tools constitute AI - Artificial Intelligence. AI is the invisible engine, or "brain", that runs the apps you use every day.

Not all AI is generative (take facial recognition technology, for example). Not all AI is presented as a chatbot. 

AI is being embedded in many of our appliances - from the vehicles we drive to the kitchen gadgets we own.

There's 3 stages of AI that you'll typically hear about.

Stage 1 - Narrow (or Weak) AI

This is any form of AI we have now or have seen in the past. This includes hard-coded AI, but typically we'll focus on more modern generative AI (ChatGPT, Gemini, etc.). This is considered "Narrow" and "Weak" AI because it generally focuses on a single task - image recognition, music generation, cancer identification, etc. ChatGPT and Gemini still fall into this category even though they seem like they can do multiple tasks (image generation, conversations, mathematics, poetry, etc.), but there's often multiple specialized agents that these tools outsource their work to. For example, when I first started using ChatGPT, it had a special bot connected to it called Dall-E that would take care of all the image generation requests. 

Stage 2 - Artificial General Intelligence (AGI, Strong AI)

This is a futuristic form of AI where it is capable of doing pretty much everything we can do just as well as we can do it. It learns and adapts to new circumstances. Essentially, it is AI that works just as well as the human brain at pretty much everything. At times, our AI systems feel like we're in this zone - AI can synthesize information from a book significantly faster than I can and it can make connections within the book I likely won't have thought of. But, it may only be trained on that one task and can't learn something new. 

Stage 3 - Superintelligent AI

This form of AI passes human intelligence in everything it does. If humanity ever got to this point, this becomes the points where AI leads human advancement (or, according to the Terminator, human destruction). 

When people talk about the different types of AI, they're not likely talking about the stages listed above. Instead, they're likely asking about the different ways that AI is created and what they can do. I will not create an exhaustive list, but I'll provide the most common current ones here:

• Rule-Based/Hard Coded AI - This is not the modern AI we think of since the advent of Generative AI like ChatGPT. This is the AI that we've had in gaming systems for a long time. There is no learning and no adaptation because the AI is hard coded into the system simply as a series of if statements (if the king is in check, calculate the all possible moves that will either move the king out of check or block the existing check). If a situation exists outside of that set of if statements, the AI has no way of addressing it. This is the AI that you play against in games and that determines whether your moves are legal. This is (or was) the AI that walks you through your online tax return filing. All programs that acted for themselves prior to the 2010's were written as rule-based, hard coded AI.

• Reactive AI - This AI analyzes a specific situation to determine the best response. It reacts to the current situation. It has no memory of past events and does not learn over time. It simply says "based on the current data, the optimal solution is _________".  Traffic management systems may use this form of AI to switch traffic lights based on the cars present at this time.

• Predictive AI - this is what companies like Netflix, Facebook, Google, Amazon, and other companies use to determine what we want to see based on what we've already seen. It analyzes historival data to find patterns and predict future outcomes or classify information. Netflix uses this to identify genres, movies, and shows you might like based on the content you've been watching. Google uses this to provide you with advertisements more tailored to your likes and to predict what you want to see when you Google "Penguins" (do you want the bird or the hockey team?). Facebook uses it to provide you with posts most similar to those you've been focusing on (based on your posts, your friends, your likes, the amount of time you appear to be viewing an add, a post, a video, etc.). Predictive AI is used in other applications as well, such as fraud detection, demand forecasting, customer advertising, healthcare risk, insurance risk, etc.

• Generative AI - This is likely what you're thinking of when you hear "AI".  Generative AI is all about creation. Creating dialog, creating images, creating videos, creating music. This AI is trained on huge data sets to find patterns. This is what you're using to cheat on your homework (knock it off!). You're likely familiar with ChatGPT, Google Gemini, Claude, and other GenAI tools. They're not all created equal and they're constantly fighting to get ahead of their competition. Many of these AI tools create profiles on their users in order to better tailor responses to them.

• Agentic AI - This AI connects to your apps and acts as an agent on your behalf. You assign it tasks and it takes care of the rest. The goal here is autonomy and completing tasks without additional prompts or human guidance. You can typically identify this because it's AI + something external (could even be something physical). Self Driving Vehicles (AI + Cars), Email monitoring AI (AI + Email), a CyberAI (AI + Computer), a news feed AI that searches for specific types of stories (AI + News). Some Generative AI services have expanded to also be Agentic AI services (ChatGPT and Gemini both allow this), but most of the time you'll use specialty software to link an AI to a system (email, calendar, etc.) and a set of goals/tasks.

• Supervised - AI is trained on existing, labeled data (AI sees the patterns and uses those to label more data)

• Unsupervised - AI is trained on existing, unlabeled data (AI has to figure out what patterns exist on its own)

• Reinforcement - AI learns what responses provide the best outcomes/rewards and adjusts, essentially like positive reinforcement in humans.

• Deep Learning - 

• Transfer Learning - 

• Y Magazine - I Chatbot - How worried should we be about the arrival—and impact—of chatty new AI machines? BYU experts weigh in.