T

Seven Patterns of AI (Video) - Fill-in-the-Blank Flashcards

When to Use AI & Cognitive Technology

  • AI & Cognitive Technologies are not always the right solution. Many problems are better solved with rules-based or traditional programming, especially if:

    • The problem is repetitive, deterministic, or requires high accuracy.

    • Inputs and outputs are always the same and predictable.

    • Automation (not intelligence) is the main goal.

  • Key questions to ask before using AI:

    • Are the inputs/outputs always the same? ext{(Yes/No)}

    • Is the process highly repetitive and predictable? ext{(Yes/No)}

    • Is high accuracy required, with no tolerance for ambiguity or error? ext{(Yes/No)}

    • Is there enough data to train an AI system? ext{(Yes/No)}

    • Would a human be more reliable, cost-effective, or provide a better experience? ext{(Yes/No)}

  • AI is best for problems with:

    • High variability in data, process, or flow. ext{(Yes)}

    • Ambiguity or tolerance for probabilistic (not guaranteed) results. ext{(Yes)}

    • Situations requiring cognitive intelligence (perception, prediction, planning). ext{(Yes)}

    • If 100% accuracy is required, or if there is no data, AI/ML is not suitable. ext{(No)}

    • Sometimes, hiring a person is more effective than implementing AI. ext{(Yes/No)}

  • A note: There is lots of value in AI & Cognitive Tech without being vague, but choose solution type carefully.

The Seven Patterns of AI

  • The patterns group AI applications into related areas. There are 7 patterns:

    • Recognition Pattern

    • Conversational (Human Interaction) Pattern

    • Predictive Analytics & Decisions

    • Goal-Driven Systems (Optimization & Autonomy)

    • Autonomous Systems

    • Patterns & Anomalies

    • Hyperpersonalization

  • Objective: Provide a categorization system to group different applications of AI into like application areas.

Recognition Pattern

  • Purpose: Make sense of unstructured data (images, audio, video, handwriting, faces, gestures).

  • Technologies: Computer vision, deep learning (e.g., CNNs), OCR.

  • Example use cases: Facial recognition, sound recognition, item & object detection, handwriting/text recognition, gesture/motion analysis.

  • Related concepts: Object recognition, classification, segmentation, pose estimation, event detection, scene reconstruction, image indexing, motion estimation, 3D modeling, image restoration.

  • Real-world uses:

    • Security and surveillance (facial recognition)

    • Retail and shopping analytics (gesture analysis)

    • Medical imaging and robotics (pose estimation, surgical assistance)

    • Entertainment and content analysis (music recognition, video tagging)

The Conversational Pattern (Conversation & Human Interaction)

  • Purpose: Machines and humans interact using natural language across voice, text, written and image forms.

  • Objective: Machines interact with humans the way humans interact with each other.

  • Technologies: NLP, NLU, NLG, STT (speech-to-text), TTS (text-to-speech), ASR (Automatic Speech Recognition).

  • Applications:

    • Chatbots (text or voice)

    • Voice assistants (e.g., Siri, Alexa)

    • Machine translation

    • Sentiment / mood / intent analysis

    • Content summarization and other NLP tasks

  • Important terms:

    • NLP = Natural Language Processing: getting machines to understand and communicate in human language.

    • NLU = Natural Language Understanding: understanding intent, entities, and context beyond words.

    • NLG = Natural Language Generation: generating human-language text or speech from data.

    • ASR = Automatic Speech Recognition: converting speech to text (often interchangeable with STT).

    • STT = Speech-to-Text; TTS = Text-to-Speech.

  • Content generation and translation involve a spectrum of NLP components (NLU/NLG) and are not limited to text-only capabilities.

Predictive Analytics & Decisions

  • Purpose: Use past or current data to help make better decisions and forecast outcomes.

  • Objective: Help humans make better decisions.

  • Technologies: Machine learning models for classification, regression, time-series analysis; pattern recognition; anomaly detection.

  • Applications:

    • Forecasting (inventory, weather, financial markets)

    • Risk/fraud detection and anomaly detection in transactions

    • Recommendation engines

    • Predictive maintenance

    • Marketing optimization and campaign planning

  • Key concepts:

    • Decision support: Presenting data and alternatives to help humans choose.

    • Predictive analytics: Forecasting trends and outcomes from historical data.

