Introduction to AI and Machine Learning

Study Notes

AI enables improvements in healthcare, overcoming physical limitations, developing smart infrastructure, and creating entertainment experiences.
AI is software replicating human behaviors and capabilities.
Key AI workloads include machine learning, computer vision, natural language processing, document intelligence, knowledge mining, and generative AI.

Machine learning is the foundation of many AI solutions.
It involves training computer models on data to make predictions and conclusions.
It combines computer science and mathematics.
A real-world example is sustainable farming using data and machine learning for informed decisions on weather, soil, and plant conditions.

AI learns from data, such as text messages, emails, social media, photos, and videos.
Millions of these data points are generated from sensors in homes, cars, infrastructure, and factories.
Data scientists use the data to train models, identifying relationships to make predictions and inferences.

Azure provides a cloud-based platform for machine learning models.
Features include automated, effective machine learning models (no-code) and graphical interfaces.
Visualization tools optimize and analyze experiments.
Jupyter notebooks allow for direct code execution.

Computer vision is an area of AI that deals with visual processing.
Examples include the Seeing AI app for the blind and low-vision community.
Computer vision uses machine learning to classify images based on their contents (classification).
It can identify objects within images and their locations using bounding boxes (object detection).
Advanced techniques like semantic segmentation classify individual pixels to create context and analyze images in more detail, possibly extracting descriptions.

NLP is an area of AI that enables software to understand human language.
It includes text analysis, speech interpretation, translation, and other tasks.
Azure's AI Language service is used for NLP solutions to better understand and analyze text and commands.

Document intelligence involves processing various document types (contracts, medical records) for insights.
Knowledge mining organizes large amounts of unstructured data into searchable knowledge stores.
Tools like Azure AI Document Intelligence automate document data extraction.

AI should solve problems fairly, reliably, safely, while respecting privacy and inclusivity.
Developers should be accountable for ethical AI applications.
Examples of risks are bias in data that affect results, system errors, and the potential for misuse.
Important considerations in responsible AI include accuracy, fairness, safety, security, and privacy of data.

Weak AI is designed for specific tasks, such as virtual assistants or chatbots.
Strong AI, also known as Artificial General Intelligence (AGI), is envisioned as having human-level intelligence and self-awareness but has not yet been achieved.
The Turing test is used to try to distinguish between human and AI reasoning ability. AI that successfully fools an interrogator into thinking it is a person would be considered intelligent.

Top-down approaches try to model human reasoning to solve a problem, and include symbolic reasoning.
Bottom-up approaches model the brain's structure, focusing on simple units interacting (neural networks).
Other methods include emerging or complex behaviors from simpler agents and evolutionary approaches (using genetic algorithms).