Can inclusive R&D disrupt and reshape the way we build, test, and scale innovations in education?
Each year, the Advanced Education Research and Development Fund (AERDF) invites the most forward-thinking minds to the AERDF AdvancED Fellowship to develop bold ideas to transform teaching and learning. Harnessing the power of research and development (R&D), AERDF unlocks scientific advancements and delivers research-backed solutions to pressing teaching and learning challenges, inclusive of educators, caregivers, and learners.
In 2024, AERDF invited eight new fellows to explore, design, and pitch their unique concept over 12 weeks. Throughout the fellowship, in partnership with our team at Intentional Futures (iF), fellows explored the ‘bleeding edge’ of AI for their developing ideas. As they developed their visionary concepts, our team of iF coaches, including an expert technologist, an experienced educator, and a seasoned strategist, brought together resources specific to the fellows’ research. Specific to each unique concept, we brought forward the latest in AI technology, developments in the field, and leaders to watch. What our team found in their ideas as we explored the intersection of education, emerging technology, and innovation, is too good not to share.
Each of the following examples reflects some of the fellows’ bold ideas, and recent developments and insights informing the evolving intersection of education and AI. Learn more about each of the fellows and their emerging ideas here.
Advancing Inclusive Multilingual Learning
Multilingual education has profound benefits, yet is generally missing or underutilized in traditional educational ecosystems. AI has the potential to transform (and learn from) multi-language learning by creating personalized, culturally relevant content and improving communication through real-time translation and speech recognition tools.
Examples & Implications
AI for content creation and personalization: Existing products use AI to create language learning content, such as using GenAI for video translation and even adjusting mouth movements to re-sync dubs. The incredibly popular app Duolingo uses AI in its Roleplay feature, allowing users to practice different languages with animated characters that adapt to the learner’s proficiency levels.
Inclusive training data: Some organizations and products are trying to confront bias in AI training data. In India, Karya pays rural individuals to record their native language, sells those datasets for training purposes, and then returns profits to the rural communities. StoryWeaver takes a similar approach. These companies offer a model for a development process that materially benefits communities instead of exploiting and extracting their cultural and linguistic assets.
Automatic Speech Recognition (ASR): Although currently limited by its ability to detect and process many languages at once or accents, some uses of ASR are already improving classroom interactions. TeachFX, for example, gives teachers real-time instruction feedback; Amira is a virtual reading assistant that listens to students read aloud and offers feedback and support, and Notta.ai is a real-time bilingual transcription and translation notetaker.
AI, Neurodiversity, and the Challenge of Measurement
How can AI help—and learn from—neurodiverse learners?
Neurodivergent students—those with ADHD, autism, dyslexia, and other cognitive differences—can face unique learning challenges that traditional education systems often fail to address. AI-powered tools and assistive technologies have the potential to create more personalized, supportive learning experiences that accommodate diverse needs while also advancing broader high-impact educational practices.
Examples & Implications
Sensory data collection: Using touch, speech, gestures, or movement, these technologies capture trends to personalize lesson planning, help teachers identify and intervene during learning, and boost student interest. Behavior Imaging and ClassProxima both capture visual behavior data, with the former diagnosing and making recommendations for specific students and the latter analyzing and measuring the quality of care and education in a classroom. Another example, LENA, is a small wearable device that records auditory children’s audio environments to analyze interactions and language development.
Task support: Goblin.tools is a kit of simple task support models built on OpenAI to help with executive functioning, ranging from estimating task completion time to analyzing content for tone.
Smart Manipulatives: These physical objects embedded with technology enhance real-time adaptations, like blocks or puzzles. Mytaptrack is a handheld button teachers use to capture student behavioral data, helping to project changes and outcomes that can enhance individualized support. LuxAI’s QTrobot is a human-like social robot that functions as an education and therapy assistant, specially designed for autistic students.
Learning from Non-Traditional Learners: There’s a burgeoning movement to rethink societal approaches to learning disabilities, fueling specific questions about what learning disabilities are. Lama Nachman developed the software and sensing system Stephen Hawking used to communicate. We see an opportunity to use technology to support disabled people while also using that technology to study data characterizing disabilities to improve society as a whole.
Creative data analysis for adaptive learning plans
Can AI make sense of disparate educational data?
