AI Governance Professional Training Notes

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Planning Phase - Define Business Problem

Define the business problem 3 common business problems) 1. Classification: A problem that requires using an AI systems to classify data into different types - types A, type B, and potentially more 2. Regression: A problem that requires using an AI system to predict what an organziation should do in the future based on past data. 3. Recommendation: A problem that requires using an AI system to make a recommendation (eg viewer recommendations and product recommendations)

Module 2: AI Impacts on People and Responsible AI Principles Lession 1: Existing and Emerging Ethical Guidance on AI What foundational controls should be in place? Mitigating the risk ethical risk posed by the use of AI?

* Develop ethical principles of AI * Develop of cross-functional and demographically diverse oversight body * Assess whether appropriate policies and procedures exist for associated risks

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Planning Phase - Use Cases

* Do you have the right data to make your AI system usable? AI systems are all about data. If you don't have the right, enough, or accurate date, the system will not perform well. * What type of data is accessible to you and usable? * Do you need to look for new data?

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Design: Data Formats

Data Gathering Considerations: Data Formats Structured vs unstructured * Structured: Labeled data, in some circumstances is categorized to identify data to be used (ie usually data that can go into a spreadsheet with rows and categories) * Unstructured: Unlabeled/uncategorized data: may need to be structured to put into the model (a large dataset that is just a collection of images) Static vs streaming * Static: data that does not change Data is put into the model and stays the way it is (ie. historical data) * Streaming: data that will change (ie. data about customers visiting a website that changes everytime they visit)

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Design: Wrangling/Preparing the Data: Data Minimazation

Data Minimazation The concept that if you do need the data for your specific application, you should not use it to train your model or as input Once again, for privacy, not including personal data will make the system more protective pf the individuals privacy

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Implementation Phase: Changes to the model

Changes to the model * Over time the model could change due to data changes * Because of the complexity of the environment in which is it implemented and the potential for data to change as the model is used, monitor and maintain the model to avoid model drift * Continue to iterate the model to improve performance as the data changes * Define a baseline to measure future iterations of the model as you iterate it * AI systems potentially require more attention than other types of systems

Module 2: AI Impacts on People and Responsible AI Principles Lesson 2: Core risks and harms posed by AI systems Characteristics of trustworthy AI

Characteristics of trustworthy AI * Operates in an expected, legal and fair manner * Accountability * Human-centric * Transparency

Module 2: AI Impacts on People and Responsible AI Principles Lesson 2: Core risks and harms posed by AI systems Company/Institutional Harms

Company/Institutional Harms * Reputational: loss of customers and renewals, etc. * Cultural: Assumption that AI is more correct than humans, so we are less likelyto challenge its outcomes even though its created by humans * Economic: Costs of internal resources, litigation costs, etc. * Acceleration: Not all risks can be anticipated, AI impact may be wider and greater than with other technolgoy solutions, etc. * Legal and Regulatory: Industry laws and regulations may apply to AI use, privacy law implications, etc.

Module 1: Foundations of Artificial Intellegen Lesson 3: AI Technology Stack Compute Infrastructure: Compute

Compute Look at through central processing units, CPUs and graphical processing units GPUs * GPUs have thrust the AI movement forward. They have specialzied chips that offload from CPUs * Provide better performance and match to algorithmic advances * Better in nmatching hardware to the AI model for optimal performance

Module 1: Foundations of Artificial Intellegen Lesson 3: AI Technology Stack Compute Infrastructure

Compute Infrastructure Main area of infrastructure: compute, storage and network, software development.

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Implementation Phase: Continuous Monitoring

Continous Monitoring * Begins BEFORE deployment * contiuously monitor how the model is performing Monitor for: * Deviations in accuracy * Irregular decisions * Drifts in data that might effect the performance of the model

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Design: Wrangling/Preparing the Data: Data Cleansing

Data Cleansing * Removing erroneous or irrelevant data from the data sets * Some of the data may not be needed for the AI system and should be eliminated * Also removed inaccurate data * If personal data is in the data sets and is not needed for the AI model, remove it so it will not cause privacy isses later.

Module 1: Foundations of Artificial Intellegence Lesson 3: AI Technology Stack Common AI Models: Decision Tree Models

Decision Tree Models * Predicts an outcome based on a flowchart of questions and answers * Explainable and not a black box * Disadvantage: changing the training data (even in a small way) can significantly impact the aglorithm; sugject to security attacks and hacks.

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Robotics and AI: Fourth Industrial Revolution or Industry 4.0: AI in manufacturing and robotics

Fourth Industrial Revolution or Industry 4.0: AI in manufacturing and robotics * Next stage of industry and manufactuing advancements, increased interconnectivity and smart automation * Robotics stems from engineering and computer science * Aims to design machines that can perform tasks without human intevention; typically, the taks are very specific. * AI introduces efficiences and effective pathways enabling the exponential improvement of robotic processes.

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Fuzzy Logic

Fuzzy Logic * A method of reasoning intended to mimic or resemble human decision making. * Convensional logic used in computing, also known as crisp logic, generally takes the form of precise inputs and outputs, often binary in nature, such as true or false, yes or no. * enables a range of possible inputs to achieve an output; allows for a situation where a statement can be partially false, and provides a method to represent uncertainty and vagueness in decision-making.

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Fuzzy Logic Systems

Fuzzy Logic Systems * Employ fuzzy logic inference mechanisms to make decisions based on the fuzzy rules and input data * These systems follwoing four standard steps: 1. Fuzzification: input data is converted to fuzzy data sets 2. Rule evaluation: determines the degree of matching between the rules and input data 3. Aggregation: rule outputs are combined 4. Defuzzification: the process through which fuzzy outputs are converted back into specific values Examples: climate control systems, image recognition systems, traffic management systems.

