“I’m a teacher and the subject of Artificial Intelligence isn’t relevant in my subject; for I teach in a subject that doesn’t need to use computers”.
Even if you are not a teacher, I’m sure that many of you reading this will empathise with this statement.
In this article, I’m hoping to help you understand what is about to take place in the educational space. It is a point in the teaching history timeline with the biggest opportunities and threats that we have likely ever faced before – game changers. This is bigger than the invention of the computer itself, bigger than the invention of the Internet.
The fact that Sherborne School educated one of the pioneers and fathers of AI, Professor Alan Turing, makes this article even more poignant for me as head of computer science here. The Turing Test was developed by him as a measure to the extent to which machines can “think” in a fashion which is indistinguishable to humans.
“The fact that Sherborne School educated one of the pioneers of AI, Professor Alan Turing, makes this article even more poignant.”
In my opinion, we are at a line in the sand which will soon be crossed, and Turing’s “Imitation Game” will become an interesting milestone relegated to the history books. Some people think that we are already just crossing that line.
In the last few months, several new AI engines have been released to the public for beta-testing. The internet euphoria around this has been unprecedented and countless tens of thousands of demonstrations exist on YouTube on how lives of artists, programmers, lawyers and other professions are doomed into redundancy. Certainly, if you are a graphic artist for an advertising agent, I would be looking to re-skill pretty quickly.
For example, DALL-E 2 was asked to create an image of “A 52-year-old, balding, red-headed computer scientist writing an article on AI for the school magazine sitting outside Sherborne School.” You’ll see from the main picture on this article, it didn’t do a bad job.
“I’m safe though – people will always need computer science teachers”. Until recently I thought this was true. I am a mathematician originally and used to lecture mathematics at university many moons ago.
I also spent over 20 years in industry writing code and designing computer-based exam systems for awarding bodies. If any of you qualified to be an accountant in the late 90’s or noughties, you are likely to have sat an exam which was presented to you in a system written by myself.
“Until recently I thought the job of computer science teacher would be safe from being made redundant from AI.”
Computer science teachers are one of the rarest commodities that we have in our profession. Schools find it almost impossible to recruit them with the levels of experience and skill that they do in so many other subjects. But, in the last few weeks after playing with more and more AI engines, I am starting to realise that in fact it is me who is more likely to be redundant than any other subject teacher.
Chat GPT is one of the latest engines created from Elon Musk’s funded company OpenAI. To my utter amazement this thing can write code. Good code! Really good code! That is right, machines that have been coded can write the code for other machines. Arnold Schwarzenegger will be teaching my Upper Sixth next year, I’m sure. Anyone who thinks that the AI engines are superficial or stupid are, arguably, deluded. They are not and they are getting better exponentially every year.
“Machines that have been coded can write the code for other machines.”
For example, I presented Chat GPT with an A-level computer science question worth six marks and an answer from a student copied and pasted from a students’ workbook. It gave a useful critique of the answer and a mark – which it updated once I had presented it with the mark scheme and explained how it worked.
Another student’s answer was fed into GPT (see how I’m starting to like him more by dropping his formal title?) and the response was really exciting. GPT understands WHY aspects of the student answer should NOT gain marks. Furthermore, it highlighted a point which is common among students when being faced with questions to compare pros and cons – they don’t, but instead only see one side of the argument.
GPT could even provide a model answer that would score full marks, although at first it used bullet points. When prompted, it was happy to put these into full sentences.
“Arnold Schwarzenegger will be teaching my Upper Sixth next year, I’m sure.”
I hand-marked the boy’s answers to this Hall (a term we use at Sherborne to mean homework). The accuracy of Gee Pee (not sure that is the best name for my new friend?) was great. But that’s only a small sample so we can’t yet be sure about just how consistently accurate Gee Pee is yet.
Moreover, Gee Pee probably gave better feedback than I could. It understood more content than I do. It writes English better than I do too – although perhaps formulaically. I am sure you will agree with this more as you read on. Remember, I am a mathematician and computer scientist…
“It writes English better than I do…but remember I’m a mathematician and computer scientist.”
By asking it in the second person, I can now just copy and paste the feedback and give it to the student. That has saved approximately ten minutes per student and with taking a sample of a few boys’ answers, I can be confident that Gee Pee is marking sensibly.
So, I can now go and spend more time preparing a marvellous lesson for my Lower Sixth class on the writing of code to simulate a Graph Data Structure. Hang on…. Can Gee Pee help me there as well? Ask any teacher and they would suggest that one sure way of increasing student results would be if they simply had more time to prepare great lessons. There’s certainly some mileage for doing that with Gee Pee here.
“GPT understands WHY aspects of the student answer should NOT gain marks.”
So, I asked Gee Pee to write some code and I was blown away by the accuracy and consistency of it. It reads beautifully and was accurate and relevant – it responded brilliantly to further instructions to adapt the codes to my requests.
For teaching code Gee Pee could also be enormously helpful. For example, students who do not understand code can ask it questions. Gee Pee can understand and explain an important concept to the student – for example algorithmic efficiency. In fact, Gee Pee knows how to optimise the code itself and be self-critical.
Indeed, this ability to criticise its own answers is key to how AI could become a tool in detecting answers that have been written by AI – something that may become important to teachers in the not-so-distant future.
In conclusion then, I hope you can see there are immense opportunities for this technology both for students and teachers. The threats are that students can circumnavigate and trick un-educated teachers (uneducated in the capabilities of AI that is) into getting good grades but not doing any work.
“I asked it to write some code and I was blown away by the accuracy and consistency of it.”
The challenge for school, including ours, is now to understand how this will affect our future in the profession. Will we need to shift the emphasis of our homework from being factual questions, text based, or do we concentrate more on student’s verbal ability to explain concepts? But how is that scalable? I am sure our colleagues in the Modern Languages Department here will agree that marking oral vivas is extremely time consuming.
But we need to identify a strategy as a whole school and indeed whole education system as to how we utilize these new tools. Until then, I am about to start reading the book How Artificial Intelligence Works and Why It’s Making the World a Weirder Place – You look like a thing, and I love you by Janelle Shane. The future is here, and Arnie has just applied to be a teacher.