In this first installment of PAI’s Summer School Series, we’re breaking down the basics of artificial intelligence. Whether you are an AI pro or just beginning to explore it, this primer will help sharpen your understanding of the technology changing the world.
What is AI?
Human intelligence is defined as the ability to learn, reason, and apply knowledge or skills to solve problems. While Artificial Intelligence, or AI, is not like human intelligence, it is designed to simulate it. There’s a very important distinction between humans and AI, and that is that AI systems do not “think,” but rather reason and solve problems based on how they are trained. AI systems are built to process information, recognize patterns, and make decisions on a much larger scale and at a faster rate than humans.
There are different types of AI. Narrow AI, or “weak” AI, is the only kind of AI that exists today. It is designed to handle and manage very specific tasks and functions like recommending certain kinds of videos on social media or generating text (via AI assistants like ChatGPT). It can do many specific tasks very well but it can’t generalize or think beyond its programming. You may hear the terms “AGI” and “Superintelligence” being used as people speculate about the future capabilities of AI, but as of now, they are both hypothetical types of AI.
Common AI Terms You Might Have Heard
Many people first heard of AI in 2022, when Open AI first released their generative AI chatbot “ChatGPT”. While not a new concept, this AI tool became a viral sensation due to its accessibility and its impressive ability to “understand” and generate human-like text. Since the initial release of GPT-3.5, AI technology has taken center stage in the tech field, flooding the media with jargon and confusing terms. So what do these common AI terms even mean?
Foundation Model: Foundation models are AI systems with generally applicable functions that are designed to be used across a variety of contexts. The current generation of these systems is characterized by training deep learning models on large datasets (which requires significant computational resources) to perform numerous tasks that can serve as the “foundation” for a wide array of downstream applications.
Large Language Model (LLM): LLMs are a type of foundation model, primarily used to program systems, like generative AI systems. They are trained on very large sets of data and use machine learning techniques to improve and refine themselves.
Generative AI: Generative AI is a kind of AI system that can produce text, images, audio, and video. Examples of generative AI systems are Claude, Gemini, and ChatGPT, Midjourney, and Sora. These AI systems work by generating Synthetic Media when a user prompts the system with a specific request.
Agentic AI: This kind of AI system can act on behalf of users, with some degree of autonomy, to achieve goals without human intervention or guidance. The goals of agents are to understand a user’s general goals and utilize context to solve specific problems without explicit instructions. For example, where ChatGPT can generate a pizza recipe for you, if asked, an AI agent can find the best pizza place near you, look for and book a reservation, and schedule the reservation for when you are available in the week.
Although generative AI has become the most popular form of AI, there’s more to it. AI has actually been around since 1956, but the concepts which led to the development of AI have been around for much longer, with some theorizing it all started as far back as the eighteenth century. AI is in many of the applications and systems you interact with on a daily basis. When you apply for a car loan, an AI system is running in the background to weigh whether or not you should qualify. When you are scrolling through Netflix to pick your next binge-watch, an AI system is running in the background to decide which shows or movies to recommend to you based on your watch history. When you go on a roadtrip and use Google or Apple maps to help you navigate, an AI system is running in the background to optimize your route and avoid traffic. AI is everywhere, but how do these machines know what they know?
Machine learning: Machine learning refers to the method in which AI “learns” over time. Incorporating computer science, math, and coding, this process involves the development of algorithms that help machines learn without any human assistance.
Algorithm: Algorithms are instructions that tell the computer how to make decisions, perform tasks, or execute a function autonomously. Algorithms look for patterns in data and over time, as they work, it improves itself.
While much of this work sounds like computers are teaching computers how to “think” and “operate,” humans play a very critical role in the development of AI systems. Apart from overseeing the development of these systems, there are hundreds of millions of people around the world who collect the data that powers AI and train these machines to identify and recognize patterns in the data. There are also millions of people around the world who are dedicated to understanding the impacts these systems have on people and society.
Data Workers: Data workers are individuals who perform data enrichment tasks, such as cleaning, labeling, and moderating large datasets, that are crucial for training machine learning models, especially those powering AI systems.
Bias: Bias in AI refers to a systematic error that leads to unfair outcomes. Bias is typically introduced through human error in programming, data collection, or training. Bias in AI systems can exacerbate preexisting risks posed to marginalized groups.
Responsible AI
While AI has become a powerful tool in our everyday lives it is also important to recognize that the ubiquity of it means it can also have great potential for harm or misuse. Used irresponsibly, AI can amplify and reinforce discrimination, violate privacy, and even be used to spread false information. That is why Partnership on AI is dedicated to advancing the responsible development, deployment, and use of AI systems. AI is a tool that can be leveraged for a multitude of purposes, but it should always benefit society. To learn more about how we advance responsible AI, sign up for our newsletter.