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Use Case and Benefit of AI - Goal Driven Optimization

Goal Driven Optimization: * Used to optimzie a particular proplems and find solutions (ie optimize a supply chain). If you are having supply chain issues and want to get a product out out faster.

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI What is AI?

Hallmarks of human intelligence: ability to think creatively, consider various possibilities, and keep a goal in mind while making short term decisions

Module 1: Foundations of Artificial Intellegen Lesson 3: AI Technology Stack Compute Infrastructure: High Performance Compute

High Performance Compute Create isolated clusters of compute power; high-speed networking; specialized chipsets

Module 2: AI Impacts on People and Responsible AI Principles Lesson 2: Core risks and harms posed by AI systems How do we achieve trustworthy AI?

How do we achieve trustworthy AI? * Embed trustworthy AI as part of the operating model * Ensure that the org and the AI are following the stated processes * Confirm AI systems are safe and secure * Ensure the integrity of the AI * Make sure AI enables human oversight and promotes human values

Module 2: AI Impacts on People and Responsible AI Principles Lesson 2: Core risks and harms posed by AI systems Operationalizing responsible AI practices

Operationalizing responsible AI practices * Understand where AI is used and its role in the organization * Set clear technical standard that are shared and adhered to * Develop AI playbooks * Update internal legal organizational structures to reflect new roles and responsiblities

Module 1: Foundations of Artificial Intellegen Lesson 3: AI Technology Stack Compute Infrastructure: Quantum Computing

Quantum Computing Processing data in three dimensions horizontally and vertically

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Use Case and Benefit of AI - Recognition

Recognition: Typically image speech or facial recognition. Facial recognition: determining if an individual's face can be matched to another picture of that individual. * Retailer or product matches: Sending a picture of a desired product to a retailer's online system. The system looks for a product match based on the description of the picture received, then notifies the consumer of product matches. * Manufacturing machines to see defects that impact product development. * Plagiarism detectors, often used in education.

Module 2: AI Impacts on People and Responsible AI Principles Lesson 2: Core risks and harms posed by AI systems Values and Opportnities of AI

Values and Opportnities of AI * Can be faster and more accurate * Helping with medical assessments and legal predicitions * Processing huge volumes and a wide variety of data * Automation of processing * Accelerating mundane and repetitive tasksHow to

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Use Case and Benefit of AI - Recommendations

Recommendations * Product recommendation or viewing recommendation for customers based on predictive analytics * Can also be used for decision support systems. AI can help human make better decisions in general. For example, AI can help health care providers make diagnoses based on past information about similar types of diseases, symptoms and previous diagnoses. * Government use for adjudicating disability cases: trying to figure out the best way to give an individual access to their benefits for disability cases.

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Reinforcement Learning Models

Reinforcement Learning Models * Uses a reward and punishment matrix to determine a correct optimal outcome * Rely on trial and error to determine what to do or what not to do; rewarded or punished accordingly * Do not ingest pre-labeled data sets; learning is solely through action and repetition, changing state or getting feedback from their environment. * Actions and decisions that result in a reward reinforce the triggering behaviour

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Robotics and AI: Robotic Process Automation (RPA)

Robotic Process Automation (RPA) * Evolving technology that uses software robots to automate repetative and rule-based tasks within business processes. * Designed to mimin human actions on digital systems (e.g. data entry and form processing) * Natural language processing or machine learning enhances RPA robots' automation capabilities.

Module 1: Foundations of Artificial Intellegen Lesson 3: AI Technology Stack Common AI Models: Robotics

Robotics Multidisciiplinary field encompassing the design, construction operation and programming of robotics. * Allows AI systems and software to interact with the physical world without human intervention. Example: Roomba using machine learning to navigate a building.

Module 1: Foundations of Artificial Intellegen Lesson 3: AI Technology Stack Compute Infrastructure: Software-Role of Open Source AI

Role of Open Source AI * A maker culture that allows organizations great freedom to innovate with AI * Creates its own massive feeback loops that drive the free spread of ideas for applying AI transformation, tuning best practices and turning ideas into viable businesses or assets for organizations

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI The Cycle of AI

* First AI Summer - Mid 1950s - mid 1970s (AI research labs established at top universities. 1st AI programming language LISP (John McCarthy). ELIZA developed at MIT; example of early natural language processing. * First AI Winter - Mid 1970s - 1980s: A period of skepticism, funding cuts and critiques * Second AI Summer - Mid 1980 - Late 1980s: renewed intereste in AI. Expert and computer systems emulated human decision making ability. 5th generation computer systems project to develop AI-powered computers (Japanese government) * Second AI Winter - Late 1980s - Late 1990s: decreased interest and funding. High cost of maintaining expert systems; end of cold war. * Renaissance and the era of big data - Late 1990s - 2011: turning point for AI. IBMs Deep Blue defeats world chess champion in 1997. Emergence of the internet. Beginning of the big data era. Advancements in computational power and machine learning improved AI capabilities (i.e. recommendation algorithms for shopping and voice assistants in Smartphones) * AI Boom 2001 - Present Advancements i-3 shown deep learning: subfield of AI invlicing training neural networks data. Victories in Googles AlphaGo over Go world champion (2016); Open AI's GPT-3 showing capabilities of language models.

