1. Foundations of AI Flashcards
What is Artificial Intelligence?
Machines performing tasks that normally require human intelligence
What is the Turing Test?
A test to determine whether a machine is intelligent. Considered intelligent if its responses fooled an interviewer into thinking it was human.
What are the common elements of AI definitions?
Technology- use of tech (engineered or machine-based system) and specific objectives (logic, knowledge, or learning algorithm)
Autonomy- discussion of autonomy by tech to achieve objectives
Human Involvement- need for human input to train tech/provide data and identify objectives
Output- tech produces output (content, predictions, recommendations, or decisions)
What is Machine Learning?
The process of training machines to display AI behavior
What are the three types of Machine Learning?
Supervised
Unsupervised
Reinforcement
What is Supervised Learning (ML)?
Labeled data grouped or classified into categories via the AI system
Ex) Two groups of labeled images- cats & dogs. AI system wants to determine what makes something a cat vs. a dog and correctly classify new images. Used for text recognition and spam.
What is Unsupervised Learning (ML)?
Unlabeled data, typically used for pattern detection
Ex) AI system looks for outliers in data- such as financial data examination looking for fraud
What is Reinforcement Learning (ML)?
AI system is rewarded for performing a task well and penalized for not performing a task well. Over time, it learns to maximize rewards and find a system that works.
Ex) Self-driving cars- rewarded for keeping car safe, penalized for getting lost/hitting object. System over time maximizes rewards.
Why is AI a sociotechnical system?
AI influences society, and society influences AI
What are some of the general risks of using AI?
AI systems are incredibly complex, and AI data will change over time. Systems will need to be updated periodically. Finally, systems are implemented in complex environments.
What is the OECD AI Framework?
OECD= Organization for Economic Cooperative Development.
Designed to classify AI systems and examine risks. Consists of 5 dimensions.
What are the five dimensions of the OECD AI framework?
1) People and Planet - identify individuals & groups that may be affected by AI system. How does system impact human rights, environment, society. Privacy comes into play here regarding use.
2) Economic Context - economic/sectorial environment in which the AI system operates (ex financial, healthcare, education), and business function for the model, whether critical to function of business, how it’s deployed/scale, technological maturity (untested vs. well-trained)
3) Data and Input - what types of data used in model, whether expert input is used (human input put into rules), structure/format of data
4) AI model - technical type, how it was built, how it was used
5) Tasks and Output - tasks AI system performs, its outputs, resulting actions from these outputs. characteristics include tasks, tasks & actions combined, evaluation methods to determine system performance
What are AI Governance Principles? Examples?
AI governance principles are guidelines that help enable consistency, standardization and responsible use of AI. Around the world, the principles that guide responsible AI governance are similar.
EX) OECD AI Principles, FIPPS, NSM
What are AI Governance Frameworks? Examples?
AI governance frameworks provide guidance for operationalizing values that come from principles. While there are similarities among AI frameworks, they are often context-sensitive and fit for specific purposes.
EX) ISO 42001, NIST, Council of Europe, NSM
What is the difference between AI Governance Principles and AI Governance Frameworks?
AI governance principles are a set of values, whereas an AI governance framework is a means to operationalize those values.
Name three categories of AI use cases
1) Recognition, Detection, and Forecasting
2) Personalization and Interaction Support
3) Goal-driven Optimization and Recommendation
Describe the use case/benefits of AI Recognition
Image/Speech/Facial recognition - look at one input, see if it matches something else.
- Retailer product matching
- Identifying individuals
- Teach manufacturing machines to see defects
- Plagiarism detection
Describe the use case/benefits of AI Event Detection
Using AI to detect specific events in a large amount of data
- Credit card fraud detection
- Identity theft detection
- Events in sports videos (touchdown, goal, etc)
- Detect cyber events and protect systems/networks
Describe the use case/benefits of AI Forecasting
Using past data to forecast future conditions
- Allow companies to predict sales, revenue, demand
- Determine demand for rides in ride-sharing apps/surge pricing
- Weather forecasting
Describe the use case/benefits of AI Personalization
Unique customer profiles created based on behavior observed during web use, used to connect users to relevant information on company sites- increases sales
Describe the use case/benefits of AI Interaction Support
Chatbots to answer frequently asked questions, guide users through steps on a website (ex. student loan applications)
Describe the use case/benefits of AI Goal-Driven Optimization
Asking AI to analyze a problem and generate multiple solutions to answer that problem
- Optimize supply chain to increase efficiency
- Optimize driving routes and reduce idle time for mass transit/deliveries
Describe the use case/benefits of AI Recommendations
- Product or viewing recommendations based on predictive analytics.
- Help humans make better purchasing decisions
- Healthcare- diagnosis support
- Government- adjudicating disability cases/determine benefits
What are the three high-level categories of AI?
1) Artificial NARROW intelligence
2) Artificial GENERAL intelligence
3) Artificial SUPER intelligence
What is Artificial NARROW Intelligence? What’s another name for this? Example?
ANI is designed to perform a single or a narrow set of related tasks at a high level of proficiency.
ANIs operate under a narrow set of constraints and limitations, which is why this type of AI is commonly referred to as weak AI
Can help boost productivity and efficiency by automating repetitive tasks, enabling smarter decision making and optimization through trend analysis
Ex) AI system designed to play chess
What is Artificial GENERAL Intelligence? What’s another name for this? Example?
