While the term “AI” has been around since the 1950s, it has come into prominence around 2022 due to recent rapid advances in machine learning. AI breakthroughs are having a profound impact on every area of our lives. AI today has truly gone mass. Here is a guide from Microsoft to help you understand key AI terminology and join the global conversation.
- artificial intelligence
Artificial intelligence (AI) is a super-smart computer system that can mimic human tasks, such as understanding language, making decisions, translating languages, analyzing sentiment, and learning from experience. AI processes vast amounts of data through algorithms to create models that automate tasks that require human intelligence. Sometimes we interact with it directly, such as asking Bing Chat for help. But AI often works in the background, suggesting words, recommending songs, providing related information, and more.
- Machine Learning
Machine learning (ML) is a subset of AI and a method for achieving it. In ML, computer systems learn to identify patterns and make predictions by repeatedly running data through algorithms with different inputs and feedback. This is much like practicing piano scales millions of times to sight-read music. ML is ideal for solving complex problems like image recognition and language translation that are difficult to solve using traditional programming.
- Large-scale language models
Large-scale language models (LLMs) use machine learning to mimic human communication. Based on neural networks (NNs) inspired by the human brain, they are trained on large amounts of text to learn linguistic patterns and relationships. LLMs can translate languages, answer questions, summarize text, and even write stories, poems, and code.
- Generative AI
Generative AI goes beyond simply repeating something that already exists or providing information about what already exists: it harnesses the power of large language models to create something new. It learns patterns and structures and generates similar but new things. It can create images, music, text, video, code, and more.
- Hallucinations
Generative AI can create stories, poems, songs, and more, but it can also produce inaccurate responses that are called “hallucinations” or “fabrications” because it can't distinguish between real and fake information. Developers are addressing this issue by “grounding” the AI with additional information from trusted sources to improve its accuracy.
- Responsible AI
Responsible AI ensures that systems are safe and fair at every level: machine learning models, software, user interfaces, application access rules, etc. This is crucial for systems that make important decisions in areas like education and healthcare, where human biases can be reflected in the training data. A key aspect is understanding the training data and mitigating bias to better represent society as a whole.
- Multimodal Model
Multimodal models can process different data types simultaneously, such as images, audio, and text, and combine this information to perform tasks such as answering questions about the images – the ultimate in multitasking.
- prompt
Prompts are instructions in language, images, or code that tell an AI what task to perform. Engineers and users must carefully design prompts to get the desired results from large language models, just like when you order a sandwich, you specify details to get exactly what you want.
- First Officer
Copilot is a digital assistant that helps with tasks like writing, coding, summarizing, and search across a range of applications. Enabled by a large language model, Copilot understands natural language and assists with decision-making and data analysis. It's built with responsible AI guardrails for safety and security: like a co-pilot on an airplane, it assists but keeps you in control.
- Plugin
AI plugins work similarly to apps on your smartphone, allowing AI applications to meet specific needs without modifying the core model, enhancing AI's capabilities by making it easier to interact with other software, access new information, perform complex calculations, and interface with other programs, integrating AI more deeply into the digital ecosystem.