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI The history of AI and the evolution of data science: The emegence and growth of data science

* Foundations - 1960s - 1980s * Age of databases - 1980s to 1990s * Advent of the internet (growth of data/data mining) - 1990s - 2000s * Rise of data science - 2000s - 2010s * Era of big data - 2010s - present

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Grounding AI Governance: The How "Frameworks"

* International Organization for Standardization (ISO) (Several may appy * National Institute of Standard and Technology (NIST) AI Risk Management Framework * Institute of Electrical and Electronics Engineers (IEEE) 7000-21 * Human Rights, Democracy, and the Rule of Law Assurance Framework for AI System (HUDERIA) - Council of Europe * Other Standard specific to juristiction/industry * See IAPP's Global AI Legislation Tracker

Module 1: Foundations of Artificial Intellegence Lesson 3: AI Technology Stack Common AI Models

* Linear and Statistical Models * Decision Trees * Robotics * Machines Learning Models * Neural Networks * Computer vision model * Speech recognition models * language models * Reinforcement learning models

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Grounding AI Governance: The Why "Principles"

* OECD AI Principles / Guidelines * Fair Information Principles FIPs * UNESCO: the Recommendation on the Ethics of Artificial Intelligence Using one or more established sets of principles helps organizations identify their own ethical principles of AI.

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Use Case and Benefits of AI

* RDFPIGR * Recognition * Detection * Forecasting * Personaliztion * Interaction Support * Goal Driven Optimization * Recommendation

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Use Case and Benefit of AI - Interaction Support

* Virtual assistants or chatbots that assist customers with transactions. Commonly used in private industry * Used in public sector as well, chatbot sometimes asisst students applying for government student loans such as FAQs

Module 2: AI Impacts on People and Responsible AI Principles Lesson 2: Core risks and harms posed by AI systems Group Harms - Facial Recognition

*Facial Recognition Algorithms: Many AI Systems using face recognition exhibit demographic differentials (the abilitity to match two images of the same person vary from one demographic group to another) * A NIST study found AI facial recognition systems to be unrealiable across many kinds of systems * Studies found those with darker skin tones and females are much more difficult to recognize, leading to discrimination and bias. * Mass suvelillance: A large potentilal harm particularly marginlaized groups * Civil rights: Harms to freedom of assembly and protest due to tracking and profiling individuals linked to certain beliefs or actions * Deepening of racial and socio-economic divides

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI 3 High-Level Categories of AI

1. Artifical Narrow Intellegence (ANI) 2. Artifical General Intellegence (AGI) 3. Artifical Superintelligence (ASI)

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI 3 Main Elements of Expert Systems

1. Knownlege Base 2. Inference Engine 3. User Interface

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Organization for Economic Cooperation and Development (OECD)'s 5 Dimensions to Classify AI Systems

1. People & Planet: Identifies individuals and groups that might be affected by the AI system (ie human rights, the environment, and society). Privacy comes into play here. 2. Economic Context: The AI system is looked at according to the economic and sectoral environment in which it operates. Characteristics include: the sector where the AI sysetem operates (financial, health care, education), the business function or model for the AI system, necessity of he AI system to operations, how it is deployed and the impact of the deployment, scale of the systems, technological maturity of the AI system (a new system may not have been tested as much data data over time; more mature systems may be more effective) 3. Data and Input: What type of data was used in the model and any exper input used. *Expert input is human knowledge that gets codified into rules * Includes characteristics such as how data was collected and what collection method was used (by machine or by human), structure of the data and data format. 4. AI Model: Discusses the technical type; how the model is built and used. 5. Task and Output: Tasks that AI systems perform, its outputs and resulting actions from those outputs. *Characterstics include system tasks, systems that combines tasks and actions, evaluation methods used to look at how tasks and systems perform.

Module 4: Implementing Responsible AI Governance and Risk Management Lesson 1: Interoperability of AI Risk Management Risks that AI algorithms and models pose: Three broad categories

1. Security and operational risk 2. Privacy risk 3. Business risk

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI 3 Machine Learning Models

1. Supervised 2. Unsupersived 3. Reinforcement

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Types of Machine Learning

1. Supervised Learning 2. Unsupervised Learning 3. Reinforcement Learning

Module 1: Foundations of Artificial Intellegence Lesson 3: AI Technology Stack AI Platforms and AI Applications

AI Platforms: software that allows an organization to develop, test, deploy, and refresh AI application. * Examples include: Google Cloud Platform, Microsoft Azure and Amazon Web Services * Platforms can: * Centralize data analysis * Streamline development and production workflows * Faciltate collaboration * Automate systems-development tasks * Monitor models and systems in production AI Applications: refers to how the AI system is used. Some common applications include: E-commerce, Education, Health Care, Autonomous vehicles, Navigation, Facial Regonition, Robotics, Human Resources, Marketings, Social Media, Chat Bots, Finance.

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Modern Drivers of AI and Data Science

AI and data science have been fuelled by tech megatrends: * Cloud computing: on-demand, scalable computing resources, high-powered computing accessible to everyone; drives AI development and data processing capabilities. * Mobile technology and social media: Proliferation of Smartphones and rise of social media platforms have led to a massive increase in data, AI models learn from this information. * Internet of things (IoT): IoT devices generate data that feeds into AI models, contributing to data science * Privacy-enhacing technologies (PETs): a viable approach to data security and privacy concerns; ensures continues responsible growth of AI and data science. * Blockchain: Blockchain technology provides a trusted interface for secure financial transaction, enhances data privacy and security in certain contexts; not universally applicable to every data privacy and AI challange. * Computer vision, AR/VR, and the Metaverse: Emerging technolgies that shape the digital landscape of AI and Data Science. * Computer Vision: Enables machines to understand the world through images and videos; creates safer more efficent, interactive human-machine interactions; transformed how AI interprets and processes visual data (ie. health care, autonomous vehicles, AI) * AR/VR: redefines how we interact with digital content. AR overlays virtual objects onto the real word. VR immerses users in entirely simulated environments. Applies to diverse fields (i.e. gaming, therapy) * The Metaverse: represents a vision of a shared virtual space where individuals can interact, conduct business and explore endless possibilities. Maybe ahead of its time.

