Transcription
Let's kick off this AI jump-start series with a high-level overview of artificial intelligence, including some key concepts that will enhance your understanding of ChatGPT, its contemporaries, and the revolution that they're spawning. And if you haven't yet, be sure to hit the Subscribe button on YouTube and sign up for emails at streamline.ai so you can make the most of AI and automation.
So, artificial intelligence is a discipline within computer science that broadly refers to the ability of computers to mimic human cognitive functions, especially reasoning, perception, and learning. And that empowers computers to perform tasks commonly believed to require the human mind, such as interpreting and generating language and multimedia. I like to think of AI as technologies that perform functions without the explicit programming of engineers; rather than humans having to define each step for processing information and returning a result, the system forms its own model through various forms of machine learning.
AI already manifests in many forms throughout your daily life. It powers web search engines like Google, voice assistants like Siri and Alexa, the content recommendations you receive on Netflix and YouTube, and the facial recognition system you use to unlock your iPhone. And if you're a digital content creator, you see a growing number of AI tools for composing and editing text, images, and video.
So, notwithstanding its form or function, AI always accepts input values and returns output values. And depending on the application, those inputs and outputs can be text (like written human or programming language), audio (such as spoken language or music), imagery or video, or mathematical or binary data. And in some cases, those outputs will trigger an action, such as an autonomous vehicle slamming on brakes. So, an AI's output represents its best prediction for the answers sought by the input. So, fundamentally, you can say that an AI system is a prediction model, and that model develops from the relationships and the patterns that the system identifies from training data. When Netflix recommends content, it's making a prediction about the content that you'll like based on your previous viewing behavior.
Artificial intelligence became an academic discipline in the mid-1950s, but for its first half-century or so, it was really only portrayed fictionally in movies and in books. But that changed in 2022 with the launch of ChatGPT, which made groundbreaking advancements in the capability and the accessibility of AI.
Let's unpack a few subfields of artificial intelligence that will help make sense of today's AI phenomenon. The AI we encounter daily is known as artificial narrow intelligence, which is also known as weak AI. It solves a single problem by performing narrowly defined, structured functions that result from human engineering; that includes web searches and content recommendations. By contrast, artificial general intelligence, or strong AI, is an aspirational level of AI where a computer's intelligence matches a human's and can be applied across many domains. We're not there yet, but it's a goal of the AI community. An artificial general intelligence may also be self-aware, but many experts classify that trait as artificial super intelligence.
Machine learning is the ability of AI to learn from its operations and improve performance over time without explicit programming from human engineers. Deep learning is an advanced form of machine learning powered by a neural network. A neural network, or an artificial neural network, is a layered network of nodes simulating connected neurons in the human brain. Each node plays a unique role in the collective algorithm, which allows the system to analyze and generate data at a high volume and a high rate of efficiency. And through training, the neural network becomes an AI model. That model takes shape as the system identifies patterns and relationships in training data.
So, engineers train these models in a combination of three ways: within supervised learning, the data set is pre-labeled (such as images containing a cat or a dog), and the AI system is given the full data set, including the classifications, which it uses to find consistencies among items with shared labels (such as all of the images with dogs and all of the images with cats). In unsupervised learning, the data set is unlabeled and often massive (such as billions of web pages), and the AI system detects patterns in relationships that are probably unrecognized by human engineers (such as patterns in natural language). Then, with reinforcement learning, AI systems learn through trial and error; so each action is going to be rewarded or it's going to be penalized, either manually or algorithmically, depending on the goal. And every time the system learns to maximize its rewards.
Back to that Netflix example, the system is going to learn from the recommendations that you actually watch, particularly those that you watch and complete. Generative AI is a class of artificial intelligence that produces content, such as text, images, audio, or video, given a prompt. It predicts the requested outcome using the model formed from supervised, unsupervised, or reinforcement learning. You probably saw those photos of the Pope in a puffy coat that went viral and fooled much of the web in March. Then, the next month, there was a song generated with the voices of Drake and The Weeknd that caused an equal stir. [Music] on my nerves [Music] And that autocomplete feature in Gmail is another example of generative AI that you likely experience frequently.
Natural language processing is a type of AI that interprets and often generates natural human language. And depending on the application, language is accepted and returned as text or as audio. Spell checkers, translators, and chatbots employ natural language processing. Neural networks that have trained and deployed are known as models. So, in the case of language processors, those models are known as large language models, or LLMs. LLMs are usually trained through unsupervised learning—that's with unlabeled data sets—and those data sets are typically massive, as I mentioned; they comprise billions of web pages and digitized books. The celebrated AI expert Gary Marcus described LLMs as "autocomplete on steroids." Then, LLMs built on a Transformer architecture have elevated comprehension and operational efficiency because they have the ability to eliminate, rearrange, and assign weights to words. That technology was discovered by engineers at Google in 2017 and it really set the stage for today's AI boom.
As a young and nerdy tech enthusiast, I quickly learned that a technology's capability correlates with its potential for good, but also its potential for evil. As technology advances, it can solve more problems or can cause greater destruction. It was true for splitting the atom, it was true for the internet, and it might be most true for artificial intelligence. The impact of AI really can't be overstated.
Not too long ago, Bill Gates published a host about AI that I highly recommend; it's linked within the companion post to this video. And in it, he says that the development of AI is as fundamental as the creation of the microprocessor, the personal computer, the internet, and the mobile phone. It will change the way people work, learn, travel, get health care, and communicate with each other. Entire industries will re-orient around it; businesses will distinguish themselves by how they use it.
Through the rest of this jump-start series, I'm going to reference the concepts I introduced in part one to help explain the current AI phenomenon. So, I encourage you to bookmark the companion posts, which will be linked in the description of this video, so you can revisit it for a refresher on these concepts as needed. And as we unpack AI, we're going to explore the opportunities, the threads, and many actionable tactics that you can employ right away to make the most of AI in this exciting moment. If this part of our series was helpful, please hit the like button and remember to subscribe at streamline.ai.