Foundations of Artificial Intelligence Certificate

Welcome to the Foundations of AI certificate! In this module, we will take a bird’s eye view of AI, and understand why the field fundamentally changed in 2020, making new applications possible that would have seemed like science fiction just a few years ago.  

By the end of this module, you will be able to: 

• Describe what makes this moment in the history of AI so special.
• Explain why, as of 2020, progress in AI accelerated so rapidly.
• Identify the eras that have led to the current AI landscape.
• Recognize the significance of GPT-3 and AI scaling.
• Identify what new industry applications of AI have been unlocked since 2020.
• Describe how AI accidents and malicious applications of AI are introducing significant global risk.
• List some measures organizations can take to avoid falling behind.
   

In this module, we’ll explore datasets and algorithms – the fundamental building blocks of AI systems. We’ll look at what models and algorithms are, as well as what types of data exist. Finally, we will spend time learning about the different machine learning tasks that we normally see in industry and research. By the end of this module, you’ll have a solid understanding of what data is, and you’ll know how two of the world’s most popular AI algorithms work, from the ground up!

By the end of this module, you will be able to: 

• Identify what an “AI model” is.
• Describe what AI data is, and how models can learn from it to develop useful capabilities.
• Recognize some basic types of AI models (including decision trees and K-nearest neighbors).
• Explain how simple AI models can be applied in industry to solve real-world problems.

In this module, we’ll complete our discussion of decision trees, and conclude our exploration of the era of AI 1.0. Then, we’ll dive into neural networks, a new kind of AI model that’s based on the structure of the human brain. We’ll see how neural networks work, how they relate to the human brain, why they kicked off the AI 2.0 era, and how they’re being applied today by companies like Netflix and OpenDoor. 

By the end of this module, you will be able to: 

• Recognize what terms like “AI bias” and “AI fairness” mean.
• Explain how a simple model called “linear regression” seeded a revolution in AI.
• Identify what neural networks are and how they work.
• Compare how deep learning relates to neural networks.
   

In this module, we’ll be taking a closer look at deep learning strategies for vision. In the process, we’ll uncover one of the most important properties of modern AI systems, a property that’s shaped the corporate strategy of world-leading tech companies, and paved the way for the deep learning revolution of the AI 2.0 era. Additionally, we will learn about “transfer learning”, and how this revolutionized the economics of AI. For the first time, companies could justify investing hundreds of thousands of dollars in training huge neural networks, knowing that their weights could be repurposed on new tasks with relatively limited training. The result was a significant consolidation of effort around a smaller number of more expensive AI systems.   

By the end of the module, you will be able to: 

• Explain transfer learning, and why it fundamentally changed the economics of AI.
• Describe how computer vision models powered by deep learning can be used to solve practical problems.
• Identify what reinforcement learning is, and how it works.
• Recognize how reinforcement learning can be used in the industry.
   

Of all the tasks you can get an AI system to master, text generation is likely the most valuable, and the most fascinating. AI models that learn to generate text are known as language models, and for reasons we’ll be exploring in this lesson, language modeling turned out to finally offer a path beyond narrow AI: a way to build individual AI systems with a wide range of far more humanlike capabilities. In the next few sections, we'll see what the world of language modeling has to offer, and why it’s going to be a key part of the future of AI.

By the end of the module, you will be able to: 

• Recognize how language models powered by neural networks can learn to read and write.
• Explain how transfer learning applies to language models.
• Identify what GPT-3 is, and how it works under the hood.
• Describe what special processors are currently being used to build the world’s most advanced AI models.
   

GPT-3 showed the world that AI systems can be made more general and capable simply via scaling, by using vast amounts of processing power and data to train gigantic models with a tremendous number of weights. There’s no known theoretical limit to how far scaling can go. Here, we will explore the new wave of AI systems, as well as key considerations when using or building these systems.

By the end of the module, you will be able to: 

• Describe what prompt engineering is, and why it’s become such an important skill in the age of powerful language models.
• Explain what language models are really trying to do, and how that can lead them to behave pathologically.
• Recognize how AI can be used to generate images.
• Identify foundation models, and why they’ve changed the economics of AI.
   

In this module, we’ll explore an important topic: malicious applications of AI and adversarial AI.  

By the end of the module, you will be able to: 

• Identify why a wide range of malicious applications of AI have become possible since 2020.
• Describe what new malicious applications are now possible, from election interference to cyberattacks.
• Recognize what organizations can do to limit their exposure to malicious applications of AI.
• Explain how AI systems can be intentionally undermined through adversarial AI attacks.
• Describe how the risk of adversarial AI attacks can be limited.
   

The problem of getting AI models to reliably generate the kind of output we want isn’t restricted to language models; it’s a universal feature of all AI systems. As those systems become more scaled and capable, it’s a problem that will get worse, and not better. Ultimately, many frontier AI researchers believe that AI systems with advanced capabilities, but poorly controlled behavior may pose an unprecedented risk to global safety. In this module, we’ll dive into what happens when AI goes wrong, and what the long-term outlook of AI accidents might be.

By the end of the module, you will be able to: 

• Identify the causes of AI accidents.
• Recognize the factors that increase AI accident risk.
• List how organizations can reduce AI accident risk.
• Describe what the future of AI accidents might look like.
• Explain why an increasing number of AI experts are converging on the view that AI accidents may represent a source of global catastrophes.