NS0-901 AI Overview

AI Overview Detailed Explanation

1. What is Artificial Intelligence?

What does “Artificial Intelligence” mean?

Artificial Intelligence (AI) is the field of computer science that focuses on creating machines and software that can perform tasks that normally require human intelligence. These tasks include:

• Understanding spoken or written language

• Recognizing pictures or sounds

• Making decisions based on data

• Learning from experience

In short, AI tries to make computers “think” or “act” like humans — but it doesn’t mean they are alive or conscious.

A real-life example:

Imagine your smartphone suggesting the next word while you’re typing. That’s a simple example of AI. It has learned from millions of messages how people usually write, and now it predicts what you might want to say.

Key characteristics of AI systems:

• Autonomous: Can work without constant human control

• Adaptive: Can improve performance by learning from data

• Intelligent: Can make logical decisions or solve problems

2. Subtypes of AI

AI is not just one thing. It comes in different levels or types, based on how smart or capable the system is. Here are the three main types:

1. Narrow AI (or Weak AI)

• Definition: AI that is trained and designed for one specific task.

• Example: A facial recognition system that can identify faces in photos but cannot understand language or drive a car.

• Why it matters: Almost all AI in use today is narrow AI. It’s very good at specific jobs but can't do anything outside its training.

2. General AI (or Strong AI)

• Definition: A more advanced form of AI that can understand, learn, and perform any task that a human can.

• Example: A robot that can learn languages, solve puzzles, write music, and hold conversations like a human.

• Important note: General AI does not exist yet. Scientists are still working on it.

3. Super AI

• Definition: AI that is more intelligent than the best human minds in every field — science, creativity, emotions, etc.

• Status: This is still science fiction. It exists only in theory or movies (like in The Terminator or Ex Machina).

3. Relationship with Machine Learning (ML) and Deep Learning (DL)

Understanding the relationship between AI, ML, and DL is essential because these terms are often used together but mean different things.

How are they related?

Imagine three circles, one inside the other:

• The largest circle is AI

• Inside it is Machine Learning (ML)

• Inside ML is Deep Learning (DL)

This means:

• All ML is part of AI

• All DL is part of ML

• But not all AI is ML, and not all ML is DL

1. Artificial Intelligence (AI)

• Goal: To make machines behave intelligently.

• Methods: Can include rules, logic, learning, and more.

• Example: A chess program that follows hardcoded rules is AI, but it may not learn or improve over time.

2. Machine Learning (ML)

• Definition: A subfield of AI where machines learn from data instead of being manually programmed.

• How it works: It uses algorithms to find patterns in data and improve from experience.

• Example: A spam filter that learns which emails are spam by analyzing millions of examples.

Key concept: ML improves automatically as it sees more data.

Types of Machine Learning:

• Supervised learning: Learns from labeled data (e.g., photos labeled “cat” or “dog”)

• Unsupervised learning: Finds patterns in unlabeled data (e.g., customer groups in marketing)

• Reinforcement learning: Learns by trial and error (e.g., teaching a robot to walk)

3. Deep Learning (DL)

• Definition: A subset of ML that uses artificial neural networks — algorithms inspired by how the human brain works.

• Structure: “Deep” refers to the many layers in the network.

• Example: Speech recognition on your phone or self-driving car vision systems.

Deep Learning works best with:

• Large amounts of data

• High computing power (like GPUs)

• Complex tasks (e.g., voice assistants, image recognition)

A simple analogy:

Think of AI as the entire field of medicine.

• Machine Learning is like surgery – a specialized branch.

• Deep Learning is like brain surgery – a very focused and powerful specialty inside surgery.

4. AI Use Cases

Artificial Intelligence is not just a theory or future concept. It's already being used in many important areas. Let’s explore some of the most common and impactful use cases.

1. Healthcare

AI is transforming how doctors diagnose and treat patients.

Examples:

• Medical Imaging: AI can analyze X-rays, MRIs, or CT scans to detect diseases like cancer or pneumonia faster and more accurately than some doctors.

• Predictive Diagnosis: AI can analyze patient records to predict the risk of future illnesses.

• Virtual Health Assistants: Chatbots or voice assistants that help patients book appointments, remind them to take medicine, or provide basic medical advice.

