Glossary of AI Terms
for Modern Marketers and Industry
30 Must-Know Terms, Definitions and Context
Learning AI can get overwhelming. Just breathe, take a few days off from it and keep this guide handy as you navigate the new frontier.
Artificial Intelligence (AI): a field of computer science and the development of intelligent systems (machines) that mimic human behavior and decision-making processes.
Artificial General Intelligence (AGI): A sub-set of AI and considered a holy grail by developers on the next frontier. A type of AI that possesses human-like ability to understand, learn, and apply knowledge across a wide range of tasks. No longer focused on “mimicking” human intelligence, AGI will come close to “being” a thinking, feeling human intelligence. (See Sentient)
AI Agents and Models: Both allow end users to interact with a computer using natural language vs. writing code. Both are designed to perform tasks and make decisions. The key difference is that Agents take actions and Models give answers. Some, like Siri, can take actions (set a time) and answer questions (weather in Hawaii today?) making it a model and an agent. Developers will eventually combine the strengths of both to become one useful and versatile application. For now they are separate.
AI Agent (takes action): Think of a robot assistant that can understand what’s happening around it, make choices and do things to reach its goal (Siri, Alexa, Roomba vacuum cleaners, self-driving cars and trading bots that buy/sell stocks or crypto).
AI Model: (gives answers): An AI model can be a Chatbot (ChatGPT, Claude, Gemini), a search engine (Perplexity.ai), a recommendation engine (Netflix, Spotify, etc.) or a Predictive Lead Scoring software (Salesforce’s Einstein).
Chain of Thought Prompting: A technique where you prompt the AI model to provide a step-by-step explanation of its reasoning process, enhancing its transparency and ensuring its understanding. A great way to train your AI and to avoid hallucinations.
Chatbot: A specific type of AI agent that uses an AI model to simulate conversation with human users, often used for customer service and engagement.
Content Window: The maximum amount of context, measured in tokens, that an AI model can consider when generating a response. A larger content window allows the AI to “remember” and take into account more of the previous conversation, enabling more coherent and relevant communication. (See Tokens)
PRO TIP: As of today, Google’s Gemini model has, by far, the largest content window available. DO know that (IMHO) Claude and ChatGPT4o are superior in other ways but if you have a large white paper to analyze, Gemini’s your bet.
Conversational AI: AI systems designed to engage in human-like conversations, such as chatbots and virtual assistants. Increasingly used for customer service, lead generation, and personalized marketing experiences.
Copyright Protections: Major lawsuits are underway where publishers, authors and designers are suing AI frontier models for using their content without permission or citation. As of November 2023, the U.S. Copyright Office position on AI-generated work:*
Only the human-created portion of the work is valid for copyright registration.
-
- Trademark due diligence should not change.
- Proceed with caution when inputting company trade secrets or proprietary information.
- Citing and checking sources remains critical.
*The U.S. Copyright Office update on usage and protections expected in Summer 2024.
Deep Fake: AI-generated media, such as images, videos, or audio, that convincingly replaces an existing person with another likeness, often used maliciously but also having potential positive applications. Deepfakes for good: positive applications are being used for creating personalized content, enhancing storytelling, and developing engaging marketing campaigns.
Ethical AI (in Marketing): The growing importance of ensuring that AI applications in marketing are transparent, accountable, and unbiased, fostering trust among consumers and stakeholders.
Expertise Paradox: The idea that as AI takes over tasks, deep human expertise becomes more valuable. (Ethan Mollick)
Few-Shot Learning: An AI technique that allows models to learn from a small number of examples, reducing the need for extensive training data.
Generative AI: A category of AI that’s focused on creating new content such as text, art/images, videos, and music. GenAI is transforming content creation strategies rapidly. (See Human-in-the-Loop)
Generative Pre-Trained Transformer (GPT): An AI model that generates human-like text based on patterns learned from vast amounts of data, such as models like ChatGPT and others. It can write essays, answer questions, or create dialogue, such as drafting an email. Many companies create their own GPT models using specific data to ensure privacy and relevance to their needs.
Hallucination: A phenomenon where an AI system generates content that appears convincing but is not based on factual information, leading to false or nonsensical outputs.
Human-in-the-Loop (HITL): A critical aspect of AI where humans are involved in the decision-making process, providing feedback and oversight. (Ethan Mollick)
Image Recognition: AI’s ability to identify and process images to detect objects, places, people, and other items.
Large Language Model (LLM): a subset of Generative AI (GAI), LLMs are specifically focused on text and designed to understand and generate human language. Applications: content generation, customer support automation, language translation, and natural language understanding (no code needed).
Machine Learning: A subset of artificial intelligence where computers learn to perform tasks by analyzing data and finding patterns, rather than being explicitly programmed with rules.
- Deep Learning: a subset of machine learning that uses multi-layered neural networks (a powerful brain-like structure) to process information, learn and make decisions. This powers many advanced AI applications.
Multi-Modal Model: They key difference between single- and multi-modal models has to do with the ability of each to handle different formats.
Single-modal model: You type a question in a text field and receive the answer in text (Text-to-text)
Multi-modal model: Has a more advanced capability to interpret and output in multiple formats including image-to-image, image-to-text, voice-to-text
Natural Language Processing (NLP): AI focused on the interaction between computers and humans through natural language.
Predictive Analytics: The use of AI to analyze current and historical data to make predictions about future events, aiding in strategic sales and marketing decisions.
Prompt: This is the input you give to an AI model so that it can address your question or problem you need solved. You can get very far down the road with a general, conversational prompt but the better your prompt, the better your answer. Results change 10X when you give your AI agent focus and context (aka Prompt Engineering). Better prompts also reduce false responses (aka hallucinations).
Red team/Red teaming: A process where a group of experts deliberately challenges an AI system to identify potential flaws, vulnerabilities, or unintended behaviors. This practice aims to improve the system’s safety, security, and robustness before deployment.
Reinforcement Learning from Human Feedback (RLHF): A technique used by AI developers to train their models; reinforcing good responses from the model and providing regular feedback to course correct along the way. Trained to respond in alignment with human preferences and values.
Sentient: You hear this word a lot when it comes to Artificial General Intelligence (AGI). It’s associated with a new future intelligence that mimics human consciousness and our ability to think and feel autonomously.
Synthetic Media: A broader term encompassing AI-generated content across various media types, including text, images, videos, and audio, which is increasingly being used in marketing and advertising.
Transfer Learning: The more you use an AI model, the better it learns–knowledge gained from one task is applied to a related task, enabling faster learning and improved performance.
Tokens: Tokens are snippets of text from a single letter to an entire phrase that AI systems use to process and understand text data such as words, phrases or symbols. The amount of interaction you can have with an AI model is based on the number of tokens afforded by the model or your budget (they vary).
Why tokens matter: you can run out of “time” on the AI model you’re working with, so be mindful of the length of your prompts and the attachments you upload.
Token counts happen behind the scenes of an AI model and you may not know you’re running low until the Chat model alerts you that you have “7 messages remaining” or similar. It may also tell you what time you can return. I have seen a visible token counter used on ChatGPT4O but it does not seem to be baked in–the counter comes from a plug-in or other integration.
Pro Tip: In longer sessions, ask the AI to summarize the session and paste it somewhere handy. Then if you run out of tokens, you can paste that summary into a new session with another AI model.
Voice-to-Text: AI technology that converts spoken language into written text, gaining new attention due to rapid advancements and new applications in AI models.
(As of 7/8/24)