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Supervised Learning

Labeled Data that is grouped or classified into categories via the AI system. Used for text recognition, detecting spam in email, etc. (i.e. for email filtering, the algorithm is training using a labeled dataset containing both span and legitimate emails. It extracts the relevant information to create patterns to predict whether future emails are spam or legitimate.

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Considerations: AI as a socio-technical system

AI influences society and Society influences AI *Strong relationship between the technology and human beings * Relavent stakeholders to consider when working with AI: * Individuals who look at the broader societal influences of AI, such as anthropologists, sociologist or others who works in social sciences. * individuals to develop and implement AI systems * Risks in the use of AI * AI systems are implemented in vast complex environments * The data used and AI will change over time. * The model may need to be changed, upgraded, or both, to reflect what is being done with the new data in the environment.

Module 2: AI Impacts on People and Responsible AI Principles Lesson 1: Existing and Emerging Ethical Guidance on AI AI-specific ethical considerations

AI-specific ethical considerations * 2019: OECD AI Principles * 2021: White House Office of Science and Technology Policy Blueprint for an AI Bill of Rights * United Nations Educational, Scientific and Cultural Organization (UNESCO) Principles * Asilomar AI Priciples * Institute of Electrical and Electronics Engineers (IEEE) Initiative on Ethics of Autonomous and Intellgent Systems * Commission Nationale de l Informatique et des Libertes (CNIL) AI Action Plan *Using these common features, companies can idetify their own ethical principles of AI and develop governance frameworks to ensure that AI is developed, produced and deployed withint the organizations ehtical boundaries and risk appetite.

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Reinforcement Learning

An AI system is rewarded for performed a task well and penalized for not performing it well. Over time, learing to maximize the reward and develop a system that works. An AI system is rewarded when it keeps a car on teh road and gets it to the destination where it is suppost to go. It is penalized if the car goes off the road or hits another object. the system learns over time to maximize the rewared, resulting in better performing self-driving car.

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Design: Wrangling/Preparing the Data: Anonymization

Anonymization One method for protecting privacy that involves removing identifiers from the data: name, Social Security number, phone number, address, anything that could indentify an individual Note: Completely anonymyzing data is difficult because individual can be identified in many ways combining data sets can potentially reindentify them.

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Artifical General Intellegence (AGI)

Artifical General Intellegence (AGI) * Strong, Deep or Full AI * Intended to fully mimic human intellegence * AGI remains beyond reach at present - moving closer * Experts expect AGI systems to have strong generalization capabliltiies, the abilitiy to think, understand, learn, and perform complex tasks and achieve goals in different context and enviroments. * Intended to closely mimic human intellegence * AGI remains beyond reach

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Artifical Superintelligence (ASI)

Artifical Superintelligence (ASI) * AI Systems with intellectual powers beyond those of humans across a comprehensive range of categories and fields of endeavor. * Capable of out performing humans self-aware, understanding human emotions and experiences and evoking its own, thuse experiencing reality like humans. * Like AGI, ASI does not yet exist.

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Artifical Narrow Intellegence (ANI)

Artificial Narrow Intellegence (ANI) a/k/a Weak AI Designed to perform a single or a narrow set of related tasks at a high level of proficiency (ie A system designed to play chess). Operated under a narrow set of contstraints and limitations. Boosts productivity and efficiency by automating repetative tasks. Broad Artificial Intelligence aka Strong AI: More advanced scope than ANI,. capable of performing a broader set of tasks * relies on a group of AI systems capable of working together and combining their decision making capabilities (eg autonomous driving vehicles). Lacks human-like capablities experts expect of AGI

Module 2: AI Impacts on People and Responsible AI Principles Lesson 2: Core risks and harms posed by AI systems Bias and Discrimination - Individual Harms

Bias and Discrimination - Individual Harms * Employment and hiring * Insurance and social benefit * Housing * Education * Credit * Privacy Concerns - Screen out personal data. If you don't need personal data it should not be used in the system. Deidentification: removing identifiers from the data such as name, address, SSN, however, it is possible to reidentify an individual if data is aggregated or combined with another data set. * Appropropriation of personal data for model training. Models are bing trained in AI from large soures of data. Data my come from social media or large datasets with information individuals may have consented for one particular use of their data but not for trianing an AI system. * Inference: An AI system model makes predictions or decisions. In some cases the systems can be used to indentify individuals but they are not always accurate. Personal data can be attributed to the wrong person. * Lack of transparency of use: AI systems should notify individuals when AI is being used ( chatbots) * Inaccurate models: Accuracy of data is very important; AI systems are only as good as the data that trains them. * Economic opportunity and jobn loss: while AI can create some opportunities for jobs, it also has the potential to affect job loss

Module 2: AI Impacts on People and Responsible AI Principles Lesson 2: Core risks and harms posed by AI systems Bias in AI Systems

Bias in AI Systems Can cause harm to a persons civil liberties, rights, safety and economic opportunity, individuals developing the systems can have bias; this should be addressed during the life cyle of AI system development. Examples: * Implicit Bias: Discrimination or prejudice toward a particular group or individual. * Sampling Bias: Data gets skewed toward a subset of a group and therefore may favor that subset of a larger group * Temporal Bias: A model is trained and functions properly at the time but may not work well at a future point requiring new ways to address the data * Overfitting to training data: Data A model works for the training data but does not work for new data because it is too fitted to the training data * Edge cases and outliers: Any data outside the boundaries of the training dataset (e.g. edge cases can be errors when you have data that is incorrect, duplicative or unnecessary * Noise: data that negatively impacts the machine learning of the model * Outliers: Data point outside the normal distribution of the data, can impact how the model operats and its effectiveness.