AGI, is also known as Strong, Deep or Full AI. It 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 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
What is Artificial SUPER Intelligence? What’s another name for this? Example?
ASI, is a category of AI systems with intellectual powers beyond those of humans across a comprehensive range of categories and fields of endeavor. Thus, it is capable of outperforming humans.
Like AGI, ASI does not yet exist. However, experts expect this type of system would be self-aware: able to understand human emotions and experiences and evoke its own, thus experiencing reality like humans.
What is Broad Artificial Intelligence?
Broad artificial intelligence is a category of AI more advanced in scope than ANI. It is capable of performing a broader set of tasks, but not sophisticated enough to be considered AGI.
Broad artificial intelligence often involves reliance on a group of artificial intelligence systems that are capable of working together and combining decision-making abilities, but still lacking the full human-like capabilities of AGI
How does the training process work for a Supervised Learning model?
an algorithm analyzes the input data and associated labels to produce an inferred function, which can then become the basis for the system to make predictions based on new, previously unseen inputs. Supervised learning models can also compare their outputs with the correct or intended output, to identify errors and improve their prediction skills.
An example of supervised learning is a model that analyzes images of road signs labeled to define the sign’s meaning or purpose
What are the two subcategories of Supervised Learning Models?
1) Classification models produce outputs in the form of a specific categorical response; for example, whether an image contains a puppy. EX) Support Vector Machine (SVM)
2) Regression models predict a continuous value; for example, estimating a stock price. EX) Support Vector Regression (SVR)
What are the two subcategories of Unsupervised Learning Models?
1) Clustering - automatically groups data points that share similar or identical attributes; for example, similarities or patterns in DNA samples.
2) Association rule learning - identifies relationships and associations between data points; for example, consumer buying habits
What are the benefits/risks of Unsupervised Learning Models?
Unsupervised learning models tend to be more cost-efficient and require less effort, but are susceptible to producing less accurate outputs and displaying unpredictable behaviors.
How does the training process work for a Reinforcement Learning model?
Reinforcement learning models use a reward and punishment matrix to determine a correct or optimal outcome. They rely on trial and error to determine what to do or not do and are rewarded or punished accordingly.
These models do not ingest pre-labeled data sets and learn solely through action and repetition, changing or not changing state or by getting feedback from their environment.
What are Semi-Supervised Learning Models?
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.
What are the benefits of Semi-Supervised Learning Models?
These models 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.
What are Large Language Models?
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.
Name three emerging crossovers between robotics and AI
Industry 4.0 (Fourth Industrial Revolution) - industry and manufacturing advancements, enabled by increased interconnectivity and smart automation
Machine Perception - systems are trained to process sensory information and mimic human senses. enables systems to sift through data, eliminate noise and analyze and categorize information at a much faster rate and order of magnitude than humans
Robotic Process Automation - using software robots to automate repetitive and rule-based tasks in business processes. RPAs are designed to mimic human actions on digital systems
What is an expert system?
a form of AI intended to mimic the decision-making abilities of a human expert in a specific field. These systems draw inferences from a knowledge base and rely on AI to replicate the judgment and behavior of a human with specific expertise
What are the three main elements that distinguish an expert system from other AI?
1) Knowledge Base - typically an organized collection of facts and information from human experts, focused on a specific field or domain. In some cases, the system is also allowed to gather additional information from external sources
2) Inference Engine - extracts relevant information from the knowledge base and uses it appropriately to solve a problem. Expert systems normally use a rule-based approach that maps data from the knowledge base to a series of rules, which the system uses to make decisions in response to input. Expert systems often include a module that allows users to review its decision-making process
3) User Interface - allows the end user to interact with the expert system. The user provides an input, like a problem or question, and obtains an output (resolution).
What are linguistic variables? How do they relate to fuzzy rules?
Linguistic 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 using if-then statements. For example, a fuzzy rule might state, “If the temperature is ‘very hot,’ then set the fan speed to ‘high.’”
What are the standard steps that fuzzy logic systems use?
1) Fuzzification - input data is converted into fuzzy data sets
2) Rule evaluation - which determines the degree of matching between the rules and input data
3) Aggregation - where rule outputs are combined
4) Defuzzification -process by which fuzzy outputs are converted back into specific values
EX) climate control systems, image recognition systems and traffic management systems
What’s the difference between an AI Platform and an AI Application?
AI Platform: Software used to develop, test, deploy and refresh AI applications. EX) Microsoft Azure, AWS
AI Application: How an AI system is used. EX) Chatbots, Self-driving, Marketing, etc.
What do AI Platforms do?
Centralize data analysis
Streamline development and production workflows
Facilitate collaboration
Automate systems-development tasks
Monitor models and systems in production
Name six types of AI Models
Linear/statistical models
Decision trees
Neural networks (such as computer vision and speech recognition models)
Language models (like NLP)
Reinforcement learning models
Robotics applications
What are Linear and Statistical Models? When are they useful?
Linear and Statistical Models
What are Decision Trees? When are they useful?
Decision Trees
What are Neural networks? When are they useful?
Neural networks
What are Language models? When are they useful?
Language models
What are Reinforcement learning models? When are they useful?
Reinforcement learning models
What are Robotics applications? When are they useful?
Robotics applications
