Artificial Intelligence

Subfields of AI

machine learning and deep learning

Machine learning

uses data and algorithms to emulate the way humans learn, relies on human interaction to gain knowledge

Deep learning

uses data and algorithms to create learning, some portions of learning are automated

Examples of deep learning

speech recognition, customer service, digital image analysis, automated stock trading

4 positive outcomes of AI

increased productivity - higher profitability

decreased human risk & injury - automates dangerous tasks like mining, etc.

diminished human errors - computers have lower chance of making mistakes

ability to better target product and service offerings - data science

2 drawbacks of AI

job losses due to automation

privacy concerns

• data repurposing - use of data outside original scope

• data spillover - use of data from individuals who were not the original targets of data collection

• data security

modernizing

prepares data for an AI and hybrid eCloud world

AI ladder approach to information architecture (IA)

infuse, analyze, organize, collect

Infuse - AI ladder step

operationalize AI throughout the business to be used across multiple oeprating areas

Analyze - AI ladder step

build and scale AI with trust and transparency, can later be scaled to maximize benefits and insights

Organize - AI ladder step

create business-ready analytics foundation by considering of data has been cleansed, if it’s complete, etc.

Collect - AI ladder step

make data simple and accessible

Cloud computing

computing model where processing, storage, software applications, and other services are provided over a network

Deloitte study on cloud computing

cloud-based software used to deliver AI to 70% of companies that use AI and 65% of companies use cloud services to create AI

Examples of cloud-based AI applications

Internet of things, ai as a service (AIaas) which is a type of SaaS, Chatbots

AIaaS

type of SaaS, allow for experimentation with AI with lower risk than full implementation

Chatbots

computer programs that process and simulate human conversations and allow organizations to give customers an experience similar to communicating with a real person

Advantages of cloud-based AI

enhanced security, cost savings, increased efficiency, improved data management

Disadvantages of cloud-based AI

data privacy, internet connectivity concerns, lack of platform control

Purpose of cognitive computing

seeks to better understand unstructured data

4 Steps of cognitive computing

observation - observing behaviors, occurrences, reading large datasets, etc.

interpretation - drawing conclusions and generating hypotheses about meaning

evaluation - determine which hypothesis makes the most sense

decision - decide course of action or which decision to take

Natural Language Processing (NLP)

used by cognitive systems, study and application of programming techniques that allow computers to understand spoken words and text that are inputted by humans

Ransomware-as-a-service (RaaS)

makes detection and prevention increasingly service because machine learning can detect anomalies with ransomware

Types of machine learning

supervised and unsupervised learning

Supervised learning

uses data class labels within data sets that specify what the data represents, used to create classification model

Classification model

designed when machine learning allocates a label value to a specific class then seeks to recognize these values to decide categories they fit into

Unsupervised learning

machine learning algorithms are used to examine and cluster unlabeled data sets

Clustering

approach used in data mining that groups unlabeled data based on parallels and variances within the data sets, uncovers hidden associations

Drawbacks of unsupervised learning

increased complexity due to large volumes of data, need for human intervention to validate results A

Application of unsupervised learning

medical imaging- assists radiologists with accurate/faster diagnoses

Reinforcement learning

focuses on using machine learning to analyze a scenario that in turn creates learning about how a behavior can be optimized to achieve maximum outcome/reward

Neural network

method in artificial intelligence that teaches computers to process data in a way that is inspired by the human brain

Layers of neural network

input, hidden, output

Input layer - neural network

execute initial handling of data (initial processing, analysis, categorization)

Hidden layer - neural network

can be multiple hidden layers, processing becomes increasingly complex as data moves to the next layer, output from each layer is analyzed & refined

Output layer - neural network

analyzes information from previous layers to make final prediction or conclusion

Application of neural networks

Rendering invariant state-prediction (RISP)

seeks to overcome cost and difficulty from motion capture technology by eliminating necessity of sensors

Generative adversarial networks (GANs)

type of machine learning model that uses 2 neural networks to compete against each other to create artificial instances of data that are interpreted as real data

Types of neural networks in GAN

generative network & discriminative network

Generative network - GAN neural network

used to create new synthetic instances of data that are erroneously identified as real data

Discriminative network- GAN neural network

tests and identifies which data it receives has been synthetically created, enables GAN to create real looking images by using the discriminative network to identify need for reworking until perceived as real image

Uses of GANs

adaptation of black and white images to color, generation of realistic images from text, etc.