Benefits:

• Faster diagnosis

• Reduced human error

• Personalized treatment plans

2. Finance

Banks and financial institutions use AI to manage risks and improve services.

Examples:

• Fraud Detection: AI models learn to recognize suspicious transactions and block them.

• Credit Scoring: AI analyzes financial history, spending habits, and even social behavior to predict creditworthiness.

• Algorithmic Trading: AI systems make high-speed decisions on stock trading based on real-time data.

Benefits:

• Better security

• Smarter investment decisions

• Faster customer service

3. Manufacturing

AI helps make factories more efficient and safe.

Examples:

• Predictive Maintenance: Sensors and AI predict when a machine is about to fail, so it can be fixed before it breaks.

• Quality Control: Cameras with AI detect defects on assembly lines in real-time.

• Robotics: Smart robots work alongside humans to assemble parts, sort packages, or move heavy objects.

Benefits:

• Less downtime

• Higher product quality

• Lower costs

4. Transportation

AI is changing how we move people and goods.

Examples:

• Self-Driving Cars: AI systems process data from cameras, sensors, and GPS to drive cars safely.

• Route Optimization: AI helps delivery companies choose the fastest or most fuel-efficient paths.

• Traffic Management: Cities use AI to control traffic lights based on real-time traffic flow.

Benefits:

• Safer travel

• Reduced fuel use

• Faster deliveries

5. AI Infrastructure Basics

AI systems often need a lot of computing power, fast storage, and high-speed networks — especially for tasks like deep learning, which involve large datasets and complex calculations.

There are three main layers in AI infrastructure:

1. Compute (Processing Power)

This is the “brain” of the AI system — where all the calculations and model training happen.

Types of processors used:

• CPU (Central Processing Unit)

  • Good for general-purpose computing

  • Slower for training large AI models

  • Still useful for small-scale inference (making predictions)

• GPU (Graphics Processing Unit)

  • Designed for parallel processing

  • Much faster than CPUs for training deep learning models

  • Widely used in AI labs and data centers

• TPU (Tensor Processing Unit)

  • Custom-built by Google

  • Optimized for deep learning using TensorFlow

  • Extremely fast for large-scale model training

Other examples:

• FPGA (Field-Programmable Gate Array): Customizable processors for specific tasks

• ASIC (Application-Specific Integrated Circuit): Built for a single function — very efficient

2. Storage (Data Storage Systems)

AI systems require fast and scalable storage to handle large volumes of training data — such as images, videos, or sensor data.

Common storage types:

• File Storage

  • Stores data in a traditional folder and file system

  • Easy to use but may not scale well for very large datasets

  • Example: NFS (Network File System)

• Object Storage

  • Stores data as “objects” with metadata and unique IDs

  • Scales easily for huge datasets

  • Ideal for unstructured data like images or logs

  • Example: Amazon S3, NetApp ONTAP S3

• Parallel File Systems

  • Designed for high-performance access to massive datasets

  • Splits files across multiple storage nodes

  • Examples: Lustre, BeeGFS

Important features in AI storage:

• High throughput (speed of reading/writing data)

• Low latency (minimal delay)

• Scalability (can grow as needed)

3. Networking (Data Transfer Between Systems)

AI workloads often require large amounts of data to be transferred quickly between servers, storage devices, and processors (like GPUs). A fast and efficient network is essential.

Key networking technologies:

• Ethernet

  • Common in general-purpose computing

  • May be slower for large-scale AI

• InfiniBand

  • High-speed, low-latency communication

  • Ideal for AI clusters or high-performance computing

• RoCE (RDMA over Converged Ethernet)

  • Allows data transfer directly between computers’ memory without using the CPU

  • Reduces latency and improves GPU-to-GPU communication

Putting It All Together:

A full AI system in a company or lab might include:

• Dozens or hundreds of GPUs working in parallel

• A shared storage system that holds terabytes or petabytes of data

• A fast network like InfiniBand to keep all the pieces connected

This infrastructure allows researchers and engineers to train models faster, test more ideas, and serve predictions to users in real-time.

AI Overview (Additional Content)

1. AI Modes: Training vs Inference

Training and Inference are the two main operational modes in the AI lifecycle, each with different goals, resource requirements, and deployment environments.

Training Phase

• Purpose: To teach the AI model by feeding it data so it can learn patterns and relationships.

