SBIR Phase I: Artificial Intelligence, Scientific Reasoning, and Formative Feedback: Structuring Success for STEM Students

This SBIR Phase I project uses artificial intelligence techniques to identify the ways that undergraduate students in scientific courses explicate problems, describe procedures, make claims, provide evidence, offer qualifications, and draw conclusions. With emphasis on forms of scientific reasoning, this new use of artificial intelligence, to identify language patterns associated with scientific reasoning, will allow students to improve their written laboratory reports before they are submitted, therefore freeing instructors to devote precious instructional time to preparing students for roles as practicing scientists. As the U.S. continues to experience rapid diversity growth, this focus on helping students through innovative uses of technology holds the potential to expand science education by cultivating student ability through autonomous writing and revision. Because artificial intelligence techniques are intended to expand capabilities, the techniques being used, available 24/4 on the web, will have the direct impact of growing our technical and scientific workforce, thus expanding the many dynamic pathways to STEM occupations. As the NSF observed in 2015 in its report Revisiting the STEM Workforce, these jobs are extensive and critical to innovation and competitiveness and are essential to the mutually reinforcing goals of individual and national prosperity and competitiveness. An investment in such a technology is thus an investment in national competitiveness, education policy, innovation, and workforce diversity. NSF SBIR support will be used to design and launch artificial intelligence techniques based on Deep Artificial Neural Network (DANN) as driven by Natural Language Processing (NLP), Latent Semantic Analysis (LSA), and the latest advances in AI algorithms. Because NLP and LSA techniques are presently used solely to identify grammatical and organizational patterns, the application of DANN is high risk in making a leap from identifying patterns of language use to capturing patterns of scientific reasoning. Trained on a proprietary corpus of 100,000 lab reports scored and annotated by instructors and students using a single rubric, the AI application will identify logic structures of scientific reasoning in student laboratory reports. Once methodically identified, categorized according to ability level, and validated by STEM instructors, digital instruction will be used to help students improve their scientific reasoning processes. With the singular goal of structuring student success through asynchronous machine learning, this innovation holds the promise to figure meaningfully in discussions of national competitiveness, education policy, innovation, and diversity as related to STEM education.