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Implementation Phase: Deploying the model

Deploying the Model Involves transitioning from a development and testing evironment to a real-world, operational setting to be used for its intended purpose. Choosing deployment environment: i.e. the infrastructure or platform for the model Three most popular environments: 1. Cloud-based: 3rd pary cloud provider 2. On-Premise: host on servers and hardware owned by org 3. Edge: hosting the model on "edge" devices like smartphone. this option may have decreased latency and greater data privacy, however, the model may be limited by the edge device's hardware, thereby limiting the model's computational power. Packaging the Model into a format that allows it to be deployed. A common option is "Containerization" which involves packaging the model and dependencies (ie. everything the model needs to run effectively) into a self-contained unit. Containers can help reduce compatiblity issues and make is easier to deploy the model in different environments (e.g. development or testing) Making the model accessible for rea-world use (also called exposing the model), allowing systems or applications to interact with the model. Many options exist, including REST APIS and embessing into an application. The best option depends on many factors including budget, IT expertise, resources, the model's purpose and computational needs and the type of data the model processes.

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Design

Design The design phase includes implementing a data strategy, including gathtering data collection. Data is critical for an AI system. You must have the right data for the AI system to work well. Garbage in / garbage out

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Use Case and Benefit of AI - Detection

Detection: Credit card transaction fraud detection or fraud detection when applying for government services or benefits: looking for patterns of fraudulent behaviour within the system * Events and sports video: example: viewing at a particular activity such as a touch down or goal *Cyber events and systems mangement help organziations better respond to indicdents.

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Development Phase

Determing the system architechture * when selecting the model, choose an algorithm according to the desired level of accuracy and interpretablity of the data. * what do you want to learn from the data? * how is it going to help you solve your problems? * What are the other requirements and contraints? Work with Subject Matter Experts Use the same features for training and testing Avoid any uncessessy features

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Planning Phase - Governance Structure

Do you have a governance structure? Who implements and maintains it? Who writes policies and procedures? Who overseas developemnts and testing or selecting the AI system product? Who champions development and implementation?

Module 2: AI Impacts on People and Responsible AI Principles Lesson 2: Core risks and harms posed by AI systems Ecosystem Harms

Ecosystem Harms * When training several common large AI model, studie found that they emit more then 626,000 pounds of carbon dioxide (the equivalent of five times the lifetime emissions of an american car) * Another study found that when looking at the top 4 natual language processing models, the energy consumed over the training process matches the energy mix used my Amazon's AWS, the larger cloud service provider * An addtional study found that each casual use of generative AI is like dumping out a small bottle of water on the ground * To address this, many orgs are seeking alternatives to the use of electrical power * possiblitly using batteries to power susteems; this can also have an environmental impact (e.g. l;ithium extraction demands enormous water usage How can AI be used to help the environment? * Self driving cars developed by AI systems can help reduce emissions. * AI use in agriculture has produced higher yields * AI use in satellite images can help identify disaster-stricken areas so they can receive help * Weather forcasting

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Design: Wrangling/Preparing the Data: Data Cleansing Design Phase: Wrangling/Preparing Data: Data Labeling

Labeling includes tagging or annotating the data to identify what kind of data it is

Module 2: AI Impacts on People and Responsible AI Principles Lesson 1: Existing and Emerging Ethical Guidance on AI What are the key ethical issues for AI? Considerations

Ethical principles: * Lawfulness * Safety for people and the planet * Protection from unfair bias * AI use is transparent and explainable * Individuals have appropriate choices about the use of their personal information to develop AI * Individuals can choose to have human intervention in key AI-driven decisions that impact their legal rights or well-being * Organizations must be accountable fo rensuring aI they develop and ensure use is secure

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Expert Systems

Expert Systems * Mimic the decision-making abilities of a human expert in a specific field. * draws inferences from a specific knowledge base and relies on AI to replicate the judgement and behavious with a specific expertise. * Widely deployed across industies: financial services, heath case, agriculture, engineering * Designed to support and assist humans, rather than replace them (e.g. a medical diagnosis systems designed to aid doctos in determining the type and stage of a canerous growth)

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Development Phase: Feature Engineering

Feature Engineering Transforming data information useful representation (features) Purposes of effective feature engineering: 1. Improving model performance: Imporving AI model or pipeline performance is the most important purpose. Data scientist attempt to derive and structure datasets so a model can optimally learn the relationships of a feature to targets. Goals: Curating and creating a subset of features and providng the greatest predictive power for an AI model 2. Reducing computational costs: Decreasting computational and storage costs of models and improving latency for training models and making predictions. Reduced cost is due to fewer coputational requirements. Computaitonal effetiveness is improved through: reduding the number of features, and this the amount of data to process and store for training, reducing the number of features and data in an API call, etc. 3. Boosting model explainability: Model explainability/interpretability: degree to which someone can consistently predict a model's result; highly valuable and required in may AI use cases. Essential to help ensure fairness, privacy, reliability, robustness, causality and trust. It affects sitatuations where modesl can significantly impact users and the larger society directly or indirectly.

