The principle that students should honestly represent their own work while using AI as a learning tool rather than a shortcut. AI misuse (e.g., submitting AI-generated content without disclosure) can be considered a form of academic dishonesty akin to plagiarism.
An autonomous system or program that perceives its environment and takes actions to achieve specific goals. For instance, a smart home thermostat agent monitors room temperature (perception) and adjusts heating or cooling (action) to maintain your desired temperature setting (goal).
Describes systems that can initiate actions or adapt behavior based on their understanding of the environment. For example, an agentic AI assistant might proactively suggest breaking up a long meeting into shorter sessions after noticing patterns of decreased participation in lengthy meetings.
The use of AI tools (e.g., Grammarly, Google Docs' predictive text) to refine writing, improve grammar, and enhance clarity without generating original content.
Software designed to identify AI-generated text by analyzing patterns in writing. These tools (e.g., GPTZero) are often unreliable and prone to false positives, sometimes misidentifying human-written work as AI-generated.
A structured document that requires students to transparently report how they used AI in their assignments, including which tools were used and for what purpose (e.g., brainstorming, editing, research).
The study and practice of ensuring responsible AI use in educational settings, including concerns about bias, hallucinations, plagiarism, data privacy, and academic integrity.
The field of computer science focused on creating systems capable of performing tasks that normally require human intelligence. For example, an AI system might help doctors identify potential tumors in X-ray images by analyzing patterns that it learned from studying millions of previous medical images, similar to how a human radiologist learns from experience.
A reasoning process where an AI model shows its work by generating intermediate steps before reaching a conclusion. For example, when solving a math word problem, the AI might first identify the key variables, then write out the equation, and finally solve it step by step, just as a teacher would encourage students to show their work.
A software application that uses AI to simulate conversation with human users. A common example is a customer service chatbot that can answer questions about order status, process returns, or help troubleshoot basic technical issues without human intervention.
The resources required to run and maintain AI models, including processing power, memory, and energy. For instance, training a large language model like GPT-3 can cost millions of dollars in computing resources, similar to how running a data center requires significant electricity and cooling costs.
A subset of machine learning using neural networks with many layers to model complex patterns. For example, a deep learning system can learn to recognize faces in photos by analyzing millions of images and understanding features at different levels - from simple edges and shapes to complex facial features.
An approach to exploring topics in great detail using AI-powered tools. For instance, a researcher studying climate change might use AI to analyze thousands of scientific papers, identifying patterns and connections that would take humans months to discover manually.
The capability of AI systems to process and interpret data for insights. For example, a retail AI system might analyze sales data to predict which products will be popular next season, considering factors like historical sales, weather patterns, and social media trends.
The principle of using AI technology responsibly and fairly. For example, when an AI system is used in hiring, it should be regularly tested to ensure it's not discriminating against candidates based on gender, race, or age, and decisions should be explainable and appealable.
The capability of AI systems to run programming code. For instance, an AI coding assistant might not only suggest code improvements but also run tests to verify the code works correctly, similar to how a programmer would test their code before deployment.
A machine learning approach where a model learns from just a few examples. For instance, an AI might learn to identify a new type of bird after seeing only three or four photos, similar to how a child can learn to recognize a new animal after seeing it just a few times.
The process of customizing a pre-trained AI model for specific tasks. For example, a general-purpose language model might be fine-tuned on medical literature to better understand and generate healthcare-specific content, similar to how a doctor specializes after medical school.
AI systems that create new content based on patterns learned from existing data. For example, a generative AI might create a new piece of music in the style of Mozart after learning from his compositions, or write a story in the style of Shakespeare after studying his works.
The capability of AI to create images from text descriptions. For instance, asking an AI to create "a watercolor painting of a sunset over mountains" will result in a unique image combining these elements in the specified style.
The process of using a trained AI model to make predictions or generate outputs. For example, when you speak to your phone's voice assistant, it uses inference to convert your speech to text and determine what action you want to take.
