Canada CIFAR AI Chair Carrie Demmans Epp on the future of AI in classrooms 

Classrooms are no strangers to the evolution of technology. From blackboards and chalk to overhead projectors, SMART Boards and ThinkPads, teachers have long adopted tools to help students learn better. 

Today, Generation Alpha (Gen Alpha), born roughly between 2010 and 2024 and colloquially known as “iPad kids,” makes up nearly all K-12 students.

Many don’t know a world without smartphones, tablets and social media, and some are now stepping into AI-enhanced classrooms.

Studying the application of AI in learning environments is Canada CIFAR AI Chair at Amii, Carrie Demmans Epp. As an associate professor of Computing Science at the University of Alberta, she is the principal investigator of the EdTeKLA Research Group, which develops educational software to support learning while studying human cognitive processes.

Her research comes at a critical moment for Canadian classrooms, which are facing both growing class sizes and multi-layered educational needs.

Data from the Canadian Teachers’ Federation shows average class sizes across the country sitting between 22 and 26 students, with some kindergarten to Grade 6 teachers reporting classes exceeding 40, 50 or even 60 children. In rooms that large, it becomes challenging to provide crucial one-on-one instruction.

AI in the classroom

Demmans Epp sees this as an opportunity for AI to support educators, giving them a clearer picture of the needs of each student.

At the core of this technology are adaptive machine learning systems, a type of AI that can update its parameters and behaviour based on new, real-time data without operator intervention.

In education, this technology works well because learning naturally happens step-by-step. Most subjects build on prerequisite concepts. A student must master basic addition before moving to multiplication, just as they must learn the alphabet before reading full sentences.

Adaptive software embeds these knowledge structures directly into its design. As a student completes exercises, the system tracks their accuracy and speed to create a digital map of what they know. The software then automatically adjusts problem difficulty, offers targeted hints or introduces new topics tailored to individual progress.

“In a public K–12 classroom, once you reach a certain class size, teachers physically can’t track what every single student knows in real time,” explained Demmans Epp. “This system gives teachers actionable data on where individual students are struggling so they can intervene appropriately. The teacher still does the core instruction; the AI system simply replaces standard independent practice for as little as 20 to 60 minutes a week.”

In a two-month pilot project led by Demmans Epp in rural Alberta, researchers deployed an adaptive learning system with young students, checking their progress three times along the way.

While the formal findings have yet to be published, the preliminary results showed measurable improvement for all students, but most notably, children with the lowest cognitive abilities experienced zero summer slide, the common seasonal loss of learning between June and September. In fact, their assessment scores in September surpassed their scores from June, prompting the school division to expand the program.

Demmans Epp pointed out that similar adaptive technologies are already being widely adopted globally.

“Other countries are already deploying similar systems at scale — the U.S. uses them across thousands of schools and countries like Brazil use them to support low-resource areas,” she said.

Managing risks

Like with any new technology or teaching method, bringing AI into the classroom comes with certain risks.

“Where we have a problem is when people just offload everything to the system. That is entirely the wrong approach and not one that anyone in this field would advocate for. As long as technology is integrated mindfully into the broader educational ecosystem, it can be a helpful tool,” said Demmans Epp. 

Further, she explained that the level of risk also depends heavily on the task at hand.

While objective subjects like math carry relatively low risk, automated essay grading introduces major concerns around fairness and systemic bias.

Ongoing evaluations of large language model (LLM) essay graders reveal clear linguistic biases, Demmans Epp noted. Models tested by researchers repeatedly awarded higher scores to Canadian and British English while penalizing African-American English.

Addressing these biases is critical before any automated grading system is used to evaluate student work. Demmans Epp emphasized that parents and educators need to know that a tool has been rigorously tested for bias, particularly against children whose home language shapes their writing.

Navigating student AI use

Beyond adaptive tools, schools are also grappling with students using generative AI for schoolwork, a challenge that is particularly prominent in high school and post-secondary classrooms.

Demmans Epp views this not as a technical problem for detection software, but as a pedagogical shift.

“[Historically] we’ve had this really big focus on assessing products and advancing people based on the output they produce, when really learning is about a process. But the process is hard to capture. We need to focus on assessing and valuing that process, and shifting that mindset will help relieve the pressure to perform.”

For educators, this could mean breaking assignments into distinct phases, such as requiring brainstorming logs, rough drafts and revision notes. 

“Maybe you do some of the brainstorming with something like ChatGPT and you submit the logs,” suggested Demmans Epp. “That isn’t necessarily bad if the space allows it, because students will eventually be working in environments where that’s permissible. But it involves teaching kids how to interact productively with these tools, which means changes to our teacher education system.”

In the field of education, everyone is a student. As technology accelerates and “learns,” educators, parents and developers are learning how to adapt alongside it. 

Yet despite these changes in tools and methods, the core mission – learning – remains. 

Beyond the AI pathways and prompts, education continues to be what Nelson Mandela famously called “the most powerful weapon which you can use to change the world.”