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Use Case and Benefit of AI - Forecasting (business forecasting)

Forecasting: (business forecasting) * predict sales and revenue, as well as potential product or service demand. * Ridesharing apps, determine when there might be a higher demand for rides; when demand is high, prices can increase * weather forecasting

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Expert Systems: Interface Engine

Inference Engine Extracts relevent information from a knowledge base and uses it to solve a problem. * used a rule-based approach that maps data from the KB to a series of rules which the system relies on to make decisions in response to the imput provided * Expert systems often include a module that allows users to review its decision making process.

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Expert Systems: Knowledge Base

Knowledge Base Typically consists of an organzied collection of facts and information provided by human experts and focused on a specific field or domain; system is also allowed to gatehr additional information from external sources.

Module 1: Foundations of Artificial Intellegence Lesson 3: AI Technology Stack Common AI Models: Linear and Statistical Models

Linear and Statistical Models Models the relationship between two variables. Example: how sales of a product are related to changes in pricing based on historical data. * Linear statistical models are not not a black box algorithm and more explainable.

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Linguistic variables and fuzzy rules

Linguistic variables and fuzzy rules * Linquistic variables describe concepts in natural language terms, such as "low", "medium" or "high" and "warm", "hot" or "very hot" * Fuzzy rules express relationships between variables by using if-then statements. For example, a rule might state, "If the temperature is "very hot", then the fan speed should be set to "high".

Module 1: Foundations of Artificial Intellegence Lesson 3: AI Technology Stack Common AI Models: Machine Learning Models

Machine Learning Models * Have black box capabilities * Have a lack of transparency and explainablitiy * Neural Networks (based on human brain) * Contain nodes, like neurons, in a layeed structure and continuously improve the ability to find the right answer * Do not need to be trained to make complex nonlinear inferences in unstructured data * Commonly behind tecnology, such as facial recognition

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Machine Learning

Machine Learning is a branch of AI that leverage data and algorithms to enable systems to repeatedly learn and make decisions. * Improves over time without explicitly instructed or programmed to do so. * Categorized based on the type of training model they reply on. Primary models are supervised learning, unsupervised learning, and reinforcement learning.

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Robotics and AI: Machine perception

Machine Perception * Evolving field of potential convergence between robotics and AI * Systems are trained to process sensory information and mimic the human senses: sight, sound, touch, smell, and taste. * Robotic sensors provide relevent data through cameras, microphones, pressure sensors, 3D scanners, motion detectors and thermal imaging. * combining sensors and AI models enable systems to sift through data at a much faster rate and order of magnitude beyond human ability; eliminating noise, analyazing and categorizing information. * E.g. food production: imporve preparation and storage by developing systems that can touch, smell, and tast the produce.

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Definition of AI

Machines performing tasks that normally require human intellgence * A branch of computer science concerned with creating technology to do things that normally require human intellegence * Alan Turing, a cryptopgrapher and mathmatician, developed a test to determine whether a machine is intellegent (1950) * A machine was considered intelligent if it produces responses to human interviewer that fool the interviewer into thinking the ressponses are human * Definitions include common elements of AI: * Technology: use of tech and specified obectives for the tech to acheive * Autotonomy: leve of autonomy by the tech to acheive the defined objectives * Human Involvement: need for input to train the tech and indetify objectives for it to follow * Output: Technology produced output ie performing tasks, solving probelms, producing content.

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Development Phase: Model Training, Testing, and Validation

Model Training, Testing, and Validation For training, testing, and validation, use representational subsets of your original dataset * Training data: Used to training the machine learning model * Test Data: Used to test the performance of the machine learning model * both should include all type of data used in the original dataset or to be used in the final product. Training * Train, test, evaluate and retrain different models to determine what the best model is to use. * Determine the best settings to acheive the desired outcome for your AI system * This is an interative process Testing and Validation * Test your models on relevent evalution metrics for consistent and expected performance within identified metrics * Based on previously developed metrics determined as son as you know your systems requirements * Develop metrics to determine how to evaluate that requirements were met * Test no new data * Helps to ensure your models generalize well and meet your business goals overall.

Module 1: Foundations of Artificial Intellegen Lesson 3: AI Technology Stack Compute Infrastructure: Networks

Networks * High-speed networks needed to support AI models: complex models, deep learning models, natual language processing, large language models like ChatGPT * Deliver to training data in time to the AI algorithm as well as training and inference at scale through high-speed networks * High-performance compute. Underlying infrastructure is housed in the same data centers usually in the same rack and connect via fiber connections, * Edge computing: the Internet of Things. It is estimated that within the next three to five years each individual will have five connect devices. * Communciation or network protocols: based on a congestion-free design, especially for larger language models and neural networks * Transmission control protocol (standard). However, this required a package to be sent.

Module 1: Foundations of Artificial Intellegence Lesson 3: AI Technology Stack Common AI Models: Neural Networks

Neural Networks 1. Computer Vision Models: Used to recognize images in videos 2. Speech recognition models: used in products like Alexa, transcription software. Analyze speech across factors such as pitch, tone, language, accent. 3. Language models: natural language processing; allow computers to understand human language using machine learning, deep learning models, linquistics. USed to process and respond to large amounts of communciations data (e.g. customer service chatbots) 4. Reinforcement learning models: train model to optimize their actions within a given environment to acheive a specific goals. Guided by feedback mechanisms of rewards and penalties. Conducted through trial and error; interactions or simulated experiences that do not require external data. Example: an algorithm trained to earn a high score in a video game by having its effors evaluated and rated according to success towarded the goal. * Disadvantage: lack of explainability and tranparency.

