In this microcourse, you’ll learn about one of the fundamental ideas in AI: general AI.
Could a computer ever be better than a human at a specific task? Of course - and you will see some examples in the AI BASICS microcourse. This kind of AI is often called weak AI.
An AI that could work just like the human mind does and could intelligently do many of the tasks that a human can do would be strong AI, or as it is often known, General AI.
But is this possible? Could we build AI that is as intelligent as (or maybe more intelligent than) a human?
So far, no one has managed to do this, but some AI systems are very intelligent indeed.
In this microcourse, we will learn about general AI by first looking at the difference between rules-based AI and pattern-matching (or machine learning) AI. You’ll see and try out some examples of both of these in this microcourse.
We will learn that rules-based AI can solve simple problems, but if we want to do things that are as complex as those that humans do, we need something more powerful that can learn how to solve problems.
Let’s get started and learn about the difference between rules-based and pattern matching AI before we think about whether general AI is even possible.
RULES BASED AND PATTERN MATCHING AI
Watch this short video from Haileybury teacher Toan Huynh that looks at the differences between these two types of AI.
RULES BASED AI
Let’s learn more about rules-based AI by looking how an AI plays a simple game.
Could you beat it? Does the AI always win? And, do you think it matters who goes first?
Now watch the following video which talked about how computers play games - naughts and crosses but also more sophisticated games like Go.
How do you think I programmed the naughts and crosses game to never lose? Do you think believe the naughts and crosses program is ‘intelligent’.
Now watch the following video from Haileybury teacher Jeff Plumb.
You should be able to see that the program doesn’t choose the simplest way to win. Does this change your previous answer about 'intelligence'?
CREATE YOUR OWN AI USING MACHINE LEARNING
We’re going to use Teachable Machine, a simple machine learning AI system, to build an image classifier.
Follow along with Haileybury teacher Toan Huynh as he leads you through the process and how machine learning techniques teach the AI how to recognise a cat or a dog.
CAN WE CREATE GENERAL AI?
You should see that the example of Teachable Machines seems to be more intelligent than simple rules-based AI. It certainly does something that humans can do - figure out what is a cat.
As you will see in the AI BASICS microcourse, some of the more sophisticated AI techniques - including machine learning, deep learning and neural networks - can do things that humans can do and often faster and more accurately. They can recognise faces, diagnose diseases and create images that look real but are, in fact, not real at all (what are called ‘deepfakes’).
But does AI do it in the same way that humans do it? Does it matter? Could machine learning make AI as smart - or smarter than - humans? Is AI ‘intelligent’?
Can we create a powerful general AI?
That's not an easy question to answer.
THE TURING TEST
One scientist thought he had the answer.
The mathematician Alan Turing proposed something called the imitation game (or the Turing Test). It’s a thought experiment - something to make you think about a problem.
The idea behind it was to figure out if a computer could fool a human into thinking that it was a human. If so, the computer could be considered to be intelligent.
Imagine three participants in the game: someone who has to ask questions, a computer, and another human.
The only way they can communicate is through text. The questioner aims to figure out by asking questions, which is the computer and which is the human.
If the computer doesn’t get found out, it has passed the test and can be considered intelligent.
Could you figure it out? What questions would you ask?
Recently, Amazon launched The Alexa Prize, a competition for university students. Teams are challenged to design conversational AI that Alexa customers can interact with for 20 minutes or more.
So maybe it's not about fooling anyone - but making it easier and more interesting to interact with AI.
You can learn more about the Turing Test in this video, and why not ask Hailey in AI ETHICS microcourse about what she thinks about the Turing Test.