Domain 1: Understanding the Foundation of AI

Compute Infrastructure

.algorithmic innovation has spurred on AI.compute; storage and network, software development.CPUs & GPUs.serverless flow.four general stages: ingestion, preparation, training, & output (inference)

OECD Framework for AI systems

1)People and planet - looks at AI system and identifies individuals/groups that may be affected by the system. How the system impacts human rights, environment, and society. Privacy is part of this dimension 2)Economic context - AI system looked at according to the economic and sectoral environment in which it operates. Sectors could be financial, healthcare, education, etc.. Looks at business function (model). How it is deployed, effect of deployment, and scale of system. Also looks at maturity of the AI system 3)Data and input - what type of data was used to train the model, whether or not expert input was used. How was the data collected, what method was used, was it collected by machine or human, structure of data. 4)AI model - technical type, how model is built, how model is used 5)Tasks and output - tasks that the AI system performs, outputs, and actions that come from those outputs.

Use Cases

1)Recognition- image, speech, face recognition (ex. Teach manufacturing tools to see defects, plagiarism detection) Event Detection - fraud detection, sports video, cyber events 2)Forecasting - predict sales, revenue, demand (ex. Ride sharing apps; when is there the most demand?, weather forecasting) 3)Personalization - unique customer profiles get created for users. Based on previous visits, AI generates a customized view for interacting with websites 4)Interactive support - chat bots 5)Goal-driven optimization- optimize a particular problem, find lots of solutions to the problem, optimize driving time (bus routes, maps, etc.)

Gen AI

A field of AI that uses deep learning trained on large datasets to create new content, such as written text, code, images, music, simulations and videos. Unlike discriminative models, Generative AI makes predictions on existing data rather than new data. These models are capable of generating novel outputs based on input data or user prompts.

Robotics

A multidisciplinary field that encompasses the design, construction, operation and programming of robots. Robotics allow AI systems and software to interact with the physical world.e.g. Roomba - using ML to navigate a building.AI in manufacturing and robotics: "Fourth Industrial Revolution".increased connectivity and smart automation: potential exponential improvement of robotic processes.robotics stem from engineering and computer science.to design machines that perform tasks (often quite specific) without human intervention."machine perception" as an evolving field of potential convergence between robotics and AI: uses data that mimics human senses through cameras, microphones, pressure sensors, 3D scanners, motion detectors, and thermal imaging.example in food processingRPA = Robotic Process Automation:.software robots to automate repetitive and rule based tasks within business processes.mimic human actions on digital systems: data entry and form processing.enhanced by natural language processing and machine learning

Transformer Model

A neural network architecture that learns context and maintains relationships between sequence data, such as words in a sentence. It does so by leveraging the technique of attention, i.e. it focuses on the most important and relevant parts of the input sequence. This helps to improve model accuracy. For example, in language-learning tasks, by attending to the surrounding words, the model is able to comprehend the meaning of a word in the context of the whole sentence.

Natural Language Processing (NLP)

A subfield of AI that helps computers understand, interpret and manipulate human language by transforming information into content. It enables machines to read text or spoken language, interpret its meaning, measure sentiment and determine which parts are important for understanding.

Socio-technical system

AI influences society, and society also influences AI

Reinforcement Learning (type of ML)

AI system is rewarded for performing a task well and penalized for not performing well. Over time, learns to maximize rewards Ex. Self driving cars. System is rewarded for keeping car on the road, penalized if car goes off the road or hits something

Artificial General Intelligence

Also known as Strong, Deep or Full AI. This type of AI is intended to closely mimic human intelligence. AGI has been a goal of AI development for decades but, as of today, it remains beyond our reach. Experts expect AGI systems will do the following things at a level that is similar to or on par with human capabilities: Have strong generalization capabilities Be able to think, understand, learn and perform complex tasks Achieve goals in different contexts and environments

AI

Machines performing tasks that normally require human intelligence Definitions of AI vary, but common elements include - technology (use of technology and specific objectives to achieve), autonomy (level of autonomy to achieve objectives), human involvement (need for human input to train technology and outline objectives), output (performing tasks, solving problems, producing content)

Linear and Statistical Models

Model the relationship between two variables (ex. A linear regression model may be used to determine how sales of a product are related to changes in pricing based on historical data). This is not a black box algorithm, so they are more explainable