Module 2: AI Impacts on People and Responsible AI Principles Lesson 1: Existing and Emerging Ethical Guidance on AI OECD AI Principles

OECD AI Principles The OECD has a set of principles specifics to promoting trustworthy AI use: 1. Inclusive growth, sustainable development and well-being Highlights the potential for trustworthy AI to contribute to overall growth and prosperity for individuals, society and the planet, and advance global development obectives. 2. Human-centered values and fairness States that AI systems should be designed in a way that respects the rule of law, human rights, demongraic values, and diversity, and include appropirate safeguards to ensyre fairness and justice. 3. Transparency and explainability Calls for transparency and responsible disclosure around AI systems so that people understand when they are engaging with them and can challenge outcomes. 4. Robustness, security and safety States that AI systems must function in a robust, secure and safe way throughout their lifetimes, and potential risks should be continually assessed and managed. 5. Accountability Proposes that organizations and individuals who develop, deploy or operate AI systems should be held accountable for their proper functioning in line with the OECD's values-based principles for AI.

Module 2: AI Impacts on People and Responsible AI Principles Lesson 1: Existing and Emerging Ethical Guidance on AI OECD and FIPs Guidelines Common Priciples

OECD Guidelines on the Protection of Privacy and Transborder Flows of Personal Data (OEDC Guidelines) and Fair Information Practices (FIPs) * The FIPs, orginated by the OECD Guidelines in 1980, are rooted in decades of ethical guidance and organizational design for privacy, security, and other data or technology related functions. These have since been echoed in various permutations by other international orgs and bu U.S. gov agencies (Dept. of Homeland Security, Federal Trade Commission) * FIPs which are pirmarily focused on data collection, use, protection, and associated individual rights relative to personal data, there have been many follow-on sets of principles to apply them in various context such as AI governance. Common Principles: Data minimization or collection limitation Use limitation Safeguards or security Notice or openness Access or individual participation Accountability Purpose specification Data quality and relevance

Module 2: AI Impacts on People and Responsible AI Principles Lesson 1: Existing and Emerging Ethical Guidance on AI Train and Education Employees to create a culture of ethical AI Creating a culture of ethical AI

Organizations should have programs to train and educate employees to create a culture of ethical AI: * Legal and compliance * Equitable design * Transparency and explainablity (also known as interpretability) * Privacy and cybersecurity * Data governance

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI What constitutes artificial intellegence?

Over the decades, society's advancements have altered our perception of what type of machine or automated process is sophisticated enough to be considered "intellegent". What constitutes indentifying a machine or automated process as artificial intellegence and what challenges do AI professionals and AI governance face?

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle AI System Development Lifecycle

PDDI Planning, Design, Development, Implementation

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Use Case and Benefit of AI - Personalization

Personalization: * Unique online customer profiles: AI systems can help develop a profuile based on an individuals previous activity and create a unique experience that better meets the individuals needs. Personalization can also improve customer engagement and sales.

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Design: Wrangling/Preparing the Data: Privacy Enhancing Technologies (PETs)

Privacy Enhancing Technologies (PETs) Differential Privacy * Blurs the data by using an algorithm that keeps the data meaningful but m,akes it nonspecific * Individuals are unidentifiable but the data is still usable Federated Learning * A new way to train models/machine learning method that does not require sharing sensitive data among different locations * The global model is in a central location (e.g. the cloud) * Different locations download the global model and train it on their own local data * only the updates of the local model, not the training data itself are sent to the central location where they are aggregated into the global model * The process is iterated until the global model is fully trained * A great way to potentially solve problems such as diagnosing a new illness - using data from different location where they might have seen symptoms of the illness.

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Semi-Supervised Learning Models

Semi-Supervised Learning Models Addition to the 3 primary types of machine learning models, a combination of supervised and undervised learning processes * uses a small amount of labeled data and a large amount of unlabled data * aims to leverage the benefits of both models; improving reliablility while reducing costs Helpful in scenarios where it is challenging to find or create a large pre-labeled dataset * Examples: * Image and speech analysis * Categorization and raking of web page search results * LLMs * ChatGPT, Dall-e and other generative AI-tools

Module 1: Foundations of Artificial Intellegen Lesson 3: AI Technology Stack Compute Infrastructure: Severless

Severless Not limited to a particular server or piece of hardware.; running code on multiple hardware devices, providing two important services or functionality for AI: 1. Loose coupling: taking data from a variety of sources 2. Scalablity: running multiple instances of the code because its not not tied to a given server, which helps drive AI forward.

Module 2: AI Impacts on People and Responsible AI Principles Lesson 2: Core risks and harms posed by AI systems Societal Harms

Societal Harms * Spread of misimformation * Ideological bubbles/echo chambers - unable to see differing views or understand broader societal implications. Causes isolcation and more division. * Deepfakes: audio, video or images manipulate to create an alternative reality * Safety: lethal autonomous weapons that identify targets to attack. Concern without sufficient oversight, systems could evolve and may be able to attack randomly without being monitored.

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI History of AI and Evolution of Data Science

The Darmouth Conference, 1956 - AI born * Conference gathered leading researchers in researchers in fields relevant to AI * Previously. the disciplines of psychology, computer science, linquistics and engineering formed the foundation of what would become AI * Term AI adopted, creating AI as a field * Participants and Events: gathered leading researchers in fields relevant to AI such as Allen Newlell and Hebert Simon (introducted the Logic Theorist, considered by many as the first AI program at the conference). Free flow of ideas and brainstorming.

