COSC 105 - INTELLIGENCE SYSTEM (MIDTERM REVIEWER)

Simulation of human intelligence in machines; machines programmed to think like humans; exhibits learning and problem-solving; broad field with many techniques.

Artificial Intelligence (AI)

Algorithms that let systems learn from data, identify patterns, and make decisions with minimal human intervention; includes supervised, unsupervised, reinforcement.

Machine Learning (ML)

Enables computers to understand, interpret, and generate human language; used in chatbots, translation, sentiment analysis, kiosks, voice assistants; determines intent and formulates replies.

Natural Language Processing (NLP)

Subset of ML that uses neural networks with many layers to learn complex patterns from large amounts of data; powers voice assistants, autonomous vehicles, medical imaging.

Deep Learning

Allows computers to see and interpret visual information from images and videos; used in facial recognition, object detection, autonomous navigation, security cameras.

Computer Vision

Proposed by Alan Turing in 1950 paper 'Computing Machinery and Intelligence'; tests a machine's ability to exhibit intelligent behavior.

Turing Test

1960s–70s period when early optimism faded, limitations became clear, and funding decreased.

AI Winter

1980s rise of rule-based systems for specific problem-solving; use knowledge bases and predefined rules; used for diagnostics and decision support.

Expert Systems

2010s–present advances in neural networks, data, and computational power.

Deep Learning Boom

AI that can perform any intellectual task a human can.

General AI

Focus on AI that is fair, transparent, and accountable.

Ethical AI

Automation of repetitive and routine tasks; affects manufacturing, administrative, and transport sectors.

Job Displacement

Demand for AI specialists, data scientists, and ethicists; new roles in AI training, maintenance, and oversight.

Job Creation

Vital to prepare the workforce for AI changes.

Reskilling / Upskilling

AI can perpetuate and amplify biases in training data, leading to unfair or discriminatory outcomes.

Bias and Fairness

Extensive data collection raises concerns about privacy, misuse, and breaches.

Privacy and Data Security

Hard to determine who is responsible when AI makes mistakes, and how complex models reach decisions.

Accountability and Transparency

AI in physical form; merges AI + mechanical engineering; designs, constructs, operates, and applies robots.

Robotics

Cutting-edge branch focused on creating novel content: text, images, audio, synthetic data; uses GANs and Transformers.

Generative AI

Sophisticated models used in generative AI to create new content.

GANs

Sophisticated models used in generative AI; generate dynamic, contextually relevant replies.

Transformers

Learns from labeled data to predict outcomes; like learning with an answer key.

Supervised Learning

Finds hidden patterns in unlabeled data; discovers structure without prior guidance.

Unsupervised Learning

Learns through trial and error, optimizing actions based on rewards or penalties.

Reinforcement Learning

Deep learning structure: input layer, multiple hidden layers, output layer; mimics human brain.

Neural Network Structure

AI-powered customer support for FAQs, order tracking, troubleshooting, escalation to live agents.

PLDT Cares

In-car voice assistants for intuitive controls and personalized assistance.

Mercedes-Benz

Use AI for order processing, recommendations, and delivery logistics.

Wendy's and Uber

Converts spoken audio into written text.

Speech-to-Text (STT)

Converts text responses into natural-sounding speech.

Text-to-Speech (TTS)

Physical self-service terminal enhanced with AI: computer vision, NLP, conversational voice recognition, predictive analytics; delivers interactive, touchless, personalized interactions.

AI-Driven Kiosk

AI analyzes vast datasets to detect suspicious patterns and prevent cyber attacks in real time.

Fraud Detection & Cybersecurity

Banks that monitor global markets and transactions to combat fraud and money laundering.

Citi and Deutsche Bank

AI learns from evolving threats, unlike traditional rule-based systems.

Adaptive Threat Response

Predicts vulnerabilities and recommends countermeasures.

Proactive Security

AI analyzes sensor data from machinery to predict failures before they occur; enables proactive maintenance scheduling.

Predictive Maintenance

Uses live signals like weather, geopolitics, social media, economic indicators to update inventory needs.

Real-Time Demand Sensing & Forecasting

Evaluates traffic, port congestion, weather, fuel prices to re-route fleets.

Dynamic Route & Fleet Optimization

ML monitors stock velocity across warehouses; automatically triggers transfers to prevent stockouts.

Automated Multi-Echelon Inventory Rebalancing

AI enables highly personalized experiences; analyzes user behavior, preferences, demographics; delivers customized content and recommendations.

Marketing & Sales

Automates data entry, invoice processing, and KPI monitoring.

Automated Data Processing

AI-powered analytics provide instant insights for faster decisions.

Real-Time Insights

AutoML tools let non-experts build, train, and deploy AI models.

Simplified AI Model Building

Example of simplified AI model building tool.

