Motivation: Rapidly Evolving, AI-Driven Job Landscape
Oklahoma faces significant challenges in educational outcomes, consistently ranking near the bottom nationwide in school quality, student success, math scores, science and engineering fields, and overall K-12 education, creating an urgent opportunity to embrace AI literacy as a catalyst for accelerated learning and practical implementation, a priority underscored by the Governor's Task Force on Emerging Technologies and OU President Harroz's Lead On, University: The Next Phase.
Traditional learning methods, such as textbooks, lectures, worksheets, and static web searches, often lack interactivity, personalization, and engagement, making them increasingly inadequate for diverse student needs. Generative and agentic AI, however, can adapt content to individual learning styles, dynamically summarize information, and leverage co-intelligence to bridge disconnected topics; capabilities traditional methods struggle to match.
Integrating AI-assisted learning tools into under-resourced Oklahoma schools and colleges can transform passive classrooms into dynamic, personalized, and interactive environments, bridging disparities and fostering equitable success. AI-powered tutoring and analytics improve test scores, deepening comprehension and engagement even outside the classroom. Major districts in Massachusetts, California, and Florida are scaling similar AI initiatives and literacy programs, reflecting growing nationwide enthusiasm. However, true AI literacy demands ongoing exposure to multiple models and tools, ensuring learners develop the critical thinking needed to evaluate diverse AI applications rather than relying on a single edTech product.
AI is also revolutionizing Oklahoma's key industries, such as energy, where it optimizes drilling and maintenance, and agriculture, where it enhances crop monitoring and precision farming, aligning with national trends where automation is expected to reshape a significant portion of work hours by 2030. To thrive in this evolving job market, students must learn to harness AI effectively, as emphasized by projections of significant disruptions in workers' core skills in the coming years. Such AI literacy can be effectively achieved only through direct, hands-on experience with AI-driven learning and experimentation.
Current State of AI: Pay-to-Play
Private companies have become major providers of advanced AI tools and API services, offering ready-to-use services for tasks such as text generation, data analysis, reasoning, programming, and computer vision. Platforms like OpenAI, DeepSeek, Claude, and Gemini deliver scalable, cloud-based solutions with extensive developer resources and ongoing research breakthroughs. Students are increasingly leveraging these AI-driven tools for independent learning, marking a significant upgrade from traditional reliance on Google or Wikipedia by providing interactive, personalized, and generative experiences that adapt to individual needs.
In parallel, open-source AI models also continue to thrive, encouraging collaboration and customization. Examples include Meta’s LLaMA, DeepSeek-R1, StabilityAI's Stable Diffusion, and Microsoft's Phi 4, all available for local deployment. Numerous open-source software platforms, like OpenWebUI, ComfyUI, Hugging Face Transformers, and Text Generation Web UI, provide interfaces for deploying and refining these models. Additional tools like LangChain, Haystack, and Stable Horde enable specialized AI workflows. Together, these models and software empower a diverse range of AI-driven applications and services.
Currently, most educational AI applications rely on commercial tools, funneling student queries and data into externally controlled servers with minimal regulation of privacy or security. Users have limited visibility into how their information is stored or used, raising concerns about privacy and other safeguards for minors. Furthermore, major proprietary models are not open source, making it difficult to evaluate them for bias, censorship, or harmful outputs. Inquiries about sensitive topics may be censored or inconsistently handled due to opaque decision-making, undermining trust and potentially harming vulnerable students.
Open-source models, in contrast, can be rigorously vetted, audited, and locally deployed in the state of Oklahoma to maintain data sovereignty and better align with ethical and moral guidelines. However, many under-resourced schools and colleges in Oklahoma lack both the technical infrastructure and specialized expertise required to effectively host and manage these systems. Consequently, these schools and colleges may forgo AI literacy entirely, forcing students to rely individually on unregulated commercial services. This situation perpetuates educational inequities by allowing students from affluent backgrounds to pay for commercial AI services, leaving their under-resourced peers at a disadvantage. Such restricted access to AI solutions hinders efforts to meaningfully address existing learning gaps.
Our Mission: Making State-of-the-Art AI Accessible to Every Student in Oklahoma
SoonerAI will be designed and implemented as a trustworthy, open-source AI platform at the University of Oklahoma, offering accessible, adaptive AI resources for both classroom and independent student use. Its primary goal is to empower students and educators with integrated lesson-plan support while also providing a free, personalized alternative to commercial tools like DeepSeek and ChatGPT, with both chat and API support.
Open-source models such as Meta’s LLaMA 4 (70B), Mistral 9B, Falcon 200B, and Google’s Gemma 2.0 will be vetted and tested for deployment on OU’s secure servers. These models enable inference, customizable fine-tuning, and retrieval-augmented generation (RAG). By linking the AI to curated knowledge bases—from digital textbooks to lecture archives—SoonerAI ensures accurate, curriculum-aligned responses. Agentic features will further enrich learning by guiding students through intricate tasks, adapting to individual needs and interests.
- Transparent, Privacy-Centric Architecture: By employing open-source models, SoonerAI ensures transparency, privacy compliance, and data sovereignty. Translating prior works on mitigating hallucinations, membership inference attacks, poisoning attacks, and adversarially robust AI/ML, SoonerAI creates a trustworthy, privacy-focused educational ecosystem.
- Retrieval-Augmented Generation for Accuracy: Linked to curated academic content, SoonerAI employs retrieval-augmented generation to deliver curriculum-aligned answers. This domain specificity fosters deeper comprehension and surpasses general-purpose models by contextually integrating real-time, relevant learning materials.
- Agentic Capabilities for Deeper Understanding: SoonerAI’s agentic features guide students through complex tasks. This personalized, adaptive scaffolding bridges topics, promotes curiosity, and catalyzes problem-solving.
- Pilot Testing and Professional Development: The pilot program will involve OU undergraduates and faculty who will offer iterative feedback through direct classroom integration and independent use. Comprehensive professional development for instructors ensures equitable adoption, meaningful impact, and continuous refinement.
SoonerAI will be freely accessible outside school hours, helping students continue their educational journey independently. As SoonerAI matures, we will expand its reach to high schools and colleges statewide with additional infrastructure funding from local and federal agencies. Workshops, training modules, and professional development opportunities will equip educators and students alike with the skills to effectively utilize SoonerAI. Ultimately, this platform seeks to narrow educational gaps and foster innovation, helping Oklahoma’s learners build the confidence and expertise needed in a rapidly evolving world.