Decision tree model

Predict an outcome based on a flowchart of questions and answers. These are also more explainable and not a black box. However, changing a little bit of the training data can have a big effect on the algorithm, also subject to hacks

Semi-supervised learning (type of ML)

Semi-supervised learning models use a combination of supervised and unsupervised learning processes. This approach generally uses a small amount of labeled data and a large amount of unlabeled data. The aim is to leverage the benefits of both models, improving reliability while reducing costs. They are particularly helpful in scenarios where it is challenging to find or create a large, pre-labeled dataset. Image and speech analysis or categorization and ranking of web page search results are classic examples. Large Language Models, or LLMs, often rely on semi-supervised learning models. They are a form of AI using deep learning algorithms to create models trained on massive text data sets to analyze and learn patterns and relationships among characters, words and phrases.

AI Platform

Software used to develop, test, deploy, and refresh AI applications Ex. GCP, Microsoft Azure, AWS. Platforms can centralize data analysis, streamline development and production workflows, automate systems-development tasks, and monitor models and systems in prod

Supervised Learning (type of ML)

Supervised learning: labeled data grouped or classified into categories via the AI system Ex. One set of images is labeled "cats", and the other "dogs". System will look at sets of images and identify common characteristics of each (cats have whiskers, dogs have wagging tails and floppy ears) Used for text recognition and detecting spam in email 2 Subcategories: Classification - is the image a cat or dog? Regression - estimating a stock price

Machine Learning

The process of training machines to display AI behavior The branch of AI that leverages the use of data and algorithms to enable systems to learn and make decisions repeatedly. It can improve over time without being explicitly instructed or programmed to do so.

cross-disciplinary collaboration

UX, anthropology, sociology, linguistics experts should all be involved

Unsupervised Learning (type of ML)

Unsupervised learning: unlabeled data; typically used for pattern detection and finding outliers in the data, identify differences without help from humans Ex. banks may use this to look through transaction data and look for fraudulent behavior 2 Categories: 1) Clustering - groups data points that share similar or identical attributes (ex. looking for similarities in DNA strands) 2) Associated rule learning - identifies associations and relationships between data points (ex. consumer buying habits)

Deep Learning

Uses neural networks: contain nodes in a layered structure that continuously improve the ability to find the right answer, much like a human brain. Do not need to be trained to make non-linear, complex inferences in unstructured data. Black Box Commonly behind tech like facial recognition.

1956 Dartmouth Summer research project AI

Artificial intelligence, as a distinct field, was born in the U.S. in the summer of 1956 during a seminal conference at Dartmouth College, a research university in New Hampshire. Prior to the conference, various strands of research across disciplines like psychology, computer science, linguistics, and engineering formed the bedrock of what would eventually become AI. However, these efforts were mostly isolated, and the concept of an "intelligent" machine was not yet formalized. John McCarthy, assistant professor of mathematics at Dartmouth, and three senior researchers penned a proposal for the conference. They proposed that "every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it." The conference gathered leading researchers in fields relevant to AI. Attendees included Allen Newell and Herbert Simon, who introduced the Logic Theorist (considered by many to be the first AI program). It was less structured, with many brainstorming sessions that allowed ideas to flow freely. During this conference, the term "artificial intelligence" was adopted, effectively creating AI as a field of research. Participants left the conference with a collective sense of mission to develop machines capable of simulating human intelligence. The Dartmouth Conference was a defining moment in AI's history. It was here that the dream of creating intelligent machines was formalized into a unified academic field. This event marked the beginning of various AI pursuits, and many continue to influence AI research and development today.

Artificial Narrow Intelligence

Artificial narrow intelligence, or ANI, is designed to perform a single or a narrow set of related tasks at a high level of proficiency. These systems may seem intelligent; however, they operate under a narrow set of constraints and limitations, which is why this type of AI is commonly referred to as weak AI. While limited in scope, artificial narrow intelligence systems can help boost productivity and efficiency by automating repetitive tasks, enabling smarter decision making and optimization through trend analysis. A system designed to play chess is an example of artificial narrow intelligence.

AI Application

How an AI system is used Autonomous vehicles, chat bots, education, facial recognition, etc.