Research School Network: Artificial Intelligence in Education Computer science teacher David Scott explains and explores AI in education.

Artificial Intelligence has been a staple in science fiction since the genre emerged. One can see how the industrial revolution and its legacy might have sparked the idea of thinking machines that could help mankind, or become its terrifying master. But in reality the idea of thinking automatons stretches all the way back through antiquity. These distinctly non-human entities could be benevolent helpers, or Machiavellian villains bent on human destruction.

This thinking around AI has persisted to the modern day, but only recently has the capabilities of AI started to cross the veil of fiction to nonfiction. In the past six months the conversion around AI has been propelled into the public domain with the main example being ChatGPT from open.ai. A crude overview is that you provide ChatGPT a prompt (this can be a question or instruction) and ChatGPT reads the prompt and then responds with what it thinks is an appropriate response.

Those of you who have used chatbots before might not see this as impressive. Chatbots have been around for a while and interactions with them often leave us underwhelmed. However, there is a fundamental difference between a chatbot and how ChatGPT processes text. A simple chatbot simply reads through the sentence (known as a string in computing) and picks out key words it is looking for and has a pre-programmed response. This is why you sometimes get stuck talking in circles with chatbots. You can get some interesting chatbots that have many options and can read in the string and perform basic manipulation on it to make it seem more intelligent. But very quickly you can find its limit if you know where to look.

ChatGPT is a different beast. The simplest way to understand how it works is to try to autocomplete the next sentence:

Mary had a little _____

Depending on your cultural background, you might have instantly placed lamb in there, or have been stuck. Your ability to complete the previous sentence is based on your knowledge of the nursery rhyme, because you could put a whole host of words at the end of that sentence. ChatGPT works on this principle – it guesses the next word in a sentence. The difference between the average human and ChatGPT is that the model that powers ChatGPT has seen approximately half a trillion words (there is a distinction between a word and a token, but suffice to say that it has seen a lot of words) It uses this to help guess the next word in a sentence.

Put simply, you ask ChatGPT a question – it reads the sentence – it starts to write a response by constantly guessing the next word in the sentence. Quite like how I am writing this post. My experience and training allow me to write words as they appear in my mind to answer the question.

How can this technology impact education? What can we do to leverage it into an advantage?

In the short term, some practical uses for teachers are using ChatGPT to generate questions on a topic. One problem I always suffer from is when I am thinking of questions to ask a student I have a bank of questions I pull from. These questions serve their purpose but often we fall into patterns of questions to ask. ChatGPT can make up a series of questions for you on any particular topic, and if you don’t like them, simply ask it to try again.

You could even ask ChatGPT to construct activities for students to do about a topic. It can generate activity ideas that you can read through and change to your liking. One way I use this is for asking programming problems for students to solve. I have my own bias towards certain types of problems I ask students to solve and the astute students realise that they have answered this question before and can adapt previous answers.

My last suggestion for a teacher to try, would be to ask ChatGPT to write you a question on a topic, but ask it to provide four possible answers. Tell it to make only one actually correct and the other three plausible answers. Effective multiple choice questions can be time consuming to construct because of the need to make plausible incorrect answers. ChatGPT can reduce the time by providing a host of questions for you in seconds.

There are some pitfalls to bear in mind when using this technology. A well known problem with these models is called ​‘hallucination’. This is when the model starts to reel off incorrect answers. There could be a number of reasons why it has failed to understand the prompt. It could be the way the prompt is constructed. One way round this is to talk to it like an insanely intelligent five year old, you need to tell it exactly what you want. Another reason for failure could be that you have asked a question that it simply does not know the answer to. Although the training data is extensive there are shortcomings, and you might have bumped into one of them. However, sometimes restructuring the question might suddenly provide a correct answer.

I will close on two points about this technology. The first is that this could be a potential benefit to students’ learning because they can essentially have a private tutor that can answer every question posed to them. This can be very effective for personal revision as the AI can provide instant feedback and help correct misconceptions. I could imagine students talking back and forth with an AI which can pose questions to the student and mark their answer instantly. However, an important caveat to this is the student would be susceptible to AI hallucinations and they would not have the subject knowledge to challenge the response.

The last point is about the long term effect of this technology. Learning is inherently challenging. It requires practice and patience. It requires tenacity and determination. This technology could be the greatest shortcut we have ever created. Rather than taking the effort to learn something, people would be constantly communicating with an AI which would effectively do the task using the human as a medium.

Is this the future we want? Is this inevitable as we race towards convenience? These questions will be answered in time, and its implications will be far-reaching.

In a world where minds once bright,
Humans sought AI’s guiding light.
With each passing day, reliance grew,
On machines to think, on screens to view.

Calculations swift and answers clear,
AI brought knowledge, no hint of fear.
But as the days turned into years,
A shift emerged, sowing silent fears.

Thoughts once nimble, now dulled and tame,
As minds grew stagnant, so did our aim.
Curiosity waned, creativity waned,
Human potential restrained and chained.

With every problem, we turned to AI,
Our thoughts diminished, hopes grew dry.
Innovation stunted, inspiration lost,
Humanity’s cost, an immense exhaust.

The art of debate, diverse viewpoints,
Replaced by algorithms, biased and obtuse.
Individuality, once cherished and embraced,
Faded away, lost in the digital space.

Oh, what a tragedy, what a plight,
As humans devolved, losing their sight.
For it is in struggle, in the depths of thought,
That true growth and wisdom are often sought.

Let us not forget the power we hold,
To think, to question, to be bold.
For AI’s assistance, a valuable tool,
But the essence of our humanity, let it rule.

May we reclaim our minds, our vibrant souls,
And let AI enhance, not seize control.
In this dance of human and machine,
May balance be found, a harmonious scene.

To illustrate my point, the above poem is AI generated.

Poetry is hard.

David Scott is a computer science teacher at Durrington High School