What role should Generative AI play in education?
For my third year, Computer Science project during university, I studied Artificial Intelligence (AI), creating a basic “Expert System” to diagnose car faults. Some of the code listings have survived – see below– but sadly not my choice of Object-Oriented language “DOMAIN”, built on an equally dead language called “Salford LISP”. No surprise really. The more observant of you will have noticed the print date at the top of the listing: March 1989.
AI isn’t new and wasn’t new even 34 years ago when I coded this rules-based decision engine to pinpoint why my rusty old Datsun wouldn’t be getting me to lectures that day.
What’s new is the recent exponential increase in processing power at our fingertips coupled with an explosion of free and cheap tools that use (or claim to use) AI to solve our educational problems. With all this power comes great responsibility, and great confusion.
In this first of a pair of blogs I address these issues by tackling the question: “What role should Generative AI play in education?”
We first need to clarify the scope of our enquiry. There are two main roles that AI can play, which I call teaching about AI and teaching with AI. This blog focuses on the former topic.
Teaching about AI
Much is talked of AI’s downsides: harmful deepfake videos, biased decision-making, the impending “singularity” when AI becomes all-powerful. I do believe we need legislation to limit the harms, but it’s not all doom and gloom.
In my book “How to Teach Computer Science” I covered a machine-learning (ML) program that improved cancer detection rates from 96% to 99.5%. And this article in the Washington Post explains how ChatGPT is helping a dyslexic tradesman communicate professionally with customers.
Like all new technologies, AI is both a blessing and a curse, and we clearly need to articulate to learners the social and ethical implications. These make up one aspect of the 4-layer SEAME model devised by Jane Waite and Paul Curzon in 2018 to classify AI concepts, discussed in this month’s Hello World magazine:
and Ethical aspects
benefits of AI versus copyright, security, privacy and bias concerns, impact on
employment and potential for misinformation.
a range of applications, explore rule-based versus data-driven approaches.
models are created, trained and tested: the different learning paradigms of
engines work: the difference between decision-trees and neural networks.
Teaching about Generative AI
While AI has been around in tools like translation software, driver-assist technology and content recommendation algorithms for years, the easy availability of generative AI for public use brings particular challenges.
All pupils should be taught some of the SE and A levels of the above model – because a basic level of AI literacy is essential for modern life. However, in computing it’s possible and perhaps vital to explore all four.
The Department for Education’s briefing paper suggests as a minimum, pupils are taught:
- Not to enter sensitive or personal information into a generative AI tool
- To be aware that the results may be inaccurate (for example, LLMs are terrible at maths and Boolean logic)
- To take great care and to check the origin and trustworthiness of all digital content, as AI can create very believable content, including text, images and even video (these could include a scam email, a fake image or a deepfake video).
- Be aware that generative AI can also be used to create harmful content, and to use only according to the terms and conditions including the minimum age (Bard’s is 18, whereas ChatGPT’s is 13 but only with parental consent).
You can read the Department for Education’s briefing here. I also recommend ChatGPT and large language models: what's the risk? from the National Cyber Security Centre. Resources exist to help you get AI into the classroom, including the Isaac Computer Science Impacts of Technology topic and the Teach Computing GCSE curriculum's Ethical impact lesson. There are additional AI lesson resources on the STEM website here.
Other resources you may find useful include Google’s Teachable Machine and IBM’s Machine Learning For Kids which are ML models aimed at the classroom, while MIT’s Moral Machine is a fun way to explore ethical decisions that AIs, such as self-driving cars, must make every day. Why not explore these with your learners, using the SEAME framework to structure lessons?
In this blog I focused on Teaching about AI. In my next blog, I will be looking at the parallel topic Teaching with AI. And one day I will rewrite my car fault expert system in Clojure – the only surviving relative of the LISP language from 1989 – and create some lessons around it for old time’s sake!
About the author
Alan Harrison is a National Specialist for Secondary Computing Leadership at the National Centre for Computing Education.