  • Real-world uses:

    • Retail forecasting, finance risk assessment, manufacturing maintenance, marketing optimization.

Goal-Driven Systems (Autonomy & Optimization)

  • Purpose: Find the optimal solution to a problem through trial and error and reinforcement learning.

  • Objective: Find the best outcome by exploring many scenarios and iterations.

  • Technologies: Reinforcement learning, simulation, optimization, planning, perception, prediction.

  • Applications:

    • Scenario simulation and scenario planning

    • Game playing and strategy optimization

    • Resource optimization (money, equipment, time, other resources)

    • Robo-advising and real-time bidding for ads

  • Key concept: Machine learning can discover hidden rules and strategic moves that lead to optimal results.

  • Real-world examples:

    • Deep reinforcement learning systems that learn to play complex games and optimize real-time decisions.

Autonomous Systems

  • Definition: Systems (physical or digital) that accomplish tasks and achieve goals with minimal or no human involvement.

  • Objective: Minimize human labor and intervention.

  • Example use cases:

    • Autonomous vehicles and drones

    • Autonomous robots and software systems

    • Autonomous business processes (software bots making independent decisions)

  • Important distinctions:

    • Automation vs. Intelligence: Automation repeats tasks; intelligence implies perception, prediction, planning and handling variability without human intervention.

    • The question to ask: Is there any ML in the system? Can the system improve over time and handle next steps without exceptions? If not, it’s not truly intelligent.

  • Note on levels and terms:

    • Automation vs. Autonomy: automation is often rule-based and can be unattended or attended; autonomy encompasses perception, prediction, planning.

    • Cobots (collaborative robots) are designed to work with humans in close proximity to augment human capabilities.

Patterns & Anomalies (Pattern Recognition & Anomaly Detection)

  • Purpose: Detect patterns and outliers in large data sets; identify data points that don’t fit patterns.

  • Technologies: Pattern recognition, anomaly detection, classification.

  • Applications:

    • Fraud detection and risk assessment in finance and commerce

    • Automatic error detection and correction (data quality, medical prescriptions, etc.)

    • Intelligent monitoring of systems (IT, cyber-physical systems)

    • HR/recruiting candidate screening and profiling

    • Content moderation and safety analytics

  • Key concepts:

    • Classification: Group data into categories

    • Anomaly detection: Identify data points deviating from the norm

  • Real-world example: Walmart’s discovery that strawberry Pop-Tarts sales spike before hurricanes (pattern discovery in time-series data).

Hyperpersonalization

  • Purpose: Treat each individual as an individual using adaptive models that personalize experiences over time.

  • Technologies: User profiling, adaptive learning, recommendation systems, perceptual ML pipelines.

  • Applications:

    • Personalized content delivery

    • Personalized recommendations and product suggestions

    • Behavioral profiling

    • Personalized medicine, finance, education

  • Key terms:

    • Personalization: Customizing offerings for user groups

    • Hyper-personalization: Customizing for each individual

    • Recommendation system: Suggests products, content, or actions based on user profile and behavior

  • Real-world uses: Social media feeds, online shopping, healthcare, finance, education

  • Summary: Personalization-driven experiences intensify engagement and relevance by leveraging individual-level data and adaptive models.

Robotic Process Automation (RPA) and Intelligent Process Automation (IPA)

  • RPA purpose: Automate repetitive software tasks at the UI level; handle back-office and front-office workflows.

  • Types:

    • Attended Bots: Assist humans in real-time; interact with users to speed up tasks.

    • Unattended Bots: Run in the background; operate autonomously on schedules or rules.

  • Technologies: Screen recording, scripting, low-code/no-code platforms, rules-based automation.

  • Applications: Data entry, invoice processing, data transfers between systems, customer support workflows, document processing.

  • IPA (Intelligent Process Automation): Add AI to RPA to handle variability, unstructured data, exceptions; helps automate perception, prediction, and planning within processes.

  • Levels of intelligent automation (illustrative ladder): Level 0 (basic automation) to Level 3+ (fully autonomous optimization), with increasing capability to handle data and exceptions.

  • Platform concepts:

    • Low-code/No-code: Enables citizen developers to build automation without heavy coding.

    • Attended vs. Unattended: Distinct modes of operation and collaboration with humans.

