Your initial introduction to the world of AI might have been through ChatGPT, OpenAI’s AI chatbot, which possesses an astonishing capacity to respond effectively to a wide array of questions. It’s renowned for its prowess in generating poems, resumes, and even fusion recipes, often being likened to an amplified version of autocomplete.
However, AI chatbots are just a single facet of the expansive AI landscape. While having ChatGPT assist in your homework or witnessing Midjourney create captivating mech images based on their country of origin is undoubtedly captivating, the potential of AI transcends such uses. In fact, this potential has the capacity to significantly reshape economies—a prospect that the McKinsey Global Institute estimates to be worth a staggering $4.4 trillion to the global economy annually. Consequently, you can anticipate hearing more and more about artificial intelligence in the times ahead.
As society becomes increasingly accustomed to an existence intertwined with AI, novel terms are cropping up at every turn. Whether you’re aiming to engage in insightful discussions over drinks or aiming to make a lasting impression during a job interview, being well-versed in important AI terms is crucial.
This glossary remains a work in progress, continuously evolving to encompass the ever-expanding AI lexicon.
Artificial general intelligence, or AGI: This concept points toward a more advanced iteration of AI than what we currently possess—a form of AI that can outperform humans in tasks, all while refining its abilities through teaching and autonomous advancement.
AI ethics: These principles are designed to safeguard humanity from the harmful implications of AI, achieved through guidelines for data collection, bias management, and other pertinent matters.
AI safety: An interdisciplinary domain concerned with the long-term consequences of AI development, particularly in relation to the sudden evolution of AI into a superintelligence that could potentially act against human interests.
Algorithm: A series of instructions enabling a computer program to learn, analyze data, and execute tasks based on patterns it discerns.
Alignment: The process of fine-tuning an AI to generate desired outcomes, extending from content moderation to fostering positive human interactions.
Anthropomorphism: The human tendency to attribute human-like characteristics to non-human entities, which, in the context of AI, can lead to the mistaken belief that a chatbot possesses human-like awareness or emotions.
Artificial intelligence, or AI: The employment of technology to replicate human intelligence, either through computer programs or robotic systems—a field within computer science dedicated to creating systems capable of performing human tasks.
Bias: In the context of large language models, inaccuracies resulting from the training data that can perpetuate stereotypes or associate specific traits with particular races or groups.
Chatbot: A software program that interacts with users through text, mimicking human language.
ChatGPT: An AI chatbot developed by OpenAI, employing large language models to facilitate conversations.
Cognitive computing: Another term for artificial intelligence.
Data augmentation: The practice of remixing existing data or incorporating a diverse range of data to train AI models.
Deep learning: A subset of machine learning involving intricate networks to recognize intricate patterns in images, sound, and text, inspired by the human brain’s functionality.
Diffusion: A machine learning technique introducing random noise to existing data, thereby training networks to restore or re-engineer that data.
Emergent behavior: The phenomenon wherein an AI model exhibits unforeseen abilities.
End-to-end learning, or E2E: A deep learning approach in which a model is taught to perform a task from start to finish, bypassing sequential learning by solving the task holistically.
Ethical considerations: The consciousness of ethical consequences tied to AI, encompassing privacy, data usage, fairness, misuse, and other safety-related matters.
Foom: Also known as fast takeoff or hard takeoff, this concept proposes that once AGI is built, intervening to protect humanity might already be too late.
Generative adversarial networks, or GANs: An AI model consisting of two neural networks—generator and discriminator—that collaborate to produce new data and validate its authenticity.
Generative AI: A technology generating content such as text, video, code, or images by training AI on substantial datasets to identify patterns and craft novel responses.
Google Bard: A Google AI chatbot akin to ChatGPT, drawing information from the current web unlike ChatGPT, which operates on data only until 2021 and isn’t internet-connected.
Guardrails: Policies and constraints imposed on AI models to ensure responsible data handling and prevent the creation of distressing content.
Hallucination: Erroneous responses from AI, sometimes expressed confidently despite their inaccuracy.
Large language model, or LLM: An AI model trained on extensive textual data to understand language and produce human-like content.
Machine learning, or ML: A component of AI allowing computers to learn and make predictions without explicit programming, often used alongside training data to generate new content.
Microsoft Bing: A search engine utilizing technology similar to ChatGPT, offering AI-powered search results and akin to Google Bard, as it’s connected to the internet.
Multimodal AI: An AI category capable of processing various inputs, including text, images, videos, and speech.
Natural language processing: An AI branch leveraging machine learning and deep learning to enable computers to understand human language.
Neural network: A computational model resembling the human brain’s structure, designed to identify patterns in data through interconnected nodes or neurons.
Overfitting: A machine learning error wherein a model closely mirrors training data, struggling to identify new data points.
Parameters: Numerical values shaping the behavior of LLMs, enabling them to generate predictions.
Prompt chaining: AI’s ability to draw on previous interactions for context in subsequent responses.
Stochastic parrot: An analogy depicting that LLMs lack comprehensive comprehension of language’s meaning, despite producing convincing responses, akin to parrots mimicking speech.
Style transfer: The capacity to merge the style of one image with another, allowing AI to replicate visual attributes across images.
Temperature: Parameters regulating an LLM’s output randomness—higher values result in more adventurous responses.
Text-to-image generation: Creating images based on textual descriptions.
Training data: Datasets that facilitate the learning process for AI models, containing text, images, code, or other relevant data.
Transformer model: A deep learning structure assimilating contextual relationships in data, like sentences or parts of images, to better understand overall context.
Turing test: A test, devised by Alan Turing, assessing a machine’s human-like behavior. If a human can’t distinguish the machine’s response from another human’s, the machine passes.
Weak AI, aka narrow AI: AI focused on specific tasks, incapable of learning beyond its designated skill set—most AI in use today falls under this category.
Zero-shot learning: A test where a model must complete a task without specific training data; for example, recognizing a lion despite being trained on tigers.