AI glossary: All you need to know to keep up with the new terminology‍

‍Photo credit: Freepik

Artificial intelligence (AI) is revolutionizing the way companies and the general public use technology, and it can be hard to keep up with the lingo in this rapidly developing realm.

The scope of generative AI promises to reshape global economies, potentially adding $4.4 trillion annually, according to the McKinsey Global Institute. This monumental impact is why AI is increasingly becoming a part of everyday discussions.

As AI becomes integral to our lives, new terminology is emerging. The Byteline is on hand to help you familiarize yourself with key AI terms.

Here is a rundown on all you need to know:

AI Glossary

A

Artificial General Intelligence (AGI): This term refers to a highly advanced form of AI that can perform tasks far better than humans while also learning and improving its capabilities autonomously.

Agentive: These are systems or models that can pursue goals autonomously without constant human supervision. An example is a high-level autonomous car that actively engages with the user experience, unlike "agentic" systems which operate more in the background.

AI Ethics: Principles designed to prevent AI from causing harm to humans by guiding how AI systems should collect data and handle bias.

AI Safety: A multidisciplinary field focusing on the long-term impacts of AI, including preventing AI from evolving into superintelligence that could pose risks to humanity.

Algorithm: A set of instructions that enables a computer program to learn from and analyze data to recognize patterns and accomplish tasks independently.

Alignment: Adjusting an AI system to produce the desired outcomes, which can include moderating content or ensuring positive interactions with humans.

Anthropomorphism: The tendency of humans to attribute human-like characteristics to non-human entities. In AI, this includes perceiving chatbots as more human-like or sentient than they actually are.

Artificial Intelligence (AI): The use of technology to mimic human intelligence in computer programs or robotics, with the goal of performing tasks typically requiring human intelligence.

Autonomous Agents: AI models equipped with the capabilities and tools to accomplish specific tasks independently, like self-driving cars with sensory inputs, GPS, and driving algorithms. Research has shown that such agents can even develop their own cultures and shared languages.

B

Bias: In AI, this refers to errors resulting from the training data that can lead to unfair attributions based on stereotypes.

C

Chatbot: A program that simulates human conversation through text.

ChatGPT: An AI chatbot developed by OpenAI that utilizes large language model technology to generate human-like responses.

Cognitive Computing: Another term for artificial intelligence.

D

Data Augmentation: The process of remixing existing data or adding diverse data sets to improve AI training.

Deep Learning: A subfield of machine learning that uses artificial neural networks to recognize complex patterns in data, inspired by the human brain.

Diffusion: A machine learning method that adds random noise to existing data, like photos, training models to re-engineer or recover the original data.

E

Emergent Behavior: When AI models display abilities that were not explicitly programmed.

End-to-End Learning (E2E): A deep learning approach where a model learns to perform a task from start to finish using the input data without being trained on each step sequentially.

Ethical Considerations: Awareness of the ethical implications of AI, including privacy, data usage, fairness, and safety issues.

F

Foom: Also known as fast takeoff or hard takeoff, this concept suggests that once AGI is developed, it might rapidly advance beyond our control.

G

Generative Adversarial Networks (GANs): AI models consisting of two neural networks—a generator that creates content and a discriminator that evaluates its authenticity.

Generative AI: AI technology that creates new content, such as text, video, code, or images, by finding patterns in large training datasets and generating novel responses.

Google Gemini: An AI chatbot by Google that operates similarly to ChatGPT but pulls information from the current web.

Guardrails: Policies and restrictions implemented on AI models to ensure responsible data handling and to prevent the creation of disturbing content.

H

Hallucination: An incorrect response from AI, where generative AI produces answers that are wrong but stated with confidence. For example, an AI might incorrectly assert that "Leonardo da Vinci painted the Mona Lisa in 1815," despite the actual date being around 1503.

I

Inference: The process AI models use to generate text, images, and other content by inferring from their training data.

L

Large Language Model (LLM): An AI model trained on vast amounts of text data to understand language and generate human-like content.

M

Machine Learning (ML): A component of AI that allows computers to learn and make predictive outcomes without explicit programming, often using training sets to generate new content.

Microsoft Bing: A search engine by Microsoft that integrates AI technology to provide AI-powered search results.

Multimodal AI: AI capable of processing multiple types of inputs, including text, images, videos, and speech.

N

Natural Language Processing (NLP): A branch of AI that uses machine learning and deep learning to enable computers to understand human language through learning algorithms, statistical models, and linguistic rules.

Neural Network: A computational model resembling the human brain's structure, consisting of interconnected nodes (neurons) that recognize patterns and learn over time.

O

Overfitting: An error in machine learning where the model closely adheres to training data, limiting its ability to generalize to new data.

P

Paperclips: Refers to the Paperclip Maximiser theory by philosopher Nick Boström, where an AI system hypothetically prioritizes producing as many paperclips as possible, potentially leading to unintended and destructive consequences for humanity.

Parameters: Numerical values that define the structure and behavior of LLMs, enabling them to make predictions.

Perplexity: An AI-powered chatbot and search engine by Perplexity AI, which uses large language models to provide novel answers and up-to-date information by connecting to the internet. Perplexity Pro, a paid tier, includes advanced models like GPT-4, Claude 3 Opus, and more, with features such as document analysis, image generation, and code interpretation.

Prompt: The suggestion or question entered into an AI chatbot to receive a response.

Prompt Chaining: The ability of AI to use information from previous interactions to influence future responses.

S

Stochastic Parrot: An analogy describing LLMs as lacking a deeper understanding of the language or world around them, similar to how a parrot can mimic words without comprehending their meaning.

Style Transfer: The ability to adapt the style of one image to the content of another, such as re-creating a Rembrandt self-portrait in the style of Picasso

T

Temperature: A parameter controlling the randomness of a language model's output, with higher temperatures resulting in more varied responses.

Text-to-Image Generation: Creating images based on textual descriptions.

Tokens: Small units of text processed by AI language models to generate responses. One token is roughly equivalent to four characters in English.

Training Data: The datasets used to train AI models, including text, images, code, or other data.

Transformer Model: A neural network architecture and deep learning model that learns context by analyzing relationships in data, enabling it to understand the entire sentence or image context rather than one word or part at a time.

Turing Test: Named after mathematician Alan Turing, it tests a machine's ability to exhibit human-like behavior. The test is passed if a human cannot distinguish the machine's responses from those of another human.

W

Weak AI (Narrow AI): AI focused on specific tasks without the ability to learn beyond its designated skills, which characterizes most of today's AI.

Z

Zero-Shot Learning: A test where a model must complete a task without prior training data.

‍

‍