Robotics, Automation, and the 4 D’s / 3K’s of Robotics

  • Robotics: Engineering discipline to design, build, operate, and apply robots

  • 4 D’s / 3K’s (why automation is needed in manufacturing): Dull, Dangerous, Dirty, Demeaning (and sometimes Dehumanizing)

  • Cobots (Collaborative Robots): Robots designed to work alongside humans to augment capabilities; not simply isolating humans from machines.

  • Note: Robotics and automation are not inherently AI; many robots are programmed or use simple detection systems without AI.

Levels of Automation & Autonomy (Examples)

  • Autonomy levels (example: autonomous vehicles): Level 0 to Level 5, where Level 0 = no autonomy and Level 5 = fully driverless with no human involvement; Levels 1–4 represent increasing autonomous capabilities and decreasing human intervention.

    • Level 0: No autonomous features

    • Level 1: One autonomous function (e.g., automatic braking)

    • Level 2: Two or more automated functions; human remains in control

    • Level 3: Capable of dynamic driving but requires human intervention

    • Level 4: Driverless in controlled environments

    • Level 5: Fully autonomous across all environments

Autonomous Retail & Examples

  • Autonomous Retail: Removing the human from the loop in retail contexts

  • Examples:

    • Amazon Go

    • LoweBot (intelligent store assistant, 2016)

    • Walmart shelf-scanning bots (2017) — note: faced challenges and failures

    • Ongoing deployment of store bots across various chains

  • Future possibility: Autonomous bot baristas and related experiments

Goal-Driven Systems: Reinforcement Learning & Game Playing

  • Objective: Find the optimal solution to a problem through trial and error using reinforcement learning in real-world settings

  • Real-world uses:

    • Scenario simulation

    • Game playing (board games, video games)

    • Resource optimization (money, equipment, time, other resources)

    • Iterative problem solving and robo-advising

    • Bidding and real-time auctions

  • Notable examples in AI history:

    • DeepMind and reinforcement learning breakthroughs

    • AlphaGo and AlphaZero achievements in Go and other games

DeepMind, AlphaGo & AlphaZero

  • AlphaGo: AI designed to play Go using reinforcement learning and deep learning; defeated top human player Lee Sedol in 2016.

  • AlphaZero: Built on the success of AlphaGo; trained purely by self-play; surpassed AlphaGo in 24 hours and beat the earlier system at superhuman levels.

  • Significance: Demonstrated that self-play can yield superhuman performance in complex tasks, and that a single algorithmic framework can master multiple domains.

Combining Patterns for Applications

  • Many real-world applications combine multiple AI patterns to achieve outcomes.

  • Example: Assistant Enabled Commerce

    • Pattern(s) involved: Hyperpersonalization (personalized data-driven tailoring), Conversation (dialog between user and bot), Pattern recognition (analyzing buying behavior) to deliver tailored product suggestions and guided conversations.

Quiz Highlights and Key Takeaways

  • Automation vs. Intelligence: Automation repeats tasks; intelligence perceives, predicts, and plans to handle variability.

  • The seven patterns are a useful framework for planning AI projects and understanding data needs and algorithms.

  • Recognize that some problems are better solved by traditional programming, rules-based systems, or human labor rather than AI.

  • AI benefits from data: Without data or reliable training data, ML/AI solutions may underperform.

Natural Language Processing: NLP, NLU, NLG, ASR, STT, and TTS

  • NLP (Natural Language Processing): Getting machines to understand and communicate using human language; umbrella term.

  • NLU (Natural Language Understanding): Adds deeper understanding of intent, entities, and semantics; part of NLP.

  • NLG (Natural Language Generation): Generating human-like text from data; a complement to NLP, adds semantic understanding to speech/text generation.

  • ASR (Automatic Speech Recognition): Converts spoken language to text; often used in STT contexts and chatbot/voice assistant pipelines; often treated as a component of NLU.

  • STT (Speech-to-Text) and TTS (Text-to-Speech): STT converts audio to text; TTS converts text to spoken audio.

  • Content summarization & analysis relies on NLU to extract key information and produce concise semantic summaries.

ImageNet, Computer Vision, and Vision Challenges

  • ImageNet: Created in 2006 by Fei-Fei Li; a large free repository with over 14{,}000{,}000 labeled images organized by WordNet hierarchy for training and benchmarking computer vision systems.