Module 1: Foundations of Artificial Intellegen Lesson 3: AI Technology Stack Compute Infrastructure: Software

Software: Driven and contributed to AI challenges 1. Democratization of AI: AI is easier to use, low code software, simpler AI interfaces 2. Tuning AI systems: allows customizations of AI models to generate more accurate outcomes and provide highly valuable insights into data. Usually done through trial and error by changing hyper parameters (very numerous in complex models), varies on model types and complexity. 3. Scale AI models: Trial-and-error tuning (doing it as you go); wnat to move from trial-and-error tuning to educated tuning. 4. Transformation: Data must be transformed into an AI model * Increases data compatability: some AI pipelines require transforming data for comatablity with the data the AI model wil be analyzing, optimzing data quality. 5. Labeling: Enriches the data used for deployment, training and tuning. Impactful and determines the quality of the AI model and its results; data labels need to be a high quality and standard. Challenges: low-quality data labels, scaling high-quality data, data labeling operationg and lack of quality assurance in data labeling operations; needing to verify and validate the data labels are high quality. 6. Observability and monitoring: intended to monitor the overall health and status of an org's data ecosystem. * AI observability is a subcomponent of data obversablity focused on monitoring the performance of an AI algorithm, that data going in and out, and the metric of an AI system. * Perform outcome validation to ensure that desired outcomes are deliverd, align with the AI model and can be more or less predicted from tuning and transformations. Challenges: Data integrity Data drift Bias and Discrimination

Module 1: Foundations of Artificial Intellegen Lesson 3: AI Technology Stack Compute Infrastructure: Storage

Storage Four general stages of AI: ingestion, preparation, training, and output (inference) * Each stage has different storage requirements that must be adhered to in order to avoid project failure. * Considerations for storage: * Expense of a storage solution for massive amounts of data * Different storage for a variety of storage types (file, object, image, etc.). Each require different storage subsystems. This may also affect expenses. * Storage types for structured vs unstructured data * Easier to process structured data than unstructured data * AI must be done at scale; flexible storage allows the ability to do an AI at scale.

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Supervised Learning Models

Supervised Learning Models * Learn from pre-labeled and classified data set * An algorithm analyzes the input data and associated labels to produce an inferred function, that becomes the basis for the system to make predictions based on new previously unseen inputs. * Compare their inputs with correct or inteded output, to identify errrors and improve prediction skills (e.d. model that analyzes images of road signs labeled to defined the sign's meaning or purpose.) 2 Subcategories: 1. Classification Models: produce outputs in the form of a specific categorical response; for example, whether an image contains a puppy. 2. Regression Models: predict a continous value, for example estimating a stock price. *Examples: Support Vector Machine (SVM), used for both classification and regression tasks, but mostly widely used for classification objectives; Support Vector Regression (SVR) most commonly used to produce continous values.

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Planning Phase - Use Cases

To determine the scope of the project, first prioritize the business problems you want to soleve. To do this, focus on three qualities: 1. Impact: of use of an AI system for the particular problem * How big of an impact will it have? * Will it solve a bigger problem or a smaller problem? * What is it going to take to do that? 2. Effort: * What types of resources do you need available to implement to new AI system? * How long is it going to take? 3. Fit: to priotiize the use case and business case * How well does the use of an AI system gfit with the goals of the org and the identified business problem?

Module 1: Foundations of Artificial Intellegen Lesson 3: AI Technology Stack Compute Infrastructure: Trusted execution environments

Trusted execution environments Good from a privacy perspective for AI, as human influence is taken out of the question.

Module 1: Foundations of Artificial Intellegence Lesson 1: Core Concepts of AI Unsupervised Learning

Unlabled data; typicalled used for pattern detection. Example: Outliers in data such as banking data; reviewing transactions for any fraudulent behaviour.

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Unsupersived Learning Models

Unsupervised Learning Models * Do not rely on labeled datasets * Designed to identify differences, similarities and other patterns without human supervision More cost effective and required less effort; susceptive to producing less accurate outputs and can display unpredictable behaviours. Two categories 1. Clustering: automatically grouping data points that share similar or identical attributes (eg DNA samples that share similaries or patterns) 2. Association rule learning: identifying relationships and associations between data points (e.g. understanding consumer buying habits) * Additional examples: anomaly detection and marketing strategies, genetics identification, consumer segementation and marketing strategies; genetics.

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Planning Phase - Use Cases

Use cases * Next, identify AI use cases, focus on the organziation mission * What is the mission of your organziation: * What do you do? * What's important to you and what are your main goals? * Then, identify the gaps: Where is the organization not meeting its goals * Use as input for cases

Module 1: Foundations of Artificial Intellegence Lesson 2: Types of AI Expert Systems: User Interface

User Interface Allows the end user to interact with the expert system by providing it an input (problem or a question) and obtaining an output (resolution).

Module 2: AI Impacts on People and Responsible AI Principles Lesson 2: Core risks and harms posed by AI systems Who is affected?

Who is affected? Individuals: (civil rights, economic opportunity, safety) Groups (discrimination toward subgroups) Society (democratic process, public trust in governmental institutions, educational access, jobs redistribution) Companies/institution (reputational, cultural, economic, accelaration risks) Ecosystems (natural resources, environment, supply chain)

Module 3: AI Development Life-Cyle Lesson 1: Understanding the AI Development Life-Cyle Design: Wrangling/Preparing the Data

Wrangling/Preparing the Data The most time consuming process (about 80%) * involves taking raw data and concering it to valuable information The five v's of data preparation: 1. Volumn: How much data do you have? How large is the data set? This is necessary to understand how much prep is needed. 2. Velocity: How often does it get updated? does it regularly change? 3. Variety: what type of data is it? Is it structured, unstructured or another type of data? 4. Veractiy: How accurate is it? How trustworthy is it? did you get from a source you know is reliable, so you don't have to worry that the data might not be correct? 5. Value: What is the outcome that you want from the use of the AI system? Will the data get you there? Is it the right data to use?