Education generates vast amounts of data, but much of it remains fragmented and difficult to analyze in meaningful ways. AI-driven knowledge graphs and longitudinal data integration have the potential to structure this information, enabling more personalized learning experiences, improving curriculum planning, and creating more accurate AI-powered educational tools.
Examples & Implications
Knowledge graphs: Similar to a mind map, knowledge graphs are a method of defining the relationships between concepts in a body of knowledge. Knowledge graphs have gained attention recently for their ability to structure complex information in a format that is easy for LLMs to understand. In education, these could be used for improved curriculum planning, personalization, and more accurate AI assistants.
Longitudinal data and integration: In a way, education systems already track longitudinal data, with GPA often measuring progress over time. But this type of data–collected repeatedly over time for the same subjects–can otherwise be rare due to the fragmented learning landscape. Some groups are trying to integrate disjointed data to help train AI systems: for example, The Coleridge Initiative Administrative Data Research Facility (ADRF) shares a cloud-based computing platform to host confidential micro-data sets that connect different reservoirs across education and work, enabling longitudinal views and evidence-based decisions, or School Passport, a data exchange between various EdTech products that enables longitudinal views while protecting confidentiality.
Solving AI Limitations and Inequities
Are there large-scale solutions to combating bias in AI?
AI has the potential to improve education by offering personalized learning experiences, but it also risks perpetuating systemic biases if trained on incomplete or historically skewed data. To ensure AI-driven tools are equitable, intentional efforts are needed to diversify datasets, refine algorithms, and create models that better serve all learners—especially those historically excluded from educational opportunities.
Examples & Implications
Adopting equitable AI: Complete College America launched the Council on Equitable AI to urge policymakers, funders, and technologists to offer AI to historically excluded institutions in order to start training big models on new data, rather than perpetuating old bias.
Investing in holistic training models: FutureSumAI’s Latimer tool draws from both public domain and exclusive partner data–like HBCU archives and Black-owned newspapers–to reduce bias and generate content that more accurately reflects the histories and cultures of Black and brown communities.
Limitation workarounds: Although AI offers tailored solutions to novel challenges and needs, it can be just as inaccurate and inconsistent as any technology, especially as it learns. Retrieval Augmented Generation (RAG) is a process to provide additional relevant and contextual data to an AI model, improving accuracy of outputs. But RAG can still struggle, especially with “noisy” data. Knowledge graphs can help to structure data to analyze bias and quality, but they’re more difficult to build than systems based on vectorized, unstructured data. Merlyn “fine tunes” smaller models for classrooms, boosting speed and accuracy but this is only applicable for smaller, specific tasks.
Beyond the Transcript: How AI Can Recognize and Validate Learning Everywhere
How can AI help identify readiness metrics beyond the norm?
Traditional education systems rely on formal credentials—such as diplomas, GPAs, and certifications—to validate learning. However, many valuable learning experiences, such as gig work, volunteering, informal education, and international coursework, are difficult to quantify and integrate into mainstream education systems. AI-powered tools offer new ways to assess, compare, and validate non-traditional learning, making education more inclusive and responsive to diverse learner pathways.
Examples & Implications
Skill and credit transfer: Several AI-based tools can factor in new data and analyze analog experiences outside formal education, helping people like immigrants and refugees to rejoin education systems at the appropriate level. For example, Arizona State University’s Credit Mobility Course Triangulator can compare course equivalency across education institutions, allowing simpler transfer of credit. The Mastery Transcript Consortium (MTC) certifies competencies across its educational member network, standardizing and enabling accurate comparison of learning records. You can view our work with MTC here.
Alternative curricula: Stanford d.school’s Wayfinder created Social Emotional Learning (SEL) curricula that help educators monitor student wellness and growth, recommending interventions timed by data analysis. The incredibly popular video game Fortnite released an education module teaching about climate change and UN SDGs through specific game levels.
Informal learning: Education doesn’t stop outside the classroom; students engage in voluntary research on their own interests, experiential learning in the real world, and social systems in collaborative relationships. These can be lifelong, changing with personal interests instead of external rewards. Boston’s Museum of Science debuted an immersive Arctic exploration exhibit where technology enhances sights, sounds, and even powers a touchable ice wall. Roblox–an online gaming platform for user-generated content–offers educational programs in coding, digital citizenship, and design. MathTalk has immersive AR experiences to apply complex concepts in everyday social settings.