This article is part of Lifehacker's “Living with AI” series. We explore the current state of AI, explain how it's helping us (and how it's not helping us), and evaluate where this revolutionary technology is headed next. Learn more.
Artificial Intelligence (AI) is the latest technological revolution. Just as the cryptocurrency boom brought a flood of new jargon to the world, the AI hype has brought with it a set of terms that are frequently used but not always explained. If you're wondering about the difference between a chatbot and an LLM, or the difference between deep learning and machine learning, you're in the right place. Here's a glossary of 20 AI-related terms and their meanings, explained in a beginner-friendly way.
Artificial Intelligence (AI)
Simply put, AI is computer or machine intelligence, especially intelligence that mimics human intelligence. AI is a broad term that covers many types of machine intelligence, but today the discussion of AI is focused primarily on tools that create art and content, summarize content, and transcribe content. Whether these tools are called “intelligent” is up for debate, but the term AI is here to stay.
algorithm
An algorithm is a set of instructions that a program follows to produce a result. Common examples of algorithms include search engines that display a set of results based on a query, or social media apps that show content based on a user's interests. Algorithms enable AI tools to create predictive models or create content or art based on user input.
bias
In the context of AI, bias refers to incorrect results produced by algorithms making false assumptions or with insufficient data. For example, a speech recognition tool may not correctly understand a particular English accent because it was only trained on American accents.
Conversational AI
AI tools that you can talk to, like chatbots and voice assistants, are called conversational AI, even when you ask the assistant questions yourself.
Data Mining
The process of sifting through large amounts of data to find patterns and trends. Some AI tools use data mining to help you understand why people are buying more products in your store or on your website, and how to optimize your business to meet increased demand during peak periods.
Deep Learning
Deep learning attempts to replicate how the human brain learns by utilizing three or more neural network “layers” to process large amounts of data and learn from examples. Each of these layers processes its own view of the data it is given, and then they are combined to arrive at a final conclusion.
Through object recognition, self-driving car software uses deep learning to identify stop signs, lane markers, and traffic lights. This is achieved by showing the AI tool many examples of what a particular object (such as a stop sign) looks like, and through repeated training, the AI tool will eventually be able to identify that object with 100% accuracy, as close as possible.
Large Scale Language Models (LLM)
Large-scale language models (LLMs) are deep learning algorithms trained on large datasets to generate, translate, and process text. LLMs (such as OpenAI's GPT-4) enable AI tools to understand queries and generate text inputs based on them. LLMs also help AI tools identify and summarize important parts of text or video.
Generative AI
Generative AI can generate art, images, text, and other results from inputs, often driven by LLMs. Generative AI has become an umbrella term for modern AI techniques that many companies are now adding to their products. For example, generative AI models can generate images with a few text prompts or turn a portrait photo into a widescreen wallpaper.
Hallucinations
When an AI presents fiction as fact, we call it a hallucination. Hallucinations occur when an AI's dataset is inaccurate or its training is flawed, causing it to output answers it is confident about based on the knowledge it has available. That said, because AI is based on a complex web of networks, it doesn't necessarily understand all examples of hallucinations. Lifehacker writer Stephen Johnson offers some great advice for spotting AI hallucinations:
Image Recognition
The ability to identify specific subjects in an image. Computer programs can use image recognition to find and name flowers in an image or identify different bird species in a photograph.
Machine Learning
When an algorithm can learn from experience and data to improve itself, it is called machine learning. Machine learning is the common practice from which the other AI terms we have discussed are derived. Deep learning is a form of machine learning, and large language models are trained through machine learning.
Natural Language Processing
If a program can understand input written in a human language, it falls under natural language processing. This is how the Calendar app knows what to do when you write, “I have a meeting tomorrow at 8pm at the coffee shop on 5th Avenue,” or when you ask Siri, “What's the weather like today?”
neural network
The human brain has layers of neurons that constantly process information and learn from it. AI neural networks mimic this neuronal structure to learn from data sets. Neural networks are the systems that enable machine learning and deep learning, ultimately allowing machines to perform complex tasks like image recognition and text generation.
Optical Character Recognition (OCR)
The process of extracting text from an image is done by OCR. Programs that support OCR can identify handwritten or typed text and also copy and paste it.
Rapid Engineering
A prompt is a series of words you use to elicit a response from a program, such as a generative AI. In the context of AI, prompt engineering is the art of crafting prompts to ensure that a chatbot returns the most useful response. It's also a field where people are employed to come up with creative prompts to test AI tools and identify their limitations and weaknesses.
Reinforcement Learning from Human Feedback (RLHF)
RLHF is the process of training an AI with human feedback: when the AI returns an incorrect result, humans show the AI what the correct response would be. This allows the AI to return accurate, useful results much faster than it would otherwise be able to.
voice recognition
The ability of a program to understand human speech. Speech recognition can be used by conversational AI to understand queries and provide responses, or by speech-to-text tools to understand spoken words and convert them into text.
token
When you input a text query into an AI tool, the text is broken down into tokens, or sequences of common characters in the text, which are then processed by the AI program. For example, if you use a GPT model, the price is based on the number of tokens you want to process. You can calculate this number using the company's Tokenizer tool, which shows you how words are broken down into tokens. According to OpenAI, a token is roughly four characters of text.
Training Data
A training set or training data is the information that an algorithm or machine learning tool uses to learn and perform its functions. For example, a large-scale language model might use training data by scraping some of the world's most popular websites to pick up text, queries, and human expressions.
Turing Test
Alan Turing was a British mathematician known as the “father of theoretical computer science and artificial intelligence.” His Turing test (or “imitation game”) is designed to identify whether a computer's intelligence is identical to human intelligence. A computer passes the Turing test if a human would mistakenly believe that the machine's responses were written by a human.