  • ImageNet Large Scale Visual Recognition Challenge (ILSVRC) started in 2010 to benchmark object detection/classification performance.

  • Concerns: Data quality and bias — estimates have suggested that up to 5\% of ImageNet data may be mislabeled or biased, highlighting training data quality importance.

  • Computer Vision components include: image acquisition, processing, analysis, object detection, recognition, segmentation, scene reconstruction, event detection, motion estimation, 3D modeling, image restoration, and indexing.

  • Applications: Facial recognition, object detection, gesture analysis, medical imaging, document processing, robotics, entertainment, and more.

Handwriting/Text Recognition, Gesture Recognition & Motion Analysis

  • Handwriting/Text Recognition: Recognizing handwritten content beyond OCR; can train on images of unstructured data; used in check imaging, intelligent document processing, forms extraction.

  • Gesture Recognition & Motion Analysis: Interpreting human gestures as commands; use cases in gaming, retail experiences, virtual try-ons, and surgical assistance.

  • Motion analysis supports pose estimation, 3D understanding, and tool/gesture-based interfaces.

Sound & Audio Recognition

  • Sound/Audio Recognition: Recognizing sounds, music, instruments, and language components; identifying songs, patterns, and environmental sounds.

  • Applications include music recognition, language identification, animal sounds, and cross-language recognition tasks.

Computer Vision: Core Concepts & Capabilities

  • Computer Vision enables machines to interpret and understand visual inputs as part of the Recognition Pattern.

  • Core components include image acquisition, processing, analysis, object detection, recognition, segmentation, scene understanding, motion estimation, 3D reconstruction, and more.

The Recognition Pattern in Depth: Real-World Uses

  • Facial recognition in security and consumer devices

  • Object recognition and image classification in surveillance, retail, and manufacturing

  • Handwriting/Text recognition in document processing

  • Gesture recognition in customer analytics and interactive experiences

  • Sound recognition in media and safety applications

  • Document processing and intelligent document processing (IDP) when combining handwriting, text extraction, and table recognition

The Conversation Pattern: Real-World Interfaces

  • Conversational interfaces include chatbots and voice assistants that converse with users in natural language

  • They rely on a combination of NLP, NLU, NLG, STT, and TTS to provide engaging interactions

Practical Examples Across Patterns

  • Chatbots and voice assistants for customer service

  • Content generation and natural language generation in consumer platforms

  • Sentiment/mood/intent analysis in social listening and feedback systems

  • Machine translation bridging languages across contexts

  • Content summarization and semantic analysis for long-form documents

  • Content recommendations and personalization driven by hyperpersonalization patterns

Content Summarization & Analysis

  • Purpose: Use AI/ML to produce concise overviews of bodies of text or other content

  • Approach: Use NLU to extract key information and generate semantic summaries

Quiz Highlights (Key Takeaways)

  • Which one is NOT a pattern of AI? Automation is not a pattern; it is a form of automation, not AI per se.

  • The first real conversational chatbot timeline and the number of AI patterns (7) are factual checkpoints to remember.

Key Definitions and Concepts (glossary)

  • Automation: Repetitive task execution, not necessarily intelligent.

  • Autonomous Systems: Minimize human involvement; require perception, prediction, and planning.

  • Computer Vision: Techniques to interpret visual data using ML/DL.

  • NLP: Teaching computers to understand and generate human language.

  • NLU: Understanding intent, context, and entities in language data.

  • NLG: Generating readable human language from data.

  • Sentiment Analysis: Classifying text by mood, intent, or opinion.

  • RPA: Software bots automating business processes via UI-level automation.

  • Attended vs. Unattended Bots: Collaboration vs. autonomous background operation.

  • Low-code/No-code: Platforms enabling automation with minimal or no programming.

  • Cobot: Collaborative robot; works alongside humans.

  • IPA: Intelligent Process Automation; RPA enhanced with AI for greater variability handling and autonomy.

  • Level-based autonomy (vehicles example): Levels 0–5 define increasing degrees of self-operation without human input.

References for further study

  • ImageNet and ILSVRC background discussions

  • DeepMind AlphaGo & AlphaZero milestones

  • RPA/IPA strategies in modern enterprises

  • Distinctions among NLP, NLU, NLG, ASR, STT, and TTS

  • The differences between automation, autonomous systems, and AI intelligence