💻 Y9 - SP2.1 - Artificial Intelligence

Slide Title

Create a OneNote page and copy this into the title:

Lesson 1 - Spring Term 2 - Introducing AI


Read the next few slides and answer the mini-whiteboard question, once your teacher has posted it

📖 What is Artificial Intelligence? (1)

Artificial Intelligence is technology that appears to behave like human intelligence.

  • When prompting Copilot, the responses almost appear to have been generated by another person.
  • When asking Alexa or Google Music to play a song, it can feel like a DJ selecting and playing music for you.
  • Self-driving cars appear to make intelligent decisions to avoid obstacles and navigate complex manoeuvres such as parallel parking.

What is Artificial Intelligence? (2)

  • While it may appear to do so, AI does not "think" like humans.
  • It is very good at identifying patterns in data.

For example: When I ask a large language model, "How are you feeling today?", it predicts what it is supposed to respond with.

There is not really an intelligence behind it — just clever design.

🐻 Mini Whiteboard Question

The leading question will be: Why is the term Artificial Intelligence problematic?

Open MWB Content

Learning Objectives

  • LO1: Explain what is meant by Machine Learning
  • LO2: Compare Narrow AI and Artificial General Intelligence (AGI)

Traditional Programming vs Machine Learning (1)

For most of their history, computers have only been able to follow instructions written by humans.

For example, this code below:

q1 = input("What is the capital of France? ")

if q1.lower() == "paris":
    print("Correct")
else:
    print("Incorrect")

Traditional Programming vs Machine Learning (2)

A good example of human-generated code can be found in computer game NPCs (non-playable characters).

  • Although a non-playable character (NPC) acts as if it has its own free will, everything it does has been pre-determined by the programmer's instructions.
  • In Minecraft, when your character offers a villager some bread, they are coded to behave in a certain way in response.
  • You may observe lots of seemingly independent behaviour and mannerisms from the villager; however, they are all pre-determined (decided beforehand).

Traditional Programming vs Machine Learning (3)

The type of programming where a human has to think of and then code every possible scenario is often referred to as Traditional Programming.

For decades, this has meant the development of software requires hours of painstaking and highly focused work to get it right.

Traditional Programming vs Machine Learning (4)

Machine Learning differs greatly from Traditional Programming.

  • Machine Learning relies upon an algorithm that improves itself over time.
  • This is done through positive reinforcement each time it produces the correct outcome.
  • A bit like training a dog.

Traditional Learning vs Machine Learning (5)

💻 Machine Learning is now part of many aspects of modern life.

Software developers (coders) are still an important part of the process; however...

  • They now develop applications that are capable of learning what to do by themselves.
  • Software developers even use AI to write large (generic) parts of code, saving them time.

That said, the process of coding new software still requires an expert programmer to make high-level adjustments to the software.

Machine Learning (1)

Machine Learning requires lots of training data.

A simple example of AI would be a spam filter on an email account (spam is junk email nobody wants):

  • The spam filter is given tens of thousands of emails to examine at a time.
  • A human trainer has already identified which are spam and which are legitimate emails.
  • Each time the spam filter successfully identifies a spam email, the trainer flags this, and the spam filter learns what it is doing correctly.
  • If it makes a mistake, the trainer flags this, and the spam filter aims to adjust its algorithm for future attempts.
  • Over time, the algorithm becomes highly effective at its job.

Machine Learning (2)

Another example of Machine Learning would be the training of Large Language Models (LLMs) such as ChatGPT and Google Gemini.

  • The LLM is given vast libraries of text.
  • A human trainer then prompts the LLM with questions.
  • The LLM predicts which words to respond with and in what order.
  • As with the spam filter, good and bad responses are flagged.
  • Over time, the LLM starts to predict exactly how it should respond.
  • In many cases, these responses are strikingly realistic.

📝 Machine Learning Activity

Copy the section below into OneNote.

📝 Machine Learning vs Traditional Programming

Explain how Machine Learning works.

In your explanation, refer to the following:

  • Training data
  • How the process of training an AI model is controlled
  • Examples of how Machine Learning has been used to train specific AI models

Narrow Field AI (1) - Definition

Some terms you may begin to hear more often over the next few years are Narrow Field AI and Artificial General Intelligence (AGI).

We will begin by looking at what is meant by Narrow Field AI:

  • All Narrow Field AI is designed for one specific task.
  • This specific task may be complex (such as the generation of text by ChatGPT).
  • However, it is still the only thing the AI can do.
  • All of the AI people are currently using is Narrow Field AI.

Narrow Field AI (2) - LLMs

The most common form of Narrow Field AI in use is something called a Large Language Model (LLM).

  • LLMs respond to text prompts from the user.
  • They have been trained on vast quantities of text sources.
  • They are capable (from their training) of predicting how they should respond to the user.
  • They are even capable of writing code.

Narrow Field AI (3) - Self-Driving Cars

Yes, these are actually a real thing and perform overwhelmingly well in safety tests.

  • The system is given sat-nav coordinates for its destination.
  • It maps out the most optimal route.
  • It drives the car.
  • It constantly analyses visual and auditory data through sensors and cameras.
  • It has been trained to react to everything on the road like a very cautious driver.
  • It can even parallel park.

Artificial General Intelligence (1)

Both examples of Narrow Field AI are not capable of doing each other’s job.

  • There is no way an LLM can learn to drive a car.
  • There is no way a self-driving car will accurately produce text or code.
  • Their initial programming was to learn how to perform a specific task.

Artificial General Intelligence (2)

It is predicted that one day there will be AI that is capable of learning how to do a myriad of different tasks:

  • The term we use for this is Artificial General Intelligence (AGI).
  • This type of AI may surpass human intelligence.
  • Some people are concerned about the impact this technology could have.

🐻 Mini Whiteboard Question

The leading question will be:

What will be the positive and negative impacts of Artificial General Intelligence?

Open MWB Content

📝 Plenary Activity

Copy the section below into OneNote

📝 Reflection on today's learning objectives

Explain how Machine Learning works.

Give an overview of the following:

  • The difference between Traditional Programming and Machine Learning
  • The difference between Narrow Field AI and Artificial General Intelligence (AGI) (giving examples of current Narrow Field AI)
♿