Responsible AI in education: where to draw the line?

There are no limits to the walks of life AI can disrupt. Education, the system that has inspired humanity to create something as Promethean as artificial intelligence, is now being turned on its head.

Today’s education is not like yesterday’s. If you’re from a generation that has completed school, or university, then education is shapeshifting in ways that you wouldn’t believe unless you’ve spoken to an educator. Almost every touchpoint of the learning experience is being transformed. Tomorrow’s classrooms will be barely recognizable.

In such a sensitive system, these changes can seem grotesque. But there are two main camps when considering the influence of AI in education: rejectionists, who see AI as an existential assault by oligarchical tech companies – and reluctant supporters, who see AI as having great potential for enhancing the learning experience. There’s not much overlap between the two.

We’re going to look at both sides of the argument and see which one, for now, has the better of the debate. First, some of the key pros and cons.

Arguments for AI in education:

• Tailored learning: Teachers are encouraged to offer feedback that’s tailored uniquely to the student but with limited resources that’s tricky to achieve. AI could improve personalization by analyzing student’s strengths and weaknesses, then adapting material for different levels and learning styles
• Reclaiming teacher time: By automating heavy bureaucratic burdens – like grading multiple-choice tests, drafting rubrics, data analysis, or organizing lesson plans – teachers are liberated to focus on the actual teaching bit. Overwhelming paperwork is one of the main challenges of the profession, globally, so that’s no small fix.
• 24/7 accessibility: AI tools offer students round-the-clock access to help students – answering questions and explaining concepts when teachers or parents aren’t available.
• Enhanced inclusion: Schools are becoming more multicultural and diverse than ever. AI could help by breaking down learning barriers. Real-time translation, speech-to-text tools, and adaptable reading levels make standard curriculum far more accessible for non-native speakers and students with learning disabilities.
• Efficiency: Some argue that tasks like rote textbook learning, essay writing, and homework do little to elevate the learning experience anyway. Think back to school: did they really help you? By making some of these traditional tasks less oppressive, AI might help – not hinder.
• Data-driven intervention: AI systems can track micro-patterns in student engagement and performance, alerting educators when a student is quietly falling behind before it reflects on a major exam and creates a major life blip.
• Literacy: Like it or not – and even if the AI ‘bubble’ bursts – AI is going to transform the future. By not teaching students about AI and how to use generative AI tools, students risk an old-fashioned education that’s out of step with how the world works. Boycotting AI would be like overlooking grammar or handwriting.

Arguments against AI in education:

• Loss of connection: AI cannot replicate empathy, emotional support, or a teacher’s ability to read the room. Over-reliance on screens and chatbots can lead to student isolation and a dehumanized learning experience that overlooks soft skills.
• Brain drain: If students use generative AI to bypass the struggle of learning (e.g., writing essays or solving math problems for them), they risk stunting their own problem-solving, analytical, creative and critical thinking. Not to mention the loss of ability to persevere through hard work if AI gifts outputs immediately.
• Data privacy and surveillance: Effective AI requires massive amounts of data. Tracking academic records, keystrokes, and behavioral patterns raises ethical questions about student privacy, consent, and the risk of data breaches.
• Algorithmic bias: AI is only as objective as the data it’s trained on. Flawed algorithms can reinforce societal stereotypes, unfairly flag the writing of non-native speakers as “AI-generated,” or favor certain demographics over others.
• Widening the divide: Premium AI tools and the hardware required to run them are expensive. The risk is that wealthy, well-funded schools surge ahead using the fanciest tech while under-resourced schools are left behind, exacerbating inequality.
• Illusion of mentorship: Chatbots may mimic the tone of supportive mentors but at heart their goal is to serve stakeholders. There is a risk that students, especially at a formative age, build a dependency on tools that prioritize keeping them on the platform – introducing new tasks and subtle ads, in obsequious camouflage, rather than prioritizing intellectual development.
• Environmental impact: generative AI has an environmental impact that gets heftier the more people use it. If classrooms and teachers across the globe introduced AI as standard, the impact would be huge. When many educational establishments are introducing mandatory teaching about climate change, that’s a jarring contradiction.

Generative AI vs. Narrow AI:

The AI debate conflates distinct types of AI that are not equal. Narrow AI which accomplishes predetermined tasks and is long established, tends to get lumped together with generative AI, which is the branch running wild. Generative AI is designed to create new, seemingly original content. If a student were to use AI to write an essay then that would be a generative AI task. Conversely, a tool to help a teacher grade paper assignments – and restricted to that task or closely related ones – would be narrow AI.

It seems many of the disadvantages are generative AI-related and the advantages narrow AI-related. So when considering the debate, we shouldn’t collide diverse types of AI together to create an amorphous blob of ‘AI’. This is unhelpful in education and also when considering other issues, like AI’s environmental impact.

What does the research say?

So far, the studies on AI’s effect on our brains have been conspicuously negative – take the recent research from MIT revealing that AI may erode critical thinking skills, which was virally cited on social media. Some studies suggest cognitive atrophy. But on the other hand, there’s also evidence that AI can enhance learning, with a meta analysis published in Nature suggesting the effects of generative AI on learning are largely positive. It’s still too early for a conclusive body of evidence,

So, where should we draw the line?

While the lines are very blurry and still being drawn we feel there are clear boundaries, even at this nascent stage of the transformation.

Fundamentally, AI needs to be distilled into its distinct types before considering the impact. Then you’ll see, while ‘narrow’ AI and specific tools may offer advantages for teachers – unrestricted ‘generative’ AI access for students is an area where there will have to be stricter controls – while bearing in mind not all generative AI destroys learning or cognition.

Better regulation, detection and teachers getting to know how to incorporate generative AI in teaching, will help. As will understanding what the real cognitive effects are. All of which takes time. We’re still at a very early stage of the curve.

Looking ahead, finding solutions might draw on how schools have handled mobile phones – now widely viewed as a major distraction and increasingly restricted, such as the bans implemented in Akepa’s region of Catalonia. Controlling AI won’t be as simple but it will have to be tamed by finding better regulatory balance. That may not be easy when dealing with an entity that’s so smart, getting smarter, and designed to deliver commercially for investors rather than help students become the best they can be.