Central to our mission, SCORE explores and elevates innovative ideas as they advance from the research and pilot stages into actual practice. The next generation of innovation is calling us to pay special attention to artificial intelligence (AI) in education, with this tool being quickly adopted by educators and integrated into schools. Given its novelty, the research on AI’s use and impact in education is only just emerging, but there is already a wealth of evidence for AI’s transformative potential as well as compelling leads for where it can do the most good for Tennessee students.
As SCORE continues to build our understanding of AI in education by keeping up with research and articles, we’ve been centering our learning around three key functions:
- Student learning: How does AI use relate directly to student learning outcomes? A focus on the impact of AI on instructional quality, student mastery, and achievement.
- Efficiency: How can AI streamline nonteaching educational processes for teachers and administrators? A look at the impact of AI on efficiency, creativity, and consistency.
- Life and career: How might we better prepare students for a workforce disrupted by AI? A focus on AI literacy, career preparedness, and considerations for AI use in work.
After spending time with many sources of evidence that reflect these three functions, we’ve collected our learnings from some initial research we found intriguing:
Student Learning
Education technology companies have shared early predictions of AI’s impact, though the research base for understanding student outcomes with AI remains sparse. Among the earliest research on AI in education, researchers from Stanford University’s SCALE Initiative have offered valuable insights through their Generative AI for Education Hub.
One paper of note is Tutor CoPilot: A Human-AI Approach for Scaling Real-Time Expertise, which highlights a randomized controlled trial where 874 tutors were randomly assigned to use a “Tutor CoPilot” AI coach to assist their pedagogy. In assessments after their lessons, students of tutors using AI were found to be 4 percentage points more likely to demonstrate mastery. This significant positive effect is promising early evidence for the potential of AI to directly improve outcomes and raises the idea to refine the approach and scale it to benefit more students.
This study fits into a wider body of research from Stanford that has shown AI can help teachers and tutors implement effective practices such as amplifying student ideas and asking focusing questions.
Efficiency
Automated grading makes up the bulk of academic research about efficiency, where generative AI can now perform at a roughly human level. Notably, Texas used AI to grade their state assessment last year, a move that was estimated to save $15 million by cutting the number of human scorers required. The state’s auto grader featured an important guardrail that sent exams back to human scorers when AI identified situations where its grading might be inaccurate and require human involvement.
Outside of grading, there aren’t many quantitative measures of efficiency improvements with AI, but current research is focusing on the tool’s uptake by educators, including teachers and administrators. While definitive numbers vary depending on the polling methodology, they seem to uniformly point toward increased use of AI. Carnegie Learning cited an increase of nearly double the educators who claimed they use AI “often” or “always,” from 22% in 2023-24 to 43% in 2024-25. A similar shift — increasing from 20% to 40% — was also reported in schools that have adopted AI policies. A LinkedIn poll conducted by Education Week found more teachers are using AI — an increase from 40% to 60%.
Increased uptake in AI usage is a strong sign the innovation is likely saving teachers time, but we do not know how increased AI use may affect students to the same extent. We’ll keep an eye out for more research in this space.
Life and Career
A bright spot in the research so far is in the work done to forecast the jobs and industries most likely to be disrupted by AI. The Brookings Institution and the Burning Glass Institute both looked at which occupations involve tasks that AI is likely able to augment or automate and classified the jobs by their exposure to AI. These research efforts both show that high-demand, high-wage jobs will likely be the most impacted by AI.
While these examples are rigorous in their approach, they are also quite theoretical and perhaps difficult to apply practically. On the other hand, Anthropic, the company behind the Claude AI chatbot, piloted a different approach with the Anthropic Economic Index. Anthropic researchers analyzed over 4 million human interactions with Claude to identify which workplace tasks are performed most frequently using AI. The results are summarized nicely in the following graph:
Anthropic found that education is the third most likely profession to use AI, accounting for 9.3% of the AI conversations, following computer occupations with 37% and communication roles with 10.3%. Within education, Anthropic found that instructional support made up the top two most prevalent AI uses. The study also confirms that people are using AI to augment their work, not to automate it, meaning these jobs are not being made obsolete, but more productive. This is promising support for promoting AI usage in classrooms.
We expect AI research to continue gaining traction, and we will continue monitoring findings that relate to education. We look forward to further exploring how research can support practice, policymaking, advocacy, and a broader understanding of AI’s novel influence on teaching, learning, school operations, and workforce development.