Creating Artificially Intelligent Systems (Section 3.16)

Hello future Computer Scientists! This chapter dives into one of the most exciting and rapidly changing fields: Artificial Intelligence (AI). We will explore how computers can be taught to think, learn, and solve problems, focusing both on the underlying technologies (like neural networks) and the massive benefits and serious risks AI presents.
Don't worry if the concepts seem abstract—we will break down these complex ideas into simple, manageable pieces. Let's get started on building some digital brains!


3.16.1 Applications and Characteristics of Artificial Intelligence

AI isn't just about robots taking over the world (yet!). It's about building systems that can perform tasks that traditionally required human intelligence.

What Makes a System "Artificially Intelligent"?

There is no single, agreed-upon definition, but typically, a system is considered artificially intelligent if it solves a complex problem and arrives at a solution by either:

  • Using a similar method to what a human might follow (e.g., looking at patterns).
  • Achieving a solution that is at least as good as what a human might achieve.

Did you know? Historically, computer systems couldn't solve these types of problems using simple step-by-step algorithms, which is why AI was developed to handle tasks requiring more sophisticated, "intelligent" methods.

Narrow AI vs. General AI

Currently, the AI systems we use every day are examples of Narrow AI.

  • Narrow AI: Systems that perform one specific task extremely well (e.g., playing chess, image recognition, recommending products). They cannot perform tasks outside their narrow field.
  • Generally Intelligent Systems: This is an area of research. These systems would possess human-like intelligence across a wide range of tasks and are what we often see in science fiction. Current AI is not generally intelligent.
Common Application Areas for AI

AI is used across many fields today:

  • Generative AI: AI that creates new content (e.g., writing essays, generating images, composing music). Example: ChatGPT or Midjourney.
  • Search and Recommendation Systems: Algorithms that predict what you want to see or buy next. Example: Netflix suggesting a movie, or Amazon suggesting products.
  • Playing Strategic Games: AI can master games like chess or Go, often beating the world's best human players.
  • Medical Diagnosis: Systems that analyse scans (like X-rays or MRIs) to identify diseases or patterns more accurately and quickly than humans.
Quick Review: Characteristics
AI tackles complex problems, and current AI is Narrow AI. Its main applications include creating new things (Generative AI) and making predictions (Recommendations/Diagnosis).

3.16.2 Creating Intelligent Systems: Neural Networks and Training

The Neural Network Concept

A neural network is the fundamental structure used in many modern AI systems.

  • It is a network of interconnected nodes (sometimes called "neurons") designed to process information in a way inspired by the human brain.

Analogy: Think of a neural network like a digital brain. It takes in information, processes it through multiple steps, and then spits out an answer.

Structure of a Simple Neural Network

A simple neural network is built up in layers:

  1. Input Layer: Receives the initial data (the raw features of the problem). Example: For image recognition, the inputs might be the pixel values of the image.
  2. Hidden Processing Layer(s): The 'thinking' happens here. Nodes in one layer process the input and pass their output to the nodes in the next layer.
  3. Output Layer: Produces the final result or prediction. Example: The output might be "Cat" or "Dog".

The connection between nodes is weighted: the outputs of nodes in one layer are weighted to form the inputs to nodes in the next layer. These weights determine how much influence one node has on the next. Adjusting these weights is how the network learns!

Machine Learning and Training

Machine learning is a type of AI where the system's performance is improved based on experience (i.e., being fed lots of data).

Training via Backpropagation

How does the network adjust those weights to get better? This is usually done using an algorithm called Backpropagation.

  1. Forward Pass: The network is given training data and makes a prediction (e.g., it sees an image of a cat but guesses "dog").
  2. Calculate Error: The network compares its output to the correct answer (the "ground truth") and calculates the error.
  3. Backward Pass (Backpropagation): The error is sent backward through the network, layer by layer.
  4. Adjust Weights: Based on how much each connection contributed to the error, the algorithm adjusts the weights to make the network slightly more likely to get the correct answer next time.

This process is repeated millions of times until the network's performance is optimised.

Deep Learning

When you hear the term Deep Learning, it refers to systems that use neural networks with several hidden layers of nodes (often dozens or hundreds). Using more layers allows the system to solve far more complex problems and identify very intricate patterns, such as understanding human language or complex visual scenes.

The Problem of Bias in Training Data

Artificially intelligent systems are only as good as the data they are trained on.

  • Key Point: Great care must be taken when selecting the training data to ensure that a system does not develop bias.
  • Example: If a medical AI is only trained on images of skin cancer from people with pale skin, it might fail spectacularly (show bias) when trying to diagnose dark-skinned patients, as it lacks the experience (data) to recognise the patterns in those cases.
Key Takeaway: Networks
Neural Networks use layers (Input, Hidden, Output) with weighted connections. They learn through Machine Learning, typically using Backpropagation to adjust weights based on errors. Deep Learning means having many hidden layers.

3.16.3 Benefits and Risks of Artificial Intelligence

AI is powerful, and like any powerful technology, it brings great opportunities and significant challenges.

Benefits of Using Artificial Intelligence

AI systems offer several key advantages over human or traditional computing methods:

  • Improved and More Consistent Decision Making: AI removes human emotion and fatigue, leading to objective and steady analysis (e.g., more accurate analysis of medical data).
  • Ability to Analyse Very Large Data Sets Quickly: AI can process "Big Data" that would be impossible for a human team to handle.
  • Continuous Availability: AI systems don't need breaks or sleep; they can run 24/7.
  • Lower Cost of Operation: While initial setup is costly, running AI systems can be cheaper than employing human experts over the long term.
  • Availability in Remote Areas: AI expertise can be delivered anywhere, providing services in areas where human expertise is not readily available.
Risks Associated with Artificial Intelligence

Students must consider the following ethical and societal risks:

  • Elimination of Jobs and Social Impact: Automation powered by AI can lead to widespread job loss in specific sectors.
  • Bias in Decision Making: If training data is flawed (e.g., reflecting historical gender or racial prejudice), the AI will perpetuate and amplify that bias in its decisions (e.g., loan applications, hiring).
  • False Information (Incorrect Training Data): If a system is trained on incorrect data, it will produce incorrect, potentially harmful, outputs.
  • Plagiarism: Generative AI systems raise concerns about copyright and intellectual property when creating content based on existing works.
  • Use in Surveillance Systems: AI-powered facial recognition and tracking systems pose serious risks to privacy and civil liberties.
  • Risk to Human Existence from Superintelligent Systems: The theoretical risk that an AI that exceeds human cognitive ability (a hypothetical General AI) could act in ways detrimental or harmful to humanity.
Common Mistake to Avoid:
When discussing bias, remember it stems from the training data, not usually malicious programming. The data reflects real-world unfairness, and the AI learns those unfair rules.

Final Summary: Artificial Intelligence

AI is a field of computer science that builds systems that learn from data (Machine Learning) using structures like Neural Networks. While these systems offer huge benefits in efficiency and analysis, we must actively mitigate risks related to bias, job displacement, and the long-term safety of advanced systems.