Leveling the Playing Field: Google’s LearnLM and the Quest for Equitable Education
A major challenge facing the world is equitable and universal access to quality education.
Google is the World’s Greatest Teacher and YouTube is the World’s Largest Classroom
Where do people start when they want to learn something?
Google. People come to Google with curiosity. Google transforms that curiosity into learning by providing search results and connecting people with amazing YouTube creators who have the answers.
Arguably, Google Search is the world’s greatest teacher and YouTube is the world’s largest classroom.
When you’re learning on YouTube, do you watch multiple videos before you understand the concept? Different videos have different instructors with different teaching styles. Eventually, you find that one video with that one instructor who is able to breakthrough to your mind and you can finally say, “ah, that’s how that works!”.
How can we repeat this feeling more broadly?
Google Combines AI Technology with Learning Science
To answer this question, Google’s current approach is to combine its AI technology with generally accepted learning science principles.
- Inspire Active Learning: AI Tutors can assist with this.
- Manage Cognitive Load: Present information in multiple modalities, which Gemini supports.
- Adapt to the Learner: Dynamically adjust the goals and needs of the student and ground the results in relevant materials.
- Stimulate Curiosity: Provide motivation through the learning journey. The YouTube Recommendation AI is a simple example of stimulating curiosity.
- Deepen Metacognition: Plan, monitor, and help the learner reflect on progress.
Google’s History Using AI in Education
Google is not an education company.
It’s strength is organizing the world’s information, connecting people with teachers who have the answers, and building great AI technology. Google supports education partners and the earliest case study I found was from 2018 when Google worked with Walden University to create an AI tutor called Julian using its BERT model, which came out in 2018 shortly after Google invented the Transformer in 2017.
Another partner Google supports is Arizona State University (ASU).
Arizona State University’s “Study Hall” Democratizes Education
Study Hall is a collaboration between ASU and YouTube that started in 2022 allowing students to test drive courses and obtain college credits before applying to ASU.
It costs $25 to register and to access content on YouTube. The curriculum is the same curriculum taught by ASU faculty on campus. Once students master a subject, students can elect to pay $400 to receive college credit. The cost works out to about 60% savings of the average course cost at ASU for in-state students and about 90% lower than the course cost of a private university.
Reducing the cost of an university education is an early attempt to democratize education.
Generative AI in Education
Building upon these partnerships, Google is researching how to leverage recent advances in generative AI to tackle the problem of equitable education.
One area of research is LearnLM. LearnLM is a new family of AI models based on Gemini fine-tuned for learning. Google created an AI tutor using LearnLM and embedded it into a Chrome extension chatbot called HallMate where ASU Study Hall students could trial it in an “Introduction to Programming” course. HallMate was grounded in the course materials and was able to guide students deeper in the content and recommend other course related content.
Many students found the Learn-LM tutor effective.
Through our participatory research, we have learned that AI tutors can be beneficial to learners by promoting active learning and providing personalized help when explaining concepts or working through problems.
While the results look positive, the Google Researchers acknowledge we are still in the early days of leveraging AI in education and suggested:
- The dream of a personal tutor for every student and teaching assistant for every teacher has not materialized yet.
- AI cannot replace teachers yet because we cannot embed pedagogical intuition into prompts.
- We need better practices to evaluate the efficacy of AI tutors.
The paper concludes with a call to action for EdTech and the science community.
We hope that the AI, EdTech, and learning science communities see this report as an invitation to join forces and work together to continue developing and iterating on a set of pedagogical benchmarks that we can all use in our daily research and product development.
The quest for equitable education is just beginning but with each step, we unlock brighter futures for all.
Partner with Google to Improve LearnLM
If you’re interested in partnering with Google to help define educational benchmarks, improve academic capabilities, and explore the possibilities when it comes to applying advances in generative AI to teaching and learning, fill out the LearnLM Interest Form.
Summary
- Google has a long history in education with Search and YouTube.
- Google supports education partners, such as Walden and Arizona State University, and wants to work with more partners.
- Google is researching how to leverage Generative AI to solve the problem of equitable education and concludes we are still in early days.
Resources
- Towards Responsible Development of Generative AI for Education: An Evaluation-Driven Approach
- How generative AI expands curiosity and understanding with LearnLM
- Grow with Google launches a new generative AI course for educators in collaboration with MIT RAISE
- Google Cloud at ASU+GSV 2023 — solutions to scale and personalize learning
- Walden University Creates AI-powered Tutor Built with Google Cloud
- ASU + GSV Summit