Excel's Data Analysis tools

Intelligent systems that perform tasks autonomously; act like digital assistants; examples include ChatGPT and Gemini.

AI Agents & Personal Assistants

Examples of AI assistants.

ChatGPT, Gemini

Analyzes X-rays, MRIs, CT scans to detect anomalies like early-stage cancer.

Advanced Image Analysis

Sifts through chemical/biological datasets to identify drug candidates.

Accelerated Drug Discovery

Tailors treatments based on genetics, lifestyle, environment.

Precision Medicine

AI-assisted robots (e.g., da Vinci, CyberKnife) improve precision and recovery.

Robotic Surgery

AI-assisted surgical system cited for precision, less invasiveness, and faster recovery.

da Vinci Surgical System

AI-assisted surgical system example alongside da Vinci; improves precision and recovery.

CyberKnife

Wearables track vital signs for proactive chronic disease management.

Remote Monitoring

AI supports, not replaces, clinicians; enhances and supports human judgment and empathy.

Augmented Intelligence

First chatbot, developed in 1966; simulated a psychotherapist using pattern matching.

ELIZA

Generate dynamic, contextually relevant replies; used in modern chatbots.

Large Language Models (LLMs)

Connects bots to systems for actions like booking, notifications, recommendations.

Backend System Integration

Step 1 in building a chatbot: clarify purpose.

Define Objectives

Step 2 in building a chatbot: Python, Dialogflow, Microsoft Bot Framework.

Choose Platform & Tech Stack

Step 3 in building a chatbot: map user interactions and dialogue paths.

Design Conversational Flows

Step 4 in building a chatbot: feed relevant data to improve accuracy.

Train with Datasets

Step 5 in building a chatbot: gather feedback and refine.

Test & Iterate

Step 6 in building a chatbot: websites, messaging apps, voice assistants.

Deploy Across Channels

Handling complex, ambiguous queries; emotional nuances; maintaining context across multi-topic conversations.

Current Limits of Chatbots

Combine rule-based and ML approaches.

Hybrid AI Models

Dynamically accesses external knowledge bases for better accuracy.

Retrieval Augmented Generation (RAG)

Robots automate order picking, packing, sorting; reduce errors, boost throughput.

Automated Picking & Packing

AI + IoT for precise tracking, demand forecasting, dynamic slotting.

Real-Time Inventory & Forecasting

Amazon system that speeds up fulfillment by up to 25%.

Amazon's Sequoia System

Consolidates orders, maximizes truck space.

Dynamic Shipment Prioritization

Detects hazards and unsafe behaviors.

Enhanced Safety Monitoring

Data preparation, modeling, and deployment are bundled together; causes bottlenecks and slows down the process.

Monolithic Workflows

Duplicated efforts, versioning issues, scaling hurdles.

Inefficiencies

Modular automation for repeatability, collaboration, and growth.

The Solution

Multiple models repeat data preparation; wasted time and resources.

Volume Overload

Expanding models lead to code duplication and maintenance chaos.

Variety Issues

Data or preprocessing updates require error-prone manual tweaks.

Versioning Nightmares

Step 1 of ML pipeline: define business goals and success metrics clearly.

Problem Definition

Step 2 of ML pipeline: gather and clean data from varied sources efficiently.

Data Collection & Prep

Step 3 of ML pipeline: transform raw data into valuable model inputs.

Feature Engineering

Step 4 of ML pipeline: train algorithms and test performance rigorously.

Model Training & Evaluation

Step 5 of ML pipeline: launch models and track ongoing health and accuracy.

Deployment & Monitoring

Cache reusable steps, eliminate redundancy; save hours on data processing.

Efficiency Gains

Independent work for data engineers, scientists, and ML experts.

Team Collaboration

Single source of truth; CI/CD for models.

Version Control & Automation

CI/CD pipeline stages: Code → Build → Test → Deploy → Monitor.

CI/CD pipeline stages

From CRM, usage logs; seamless intake.

Data Ingestion

Tenure, frequency; build meaningful inputs.

Cleaning & Features

Algorithms used in churn prediction.

Random Forest, Gradient Boosting

Metrics used in churn prediction evaluation.

Precision, recall, F1-score

Deep learning pipeline using Convolutional Neural Network: input image → convolutional layers → fully connected layer → output class.

CNN for Hand Gesture Recognition

ASR → Transcribed String → NLU.

Automatic Speech Recognition Pipeline

Break into reusable steps for easy updates.

Modular Design

Track degradation; alert on performance drops.

Monitoring Tools

Leverage AWS, GCP for on-demand resources.

Cloud Scaling

Validate each stage to catch issues early.

Automated Testing

Use Docker for consistent, portable environments.

Containerization

Large static datasets.

Batch

Real-time data flow.

Streaming

AWS tool for streaming real-time data flow during data ingestion.

AWS Kinesis