The Differences Between Common Artificial Intelligence (AI) Types Explained

Artificial intelligence (AI)—you’ve heard of it, you’re likely using it, and you know it’s already used everywhere and its reach will only likely increase. These days, the term "AI" is thrown around frequently, but because this technology is actually made up of many different subsets that generally all get thrown under the umbrella of AI, it can sometimes lead to confusion.

Being cybersecurity experts, we probably know more about artificial intelligence than the average person, as it’s our job to help our clients maintain the security of their digital systems—AI included. To help you, and them, understand the differences between various AI tech, we’re going to break down some key terms and explore their unique characteristics, applications, and impact on our lives so that, moving forward in this digital world, you’ll be more able to differentiate the differences.

Defining Artificial Intelligence (AI)

Where else to start but with the blanket term “artificial intelligence?”

Merely saying a system is or contains AI is like saying “I’m an engineer,” or “I’m in marketing”—e.g., AI is a basic concept, but when you go deeper, there are a lot of diverse details, aspects, and applications.

Before we get into its details, we must define this blanket term. Simply put, AI is computer-based operations used to mimic human intelligence, meaning they go beyond standard automation and computation to simulate the human ability to:

• Learn by applying context and improving decisions based on previous experiences.

• Reason by building concepts and determining their likelihood based on previous or associated events.

• Take chances in terms of how we make best-guess decisions and commit to them.

• Innovate.

General Examples of AI

AI represents a huge breakthrough in technological advancement, and we’re already putting it to good use in many different ways that comprise, among other applications:

• Fraud detection/prevention: AI can leverage transaction patterns and metadata to identify and prevent fraudulent activities.

• Self-driving cars: AI can utilize real-time inputs from sensors and harness algorithms to navigate your car on roads and make driving decisions—these are hugely complex operations, which is likely why fully autonomous cars are not available yet.

• Virtual assistants: AI bots like Siri, Alexa, and Google Assistant recognize voice commands and respond with the desired information. (And if they don’t, it may be that you just need to speak a little louder and slower. 😉)

But these use cases—as well as the many others out there—are representative of different kinds of AI, and these differentiations are and will be important to know as this technology continues to take root across society.

3 Types of Artificial Intelligence

Let’s break down the big three of AI—the categories of machine learning, large language models, and generative AI, along with examples of each.

1. Machine Learning (ML)

Machine learning is a subset of AI that includes systems that can simply be fed data from which they will learn and improve over time. Because ML algorithms can identify patterns and make predictions based on the data they are trained on, as more data is fed into a system, ML uses that expanding knowledge base to improve its accuracy and performance.

It may sound simple, but we shouldn’t understate the complexity of this technology. ML requires a lot of:

• Preprocessing (transforming data into a usable format);
• Feature selection (identifying and selecting the variables necessary to make a decision to get rid of noise); and
• Model tuning (optimizing performance and results).

 Some examples of ML you might be familiar with include:

• Content recommendations: Services like Netflix, Flipboard, Spotify, and Amazon use ML to push new movies, music, articles, etc. to users based on each individual’s history, associated actions, and chosen preferences.

• Spam protection: ML algorithms can review the sender, subject, text, format, and content of text messages and emails to identify and filter out spam.

2. Large Language Models (LLMs)

A subset of machine learning is large language models, which are AI systems specifically designed to understand, interpret, and generate text—so, whereas ML can serve a broader purpose using many different data types, LLMs only process and generate human language (and appear human).

Built upon deep learning techniques—particularly transformer architectures that allow them to capture complex relationships between words and sentences—examples of LLMs include:

• Chatbots: LLMs power those advanced chatbots you engage with on websites—they are designed to conduct natural conversations, answer questions, and provide assistance.

• Machine translation: Google Translate and other translation services utilize LLMs to translate text between languages with increasing accuracy.

• Text summarization: LLMs can condense lengthy articles or documents into concise summaries.

3. Generative AI (No acronym—these parentheses are included because we didn’t want to it to feel left out)

Despite the incredible usefulness of ML and LLMs, the iteration of artificial intelligence currently capturing the public’s attention is generative AI—or the systems/applications capable of creating entirely new content, be it text, images, audio, or even code.

Think about this—you go to a museum with a watercolor set and start copying a famous piece of art. Your finished painting may be your output, but your inspiration came from another’s work. In the same way, generative AI relies upon training data to generate unique outputs that correlate to what was provided before.

Practical examples of generative AI include:

• Image generation: Services like DALL-E 2, Stable Diffusion, and Midjourney use generative AI to build images from textual descriptions.

• Music composition: Soundraw, Musicfy, and Suno are just a few examples of algorithms that generate musical pieces in various genres and styles.

• Code generation: Generative AI tools such as GitHub Copilot or Cursor assist in code development by suggesting code snippets and completing code based on the provided context.
  • Both of those referenced tools rely upon coding data and not all information about coding, which is akin to learning how to be a chef by cooking with a chef rather than reading books with favorite recipes.

Addressing the Opportunities and Challenges of All Artificial Intelligence

We wrote this blog post to provide definitions for the real-world applications of and context for artificial intelligence and associated terms because they’ll be important to know moving forward—ML, LLMs, and generative AI will surely continue to provide unparalleled potential for solving complex challenges across various industries as they are already.

(It’s also important to note that there’s overlap between these concepts, as generative AI often leverages machine learning and large language models to perform its tasks, creating a synergistic relationship between these different AI subsets.)

Right now, universities are using these types of tools to create new medications or give personalized medicine, while engineering projects are bolstered by measurements and math that would previously been limited by humans needing more time to calculate or consider inputs. In light of all this innovation and progress that will extend into the future, we also need to stay aware of the flipside of all this AI, including the social, economic, and ethical considerations of its use that could affect data privacy and the issue of bias.

To make the best decisions regarding all kinds of AI, we must stay informed and continue to ask its providers how they ensure the security and ethics of their applications. After all, when engaging human engineers and marketers, we ask for details on their prowess and assurances for a realistic outcome—we should treat AI the same way.

For more information on how AI developers and providers can demonstrate how they responsibly manage their applications, check out our other content:

• AI Data Considerations and How ISO 42001—and ISO 9001—Can Help
• NIST's AI Risk Management Framework Explained
• An Explanation of the Guidelines for Secure AI System Development