Core AI Concepts & Technologies
These are a few essential terms to understand regarding AI. Additionally, the NSF-supported Engage AI Institute has published a more substantial glossary of key terms which we strongly recommend as a helpful resource.
Algorithm
The step-by-step instructions that guide AI's problem-solving process. Algorithms are the recipes that tell AI systems how to process data and arrive at conclusions. They range from simple if-then statements to complex mathematical models.
Artificial Intelligence (AI)
(AI) "involves computational technologies that are mimicking, inspired by, or related to what we see from humans - how humans work, how humans think, and how they function. At a level [AI] attempts to augment or potentially even replace some cognitive tasks that human beings engage in." -George Siemens
Deep Learning
The brain-inspired approach to machine learning. Deep learning excels at complex tasks like image recognition and natural language understanding. It uses layered neural networks to process data in increasingly abstract ways.
Generative AI
Generative AI is trained on existing data (music, art, writing, code, etc.) and, extending from its pattern finding capabilities, can generate new content. ChatGPT is a well-known form of Generative AI that uses a Large Language Model to undertake natural language processing and natural language generation. In short, ChatGPT understands human language input and generates human-like text responses.
Intelligence Amplification (or Intelligence Augmentation)
"Intelligence Amplification exploits the opportunities of artificial intelligence, which includes data analytic techniques and codified knowledge for increasing the intelligence of human decision makers" (Wijnhoven, 2022).
Large Language Model (LLM)
Large language models are neural networks that have been trained on natural language data. The training of these models is self-supervised in that they are initially trained on an unlabeled data set to obtain their initial parameters before being tested and further trained with supervised or unsupervised tasks. Tasks in these latter stages might consist of sentence completion exercises (Think of a cloze-type task). LLMs have gained widespread interest through the popularity of ChatGPT and Bard.
Machine Learning (ML)
AI's ability to improve without explicit programming. ML algorithms get smarter with experience, much like humans do. This adaptive capability is what allows AI to tackle complex, ever-changing problems.
Neural Network
A web of interconnected nodes mimicking brain structure. Neural networks form the backbone of many modern AI systems. They're particularly adept at recognizing patterns and making decisions based on complex, multifaceted data.
Training Data
The educational foundation of AI systems. Training data is the information used to teach AI models how to perform tasks. The quality and diversity of this data significantly impact an AI's performance and potential biases.
AI Learning Methods and Approaches
RAG (Retrieval-Augmented Generation)
Enhancing AI responses with real-time information retrieval. RAG combines the generative capabilities of language models with the ability to fetch and incorporate relevant information from external sources. This approach improves the accuracy and relevance of AI-generated content.
Reinforcement Learning
AI learning through trial, error, and rewards. This approach mimics how humans learn from experience, with the AI receiving feedback (rewards or penalties) based on its actions. It's particularly effective in gaming AI and robotics.
Supervised Learning
A teaching method for AI where the “correct answers” are provided. In supervised learning, the AI is trained on labeled data, learning to associate specific inputs with specific outputs. It's widely used in classification and prediction tasks.
Unsupervised Learning
AI's way of discovering hidden patterns without explicit guidance. Unlike supervised learning, unsupervised learning works with unlabeled data, finding structure and relationships on its own. It's useful for tasks like customer segmentation or anomaly detection.
AI Applications & Specialized Use Cases
Chatbot
AI's conversational interface with humans. Chatbots use NLP to understand and respond to user queries, providing 24/7 customer service, assisting with bookings, or even offering therapy-like interactions.
Computer Vision
Giving machines the power to “see” and understand visual information. Computer vision enables applications ranging from facial recognition systems to self-driving cars that can interpret their surroundings.
CustomGPT
Tailor-made AI for specific tasks or industries. It's like having a personal AI assistant fine-tuned to your field's jargon and needs. A CustomGPT could be trained on legal documents to assist lawyers or on scientific papers to aid researchers.
Natural Language Processing (NLP)
The bridge between human communication and computer understanding. NLP makes possible everything from translation apps to voice-controlled devices. It's a complex field that combines linguistics, computer science, and AI.
Prompt Engineering
The art and science of crafting effective AI instructions. Prompt engineering involves designing and refining the text prompts given to AI models to elicit desired outputs. It's a crucial skill in leveraging the full potential of large language models.
Prompt Injection
A technique to manipulate AI behavior through carefully crafted inputs. Prompt injection involves inserting specific phrases or instructions into a prompt to override or bypass an AI's intended behavior. It can be used benignly to enhance AI capabilities or maliciously to exploit vulnerabilities.
Sentiment Analysis
The AI's emotional intelligence for text. Sentiment analysis uses NLP techniques to determine the emotional tone behind words, helping businesses gauge public opinion about their products or services from online comments and reviews.
AI Challenges, Ethics, and Testing
Bias
The unconscious tilt in AI decisions. Just as humans have prejudices, AI can inherit biases from its training data or creators. Recognizing and mitigating these biases is crucial for developing fair and ethical AI systems.
Explainable AI (XAI)
Making the black box of AI decision-making transparent. XAI aims to create AI systems that can explain their reasoning in human-understandable terms. This is crucial for building trust and accountability in AI-driven decisions.
Hallucinations
When AI spins fiction as fact. These convincing but fabricated outputs remind us to always verify AI-generated information. Hallucinations can range from minor inaccuracies to completely imagined scenarios or data.
Overfitting
When AI becomes too specialized in its training data. An overfitted model performs well on familiar data but fails to generalize to new, unseen information. It's like memorizing test answers without understanding the underlying concepts.
Synthetic Data
Artificially generated information mimicking real-world data. Synthetic data is used to train AI models when real data is scarce, sensitive, or difficult to obtain. It helps in preserving privacy and expanding limited datasets.
Turing Test
A classic benchmark for machine intelligence. Proposed by Alan Turing, this test evaluates a machine's ability to exhibit intelligent behavior indistinguishable from a human. While controversial, it remains a thought-provoking concept in AI development.
Underfitting
The opposite of overfitting, where AI fails to capture the underlying patterns. An underfitted model is too simplistic, missing important trends in the data. It's akin to using a linear equation to model a complex, non-linear relationship.