• Processes:

  • Forward and backward propagation

  • Gradient descent and weight optimization

• Hardware Requirements:

  • High-performance GPUs or TPUs are typically needed due to the massive volume of matrix operations.

  • May require multi-node clusters for distributed training.

• Duration: Time-consuming (can range from hours to weeks depending on model size and dataset).

• Location: Often done in cloud platforms or dedicated on-premise data centers.

• Tools Used: TensorFlow, PyTorch, distributed frameworks (Horovod, Ray, etc.)

Inference Phase

• Purpose: To use the trained model to make predictions on new data.

• Processes:

  • Only forward pass (no backpropagation)

• Hardware Requirements:

  • Lightweight models can run on CPUs.

  • Real-time, high-throughput tasks may still use GPUs or ASICs.

  • For latency-sensitive applications, edge devices are commonly used.

• Location: Cloud, on-premise servers, or embedded edge devices (phones, cameras, IoT devices).

• Tools Used: TensorFlow Serving, NVIDIA Triton, ONNX Runtime, REST APIs

Key Differences:

Feature	Training	Inference
Objective	Learn from data	Make predictions
Frequency	One-time or periodic	Continuous
Hardware	GPU/TPU intensive	CPU or lightweight hardware
Complexity	High	Lower
Latency	Not critical	Often low-latency required

2. AI Deployment Models: Cloud vs On-Premises vs Edge

AI systems can be deployed on various platforms, each offering different levels of control, performance, cost, and latency.

Cloud Deployment

• Definition: Hosting AI workloads on platforms like AWS, Azure, Google Cloud.

• Advantages:

  • Scalable and flexible resources

  • Access to managed services (SageMaker, Vertex AI, etc.)

  • Lower upfront costs

• Disadvantages:

  • Data transfer latency

  • Ongoing subscription costs

  • Regulatory concerns for sensitive data

On-Premises Deployment

• Definition: Hosting the entire AI infrastructure in a local data center.

• Advantages:

  • Full control over data and hardware

  • No dependency on internet connection

  • Suitable for regulated industries

• Disadvantages:

  • High capital expenditure

  • Requires in-house technical expertise

  • Less elasticity in scaling

Edge Deployment

• Definition: Deploying AI models directly onto local devices (smartphones, cameras, drones).

• Advantages:

  • Extremely low latency

  • No dependency on cloud or network

  • Ideal for privacy-sensitive use cases

• Disadvantages:

  • Limited compute and storage

  • Requires model compression and optimization

  • Harder to manage and update at scale

3. Integration of AI with HPC and Big Data Analytics

AI workloads increasingly intersect with High-Performance Computing (HPC) and Big Data Analytics, forming a unified ecosystem.

AI + HPC

• HPC platforms provide the parallel compute power necessary to train large-scale deep learning models.

• GPU clusters and supercomputers (like NVIDIA DGX or NetApp + SuperPOD) are used for high-volume AI training.

AI + Big Data (Analytics)

• AI systems consume large volumes of data from platforms like:

  • Apache Hadoop (distributed file storage and processing)

  • Apache Spark (in-memory processing engine for streaming and batch data)

• Integration examples:

  • Using Spark to preprocess massive datasets and pipe them into a PyTorch or TensorFlow model.

  • Merging business intelligence tools with ML pipelines for predictive analytics.

Trend: Converging AI + Analytics + HPC forms the foundation for modern data-driven enterprise architecture.

4. Current Challenges in Modern AI Architectures

While AI continues to advance, several critical challenges persist:

a. Cost of Large Models

• Training large language or vision models requires immense compute and storage.

• Energy consumption and carbon footprint are growing concerns.

b. Data Privacy and Governance

• Managing sensitive data (e.g., medical, financial) while training models is complex.

• Regulations (GDPR, HIPAA, etc.) require strict compliance in both data handling and inference pipelines.

c. GPU Resource Bottlenecks

• GPU shortages or allocation mismanagement can severely delay training.

• AI workloads often compete for limited GPU capacity in shared clusters.

d. Maintenance Complexity

• AI models degrade over time due to data drift.

• Continuous monitoring, updating, and retraining are resource-intensive.

These issues serve as the foundation for the broader "AI Common Challenges" module and are crucial for architectural planning in real-world systems.