Small Language Models in Education: A new approach for Learning and Democracy?

AI in education has long been dominated by massive, resource-hungry language models—the kind that require data centres to function and corporate budgets to access. But this may be about to change. Small Language Models (SLMs) are emerging that can be focused on educational technology, offering AI that’s affordable, adaptable, and most importantly accountable to the communities using it.
Education has long been a paradox - a system meant to equalize opportunity yet frequently full of inequities. Traditional AI tools, built for scale, often amplify these divides. A high school in Oslo can tap into GPT-4 for essay feedback, while a rural school in Ghana struggles with a weak internet connection just to load a webpage. SLMs change this equation. Trained on focused datasets supporting domain knowledge and with pedagogical objectives and capable of running on a teacher’s laptop or a student’s smartphone, they can equalise access to technology and education.
One example is language learning. While Duolingo relies on cloud-based AI, an SLM could power a similar app entirely offline, customized to a region’s dialects and idioms. Or for special education a compact model fine-tuned for dyslexia could provide real-time reading support without sending sensitive student data to a server halfway across the world.
The Democratic Potential of Smaller AI
The promise of SLMs is not their just their technical specification but their potential both to move from a technocentric approach t0 a pedagogical basis and to distribute agency and power. Large AI models centralize power; small ones distribute it. In education, this means:
- Local Control Over Tools
When a teacher in Mexico City can tweak an SLM to align with her curriculum—without begging a tech company for API access—education technology stops being a one-way street. Schools regain agency. - Privacy by Default
The era of "trust us with your students’ data" is ending. SLMs that process essays or math problems directly on a device, with no cloud middleman, offer a path to compliance with laws like GDPR and the new European AI Act—not as an afterthought, but by design. - Bridging the Resource Gap
In India, the nonprofit Project Bina built a Hindi-language tutor that runs on recycled tablets. In rural Canada, an open-source SLM helps Indigenous communities preserve languages without relying on Silicon Valley.
None of this is simple. Small models require thoughtful training data to avoid inheriting biases. Teachers need support to integrate them meaningfully—AI shouldn’t replace pedagogy, just as calculators didn’t replace math teachers. And we must resist the temptation to view SLMs as a panacea; they’re tools, not saviours.
But the direction is clear. For decades, education technology has chased "disruption" while often reinforcing existing power structures. SLMs offer something quieter but more profound: the chance to rebuild edtech from the classroom outward—not as a product to consume, but as a resource to shape. The rise of SLMs in education isn’t just a technical trend. It’s a test of what we value. Do we want AI that serves the few or empowers the many? Models that lock knowledge behind paywalls or ones that communities can adapt and own? The answers will determine whether AI becomes just another unequal institution—or finally, a true equalizer.
About the Image
In this image, I wanted to explore how employers are beginning to adopt AI for specific, often low-level and repetitive tasks. I represented this through the figure of a businessman literally carrying AI into the workplace through a back door. This door sits at ground level which is a reference to the traditional pathway where young people entered the workforce at the bottom and worked their way up. In contrast, a group of young people are shown gazing up at a front entrance high above their heads, symbolising how the entry point into the workforce is becoming less accessible. By outsourcing foundational tasks to AI, employers may close off opportunities for young people to build digital skills and gain early career experience. The goalposts being moved, as it were, as a result of AI adoption. This image was created using collage tools on Canva.com. This image was selected as a winner in the Digital Dialogues Art Competition, which was run in partnership with the ESRC Centre for Digital Futures at Work Research Centre (Digit) and supported by the UKRI ESRC.