Deep Learning Process used to build, train, deploy systems

collect data, choose and optimize algorithm, setup and manage training environment (neural networks, etc.), train/retrain/tune models, deploy models into production, scale and manage production environment

Common types of inference for deep learning models

batch inference: scheduled on recurring basis

real-time inference: gathered on request

Federal Trade Commission (FTC) policies for use of AI

be transparent with consumers, explain how algorithms make decisions, ensure good decisions, hold themselves accountable for ethics/fairness/non-discrimination

10 principles to consider when articulating methods for the development/use of AI

1. create public trust

2. public participation in rule-making processes

3. scientific integrity & information quality

4. risk assessment

5. benefits and costs: analyze which tools produce net benefits

6. flexibility

7. fairness & nondiscrimination

8. disclosure & transparency

9. safety & security

10. interagency coordination: collaboration with governmental agencies on best uses

Ethical concerns surrounding AI

privacy and surveillance, bias and discrimination, role of human judgement

Differential Privacy (DP)

mathematical framework that can be used to analyze how much an algorithm recalls data and information about individuals

Surveillance

used to monitor customer patterns and behaviors to identify shoplifting or shooting suspects, can also cause data privacy problems

AI bias

variance that exists in the output provided by AI algorithms

Factors causing AI bias

cognitive bias or incomplete data

Cognitive bias

conscious or unconscious errors in cognition that impact individual’s judgments, assumptions, and decision making

3 methods to gain human trust in AI

transparency to dispel misconceptions, instilling human values in AI, collaboration through establishing a common set of standards and platforms

5 Guiding principles for ethical use of AI

socially beneficial, avoid forming unfair bias, follow laws, incorporate transparency and accountability, implement data security & privacy

AI in business

sales & marketing, accounting, IT operations, manufacturing

AI in healthcare

robotic surgery, health tracking, pharmaceuticals (increase efficiency in processing)

AI in government

automation of routine tasks, military (precise targeting), policing aka law enforcement, predictive policing

4 careers associated with AI

user experience: analyzes customer data to optimize products and services

AI engineer: build AI models

machine learning engineer: building of machine learning models and training models

data scientist: develop data models and algorithms to automate processes

Tips to learn AI

learn about the math used, study computer programming languages, take free courses

AI chatbot

computer program that incorporates artifical intelligence and natural language processing to interpret user-provided questions and provide automated responses

examples of AI chatbots

ChatGPT, Caktus AI, AI-Powered Bing

Artificial Intelligence language models

computer software programs that use artifical intelligence to process and generate human languages

NLU

natural language understanding, used to understand the user’s needs better in AI

Chat GPT uses

idea generation & research, language translation, sentiment analysis (tone detection), analysis of computer code, marketing

Caktus AI

first AI designed specifically for education, pay to use service

Uses of Caktus AI

essay writer, paragraph generator, content improver, python writer

Features of AI powered bing

targeted search results, completed responses, display options, create and compose emails/brand image/etc., image creator

AI plagiarism detector examples

AI text classifier by open AI, GPTZero, Turnitin

Ethical problems with chatbots

transparency, chatbot identity reinforcing gender stereotypes, proper communication in regards to sensitive topics, accurate data representation

Ethical problems with chatbots for students

misinformation (wrong answers), plagiarism due to chatbot’s lack of citations and students thus not citing their sources

Areas of research for chatbots

personalization, interactions in multiple languages, technology integration (virtual reality), emotional intelligence, AI-assisted shopping and task completion, artificial general intelligence (being able to complete any tasks that the human brain can do)