Brilla AI: AI Contestant for the National Science and Maths Quiz
Project Overview
The document focuses on the innovative use of generative AI in education, specifically highlighting Brilla AI, which was developed to participate in Ghana's National Science and Maths Quiz (NSMQ). This initiative addresses the scarcity of qualified teachers in Sub-Saharan Africa by leveraging machine learning techniques such as speech-to-text, question extraction, question answering, and text-to-speech for real-time competition. In its inaugural live competition in 2023, Brilla AI achieved a 25% accuracy rate, demonstrating the potential of AI to enhance science education. The project underscores the transformative role of AI in providing personalized tutoring and expanding access to educational resources, thereby democratizing learning opportunities across the region. Overall, the findings suggest that generative AI can significantly contribute to improving educational outcomes and addressing teacher shortages in underserved areas.
Key Applications
Brilla AI - AI contestant for the National Science and Maths Quiz
Context: Secondary education, specifically for senior secondary school students in Ghana participating in the NSMQ.
Implementation: Brilla AI uses a web app to livestream the quiz and employs four machine learning systems for real-time interaction.
Outcomes: Unofficially placed second in its debut, answered one riddle correctly, and achieved a 25% exact match accuracy.
Challenges: Faced issues with speech recognition accuracy, latency in processing, and the need for improved training data.
Implementation Barriers
Technical Barrier
Challenges with speech-to-text accuracy, especially for Ghanaian accents, leading to misinterpretation and failure to detect riddle clues.
Proposed Solutions: Future plans include fine-tuning the speech-to-text model with Ghanaian-accented data to improve performance.
Implementation Barrier
Latency issues caused by the sequential processing of API calls, affecting real-time performance during live quizzes.
Proposed Solutions: Exploration of parallelizing operations to reduce latency and improve efficiency.
Data Barrier
Difficulty in curating accurate NSMQ data due to scattered information and inconsistent online resources.
Proposed Solutions: Plans to enhance data collection methods and automate data extraction processes.
Project Team
George Boateng
Researcher
Jonathan Abrefah Mensah
Researcher
Kevin Takyi Yeboah
Researcher
William Edor
Researcher
Andrew Kojo Mensah-Onumah
Researcher
Naafi Dasana Ibrahim
Researcher
Nana Sam Yeboah
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: George Boateng, Jonathan Abrefah Mensah, Kevin Takyi Yeboah, William Edor, Andrew Kojo Mensah-Onumah, Naafi Dasana Ibrahim, Nana Sam Yeboah
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai