🎯 Learning Objectives
Develop the Information Technology learning strand:
- Define artificial intelligence and machine learning
- Explore examples of where they are being applied
- Teach a machine how to recognise different types of images
- Discuss moral issues associated with these technologies
💬 Key Vocabulary
- Artificial intelligence
- machine learning
- data
- training
- testing
- programming
📝 Starter Activity – Words Matter
Think, write, pair, share – Use the worksheet below to write down your answers and discuss them in pairs.
Define ‘artificial’
examples: flavour, flower, hair, heart, lake, light, materials
Provide synonyms for ‘intelligent’
synonym – a different word with the same meaning
When would you call a person, an animal, or a machine ‘intelligent’?
📖 Defining artificial intelligence
Question
What is artificial intelligence?
Answer (suggestion)
Any machine that performs tasks that typically require intelligence in humans
There is no single, agreed definition for artificial intelligence.
- Because words like ‘intelligence’ and ‘thought’ are very difficult to pin down.
- Because a machine considered ‘intelligent’ now will probably be commonplace in a few years.
📖 Artificial intelligence: what it is not (yet)
We are years away from achieving the kind of general AI portrayed in books and films.
At present, artificial intelligence research mostly focuses on individual aspects of intelligent behaviour.
The legendary HAL 9000 computer, from the film
2001: A Space Odyssey
📖 Can a machine do this?
| Task | Progress so far |
|---|---|
| Play board games: e.g. checkers, chess, Go | Checkers was solved in 2007: computers play perfectly. Deep Blue by IBM beat the top human player in chess in 1996. Humans haven’t beaten a top chess program since 2005. AlphaGo by DeepMind beat the top human player in Go in 2017. |
| Prove mathematical propositions | Automated provers have deduced thousands of known or new propositions and also discovered shorter proofs. |
| Planning and scheduling | Computers are used extensively in manufacturing, crew scheduling, self-driving vehicles, and space exploration. |
Questions
- Do these tasks require ‘thinking’ by humans?
- Do you think computers can perform these tasks well?
AI has by now succeeded in doing essentially everything that requires ‘thinking’ but has failed to do most of what people and animals do ‘without thinking’ – that, somehow, is much harder!
Donald Knuth, author of The Art of Computer Programming, in 1981
| Task | Progress so far |
|---|---|
| Identify objects in images | Accuracy has jumped from 50% to 90% since 2011 |
| Identify words in sound | Major advances since 2009 Error rates have dropped to around 5% Comparable to professional transcribers |
| Generate speech from text | Major advances in 2016 Now almost indistinguishable from a real human voice |
| Handle and manipulate objects | Robotic arms that pick up objects constantly improving Mostly in research phase as of 2020 |
| Walk | Two- and four-legged robots constantly improving Mostly in research phase as of 2020 |
📖 Can a machine do this?
The scientists below had this to say in the past:
Every time we figure out a piece of it, it stops being magical; we say, ‘Oh, that’s just a computation.’
Rodney Brooks, director of MIT Artificial Intelligence Lab, from a 2002 Wired magazine article
Once something becomes useful enough and common enough it’s not labelled AI anymore.
Nick Bostrom, director of the Future of Humanity Institute at Oxford University,
from a 2006 article at cnn.com
AI is whatever hasn’t been done yet.
Douglas Hofstadter, in his 1979 book Gödel, Escher, Bach
| Task | Progress so far |
|---|---|
| Hold a conversation | Holding an open-ended conversation with a human is considered a benchmark for AI (the Turing test) No chatbot has really achieved that goal as of 2020 |
| Translate between languages | Major advance by Google in 2016 Systems produce useful output, but still an open problem |
| Understand and answer questions | Watson by IBM beat the top human players on Jeopardy! in 2011 It is capable of providing evidence to justify its answers |
| Drive a car | Highly complex problem Major breakthroughs since 2005 |
| Diagnosing medical images | Cases of performance comparable to human experts reported since 2012 |
Most of these recent advances in artificial intelligence are due to breakthroughs in machine learning.
📖 AI development so far
Goal: Create a machine that performs a specific task.
Method: Program the machine to perform the task.
Provide the machine with explicit instructions.
For some tasks, providing explicit instructions is far too complicated.
Tasks
- Identify objects in images
- Identify words in sound
- Generate speech from text
- Handle and manipulate objects
- Walk
- Hold a conversation
- Translate between languages
- Understand and answer questions
📖 AI development from now on – Machine Learning
Goal: Create a machine that performs a specific task.
Method: Teach the machine to perform the task.
Provide the machine with examples
Program the machine to learn from examples
This is called ‘supervised learning’.
Programming computers to learn from experience should eventually eliminate the need for much of this detailed programming effort.
Some Studies In Machine Learning Using the Game of Checkers
Arthur Samuel (1959)
Machine learning does not eliminate programming.
It replaces the problem of programming a machine to perform a task with two separate problems:
- programming the machine to learn, and providing it with the necessary training
📝 Activity 4 – Be the teacher
Teach your computer to tell the difference between apples and oranges, using Google Teachable Machine.
Follow the instructions in the worksheet below.
📝 Plenary – Thinking beyond ‘coolness’
Look at these applications of AI, what moral considerations should be made when developing these systems?
- Self-driving cars
- Medical diagnosis
- Banking – detecting fraud, approving loans
- Automation – performing tasks instead of humans
Answer
- Self-driving cars – Who is responsible in an accident? (Accountability)
- Medical diagnosis – How can decisions be explained? (Transparency)
- Banking – detecting fraud, approving loans – How can we guarantee that machine training does not lead to discrimination? (Bias) How can decisions be explained? (Transparency)
- Automation – performing tasks instead of humans – How will humans handle lower demand for labour? How will the benefits of AI be fairly distributed?
In this lesson, you…
- Defined artificial intelligence and machine learning
- Explored examples of where they are being applied
- Taught a machine how to recognise different types of images
- Discussed moral issues associated with these technologies
Next lesson, you will…
- Take a quiz, to assess learning
- Explore the implications of sharing programs, and learn about free and open source software
🏅 Badge it
🥈 Silver Badge
- Complete the Starter Activity – Words Matter and upload the Word document to www.bournetolearn.com.
🥇 Gold Badge
- Complete Activity 4 – Be the teacher and upload screenshots of your model to www.bournetolearn.com.
🥉 Platinum Badge
- Complete Activity 4 – Be the teacher – Explorer Task and upload screenshots of your model to www.bournetolearn.com.