The degree to which humans can understand an AI's decisions. For instance, a loan approval AI should be able to explain why it denied a loan application by pointing to specific factors like credit score or debt-to-income ratio, rather than just saying "application denied."
A type of AI model trained on vast amounts of text data to understand and generate human-like text. For example, an LLM can write emails, answer questions, or even write code based on natural language instructions, similar to how a human might complete these tasks.
A branch of AI where systems improve through experience. For instance, an email spam filter learns to better identify junk mail over time by analyzing which emails users mark as spam, similar to how a person learns to recognize spam through experience.
A feature allowing AI systems to operate in real-time. For example, a live translation system that converts spoken words from one language to another during a video call, enabling real-time communication between people speaking different languages.
A method where computers learn from data without explicit programming. For instance, an ML system might learn to recognize cats in photos after analyzing thousands of labeled images, rather than being programmed with specific rules about what makes a cat look like a cat.
Standards used to measure AI performance. For example, a medical diagnosis AI might be evaluated on its accuracy (correct diagnoses), precision (avoiding false positives), and recall (not missing actual cases), similar to how a doctor's performance might be assessed.
A field focused on enabling computers to understand and process human language. For example, when you ask your smartphone's assistant "What's the weather like today?" it uses NLP to understand your question and provide a relevant response.
A computing system inspired by biological brains. For instance, a neural network might learn to recognize handwritten digits by processing millions of examples through layers of interconnected nodes, each layer learning to recognize increasingly complex patterns.
The protection of sensitive information in AI systems. For example, when an AI assistant processes your health data to provide personalized recommendations, it should encrypt the data, remove identifying information, and only use the data as explicitly authorized.
The input text given to guide an AI model's response. For instance, instead of asking "Write about dogs," a better prompt might be "Write a 300-word explanation about how dogs use their sense of smell, suitable for a 10-year-old reader."
The practice of crafting effective prompts for AI models. For example, a prompt engineer might discover that asking "Analyze this text for tone and key themes, then explain your reasoning step by step" produces better results than simply saying "What's this text about?"
An educational approach where students are encouraged to disclose their AI use without fear of penalties, allowing for open discussions about ethical AI engagement.
The ability of AI to process and understand document content. For example, an AI might read a lengthy legal contract and summarize the key terms and conditions, or extract specific information like dates, parties involved, and financial terms.
The process of making logical decisions based on information. For example, when an AI chatbot helps troubleshoot a computer problem, it reasons through possible causes and solutions, asking relevant questions before suggesting the most likely fix.
A learning method where AI improves through trial and error with rewards and penalties. For instance, an AI learning to play chess might receive positive rewards for winning moves and negative rewards for losing pieces, gradually developing better strategies through practice.
The ability of AI to process and understand visual information. For example, an AI system might analyze security camera footage to detect unusual activities or help visually impaired people understand their surroundings by describing what it sees.
The capability to analyze or generate video content. For instance, an AI system might analyze workout videos to provide real-time feedback on exercise form, or generate animated videos from text descriptions.
The distinction between the user-facing AI service and its underlying model. For example, ChatGPT is a service that provides a user interface and additional features, while GPT-4 is the underlying model that processes and generates the text.
The process of breaking text into smaller units for AI processing. For instance, the sentence "I love AI!" might be broken into tokens like ["I", "love", "AI", "!"], allowing the model to process each piece separately while maintaining their relationships.
The examples used to teach AI models. For instance, to train an AI to recognize different dog breeds, you might provide millions of labeled photos showing different breeds from various angles and in different lighting conditions.
A practice where students and faculty openly disclose when and how AI is used in academic work, fostering trust and responsible AI integration.
The ability of AI systems to retrieve online information in real-time. For example, an AI assistant with web access could provide current weather forecasts, latest news updates, or real-time stock prices as part of its responses.
This glossary is current as of February 8, 2025, and provides an accessible introduction to key concepts in AI, enhanced with practical examples for newcomers to the field.