BU’s EVAL collaborative researches AI tools in education, seeks to ‘understand the limits’

In classrooms across the country, students are increasingly using artificial intelligence in their work. According to Microsoft’s 2025 AI in Education Report, 86% of educational institutions use generative AI in their coursework.

Boston University’s multidisciplinary Evidence-Based AI in Learning Industry Collaborative aims to evaluate AI as a tool to help K-12 students improve academic performance. 

Ola Ozernov-Palchik, a cognitive neuroscientist, research assistant professor at BU and director of the EVAL initiative, said AI tools can help identify academic risks for children with disabilities and boost students’ performance. 

“We want to make sure that we build the evidence base for AI in K-12 education and really [understand], ‘What are the limits of these technologies [and] who can they help?’” Ozernov-Palchik said.

During the past academic year, the collaborative launched its first challenge, in which companies could apply to have their student-facing AI tools tested and researched. The collaborative’s second annual challenge is currently in the second stage of its application phase. 

The winner was Amira Learning, an educational technology company that offers a suite of AI-based reading tutor products. EVAL’s research on Amira is still ongoing.

“We don’t really have evidence that [AI] can actually teach a child how to read yet. And so that is where we are right now, and that’s the research we’re trying to build,” Ozernov-Palchik said.

Ozernov-Palchik said students as young as first graders and kindergarteners have shown they understand AI agents as non-human, emotionless tools that may be good information sources.

“They have a really nice boundary between whether the AI is human or not a human entity, which I think is really interesting,” she said. 

In the current study, K-12 students are randomly assigned to work with either Amira or a human tutor. 

Mary Bready, research coordinator for EVAL, said students working with the Amira model are also paired up with an Amira “buddy” who oversees the sessions and provides technical assistance.

Bready said careful facilitation and feasible implementation are crucial when introducing an AI tool into the classroom.

“Classrooms are chaotic,” Bready said. “Even with the best of intentions, if there’s not really clear implementation guidelines and supports, it’s really easy to be like, ‘I meant to get to that this week,’ and it’s never actually implemented.”

Sammi Cruz, a junior at Tufts University and a psychoevaluator for EVAL, said her main responsibilities are administering the baseline and post-assessment tests for the Amira study.

Cruz said she outlines the research process for parents, informs them of the research’s scope and provides them with score reports for their child’s assessments. 

The assessments consist of word games that test decoding, speed, efficiency and vocabulary. 

The data, Cruz said, helps the team effectively place students in intervention methods that best fit the student’s needs.

Some students went from reading 15 words per minute to 60 words per minute in 10 weeks, she said.

However, others did not improve or worsened, possibly because of summer break or other factors, she said. Due to the study’s design, she did not know if the students whose results worsened were in the group that used the AI software. 

Cruz said parents are more receptive to the technology because they want tools that will help their children, but teachers tend to be concerned about its effects on critical thinking skills. 

“[AI is] not going away. It’s only going to start growing,” Cruz said. “Are we teaching them how to use it, or are we only giving them the correct models?”

Ozernov-Palchik said she understands the pushback against AI in education, but “we tend to think about it pretty narrowly.” 

“Ineffective approaches harm students with disabilities and students who are from underserved backgrounds,” Ozernov-Palchik said. “It is our duty to protect these students and make sure that we have evidence for these tools.”

Read more here: https://dailyfreepress.com/11/16/22/222596/bus-eval-collaborative-researches-ai-tools-in-education-seeks-to-understand-the-limits/
Copyright 2026