Why human-centered AI should be embraced in education
By
James Smith
·
3 minute read
Whilst providing a demo of Mark My Words to a school late last year, the fear of AI was put to me bluntly: “Well,” one teacher said, “there goes our jobs.”
As an educator passionate about technology's role in learning, I am fascinated by what AI could achieve in the classroom. Unfortunately – yet understandably – the dialogue around AI is so often led by fear that the significance of this technology as a tool for teachers can become lost.
The truth is that AI is a long way from being able to stand up in front of 20 students and teach a class in an interactive, personalised, and reactive way – and no matter how good this technology gets, human oversight is a non-negotiable.
Ensuring humans remain central to AI decision-making is a topic high on the international agenda.
Back on November 15, 2022, Stanford hosted a conference that reinvigorated the dialogue around the need for a "human in the loop" approach to AI development. Thought leaders, policy makers, and technologists convened to consider a future where people remain at the heart of decision-making in AI systems.
The central idea of the conference was simple yet very significant: AI should augment us, not usurp us. I am yet to meet anyone who disagrees with this sentiment – and would be terrified of anyone that does.
Stanford HAI's Associate Director, Russ Altman, emphasised the pivotal role humans should play, "The default should be human control, and we should use autonomous AI systems only rarely under special circumstances... all AI systems should be designed for augmenting and assisting humans – and with human impacts at the forefront."
This philosophy of assisting not usurping has guided the development of Mark My Words.
A large part of my work with teachers, school leaders and students over the last few years has been about where artificial intelligence can and should be adopted to ease the marking marking and feedback load. Everyone that I have spoken to has agreed that incorporating AI should be about enhancing the teacher's role and empowering students to learn more effectively, not replacing the unique human elements that are essential to education.
The risks of getting the balance between human control and AI wrong are non-trivial.
The conference speakers acknowledged this, highlighting the potential consequences on the human experience and society at large if AI systems proliferate without prioritising human control.
As suggested at the Stanford HAI conference, interdisciplinary partnerships will be pivotal in building truly human-centred AI applications. This means teachers working alongside technologists to craft AI tools that reflect our pedagogical objectives and uphold ethical standards.
So, guided by this, I have prioritised the “human-in-the-loop” or “human-centred” approach when working with teachers and engineers to develop the solution.
The assessment rubric is the engine-room of the entire workflow – and so, we must keep this human-led.
The teacher must spend time writing out everything they want to see in any given assessment - a 'checklist', so to speak.
The teacher needs to outline the different skill groups, specific skills, the levels within these skills, and how students can progress within these skills.
Using the teacher's knowledge to inform the software’s knowledge-base means that any AI ‘decisions’ from there on out are, inherently, “human-centred”.
Checks and balances are key. The software facilitates teacher oversight.
1. Document allocation must be verified.
After standing at the printer for twenty minutes scanning individual essays, I have prioritised a feature allowing teachers to upload a collection of handwritten essays in one PDF – using an AI algorithm to automatically sort essays by student, splicing the PDFs where necessary. However, once the scan is complete, the teacher must confirm whether each essay has been spliced appropriately, making changes where necessary.
2. Teachers must check how handwritten essays have been transcribed.
Ethically, it is not appropriate to allow the AI agents to mark and provide feedback on handwritten essays without allowing the teacher to see how it has been interpreted. The model is very, very accurate, but never (or at least, rarely) perfect. It is essential that the teacher can see the transcription so that they can address hallucinations and understand exactly what the model is basing its assessment on.
3. All marking decisions must be reviewed by the teacher.
When the teacher clicks ‘Professor Mark’ and asks him to fill in the feedback, the software will fill the rubric in first. Although we have the ability to create models for schools that are extremely accurate, teachers will always have the ability to override the decisions of the AI. By keeping the “human in the loop”, and allowing them to correct any erroneous decisions, we can actually create a workflow where the AI becomes iteratively better. It is this symbiotic relationship that I believe should characterise the integration of AI into a teacher’s workflow.
4. Written feedback provides a framework for the teacher to change as necessary.
The software uses language models to produce very detailed feedback for students. We ensure that the feedback provided is precise, with the software able to cite the essay and offer constructive suggestions for improvement. This is a tremendous time-saver, and infinitely more effective than just starting from scratch. Having said that, it is designed to be a template – all feedback can and should be edited by the teacher.
5. All insights received by the student are validated by their teacher.
Mark My Words provides teachers – and students – with a detailed dashboard that outlines a student’s performance. These dashboards only present information that has been curated and confirmed by the teacher, even if it was in fact predominantly written by AI.
AI will transform the classroom. The education landscape will look dramatically different in fifteen years time. There is no doubting it.
However, we are currently faced with some critical decisions about how we can use and structure these resources in a way that keeps us “in the loop”. Being open and receptive to this technology is critical, but so is being cautious and thoughtful about the way we integrate it.