Online education platforms scale college STEM instruction with equivalent learning outcomes at lower cost

GAICIL Insights

The scarcest input in STEM education may be the professor, and the most promising substitute may be a well-built platform.¹ Online education platforms scale college STEM instruction with equivalent learning outcomes at lower cost, by Igor Chirikov, Tatiana Semenova, Natalia Maloshonok, Eric Bettinger and René F. Kizilcec, attacks a binding constraint on economic growth: the global shortage of qualified science, technology, engineering and mathematics instructors. National online education platforms let resource-constrained universities adopt online courses produced by a country's top universities and departments.

The evidence comes from a multisite randomized controlled trial on Russia's national online education platform, testing fully online and blended instruction against traditional in-person teaching. Online and blended instruction produced similar student learning outcomes to in-person instruction at substantially lower costs. Adopted at scale, the model reduces faculty compensation costs, and the savings can fund increases in STEM enrollment. Equivalent learning, cheaper, and the difference can buy more students.

The architecture deserves as much attention as the result. The platform licenses courses created by top universities to institutions that lack qualified instructors. It provides the information technology infrastructure and supports universities with the design, production and delivery of courses. Learning outcomes, topics, required reading and assignments are standardized, with iterative quality assurance from domain experts and psychometric evaluators. Universities can offer the courses for credit fully online or blended, with local instructors supported by lecture slides, assigned reading and problem sets. This is a supply chain for knowledge, with quality control at every stage.

The parallel with today's technology is close. Artificial intelligence slots naturally into that chain. AI tutors, automated assessment and adaptive content are the next layer on such platforms. The central finding establishes that instructional quality can be decoupled from the physical presence of a top instructor when content is carefully designed and quality-assured. AI promises to push the decoupling further, personalizing feedback and practice at a scale no faculty could match.

The order of operations is the hidden lesson. The gains documented here rest on standardized learning outcomes and expert review. AI tools layered onto weak curricula would scale weakness. AI tools layered onto platforms with this rigor could scale excellence, extending access to students at institutions that cannot hire enough qualified faculty. Infrastructure and quality assurance come first; the chatbot comes later.

The authors are candid about the hurdles. There are substantial start-up costs for the platform, for faculty professional development to improve online teaching and for introducing new models of instruction. The state or a consortium of course-providing universities could cover them. Platforms work more efficiently where academic programs are synchronized in learning outcomes and calendars, and accreditation boards and professional associations could help harmonize curricula. Cross-country variation in impact warrants investigation, and more research is needed on how combining modalities affects learning and career outcomes for students with diverse backgrounds.

Each hurdle has an AI twin. Start-up costs for AI-enabled platforms include model development, data infrastructure and teacher training. Harmonized curricula make AI tools easier to deploy across institutions. The open question about diverse students is especially pressing, since AI systems may serve some groups better than others. Trials like this one are the right method for answering it before large-scale deployment.

For policy, this is a growth blueprint. Governments seeking a larger STEM workforce face an instructor shortage that training more faculty cannot solve quickly. Shared national platforms, combined with blended teaching, can expand access while maintaining outcomes and lowering costs. Savings reinvested in enrollment create a virtuous cycle of skill formation that supports innovation across the economy.

The thread running through it all: technology can relax constraints that once seemed fixed. When the best courses can be shared across institutions with equivalent results, the scarcity of top instructors stops capping the supply of STEM education. As AI enhances these platforms, the potential grows, and this study shows how to realize it responsibly through design, quality assurance and experimental evaluation.

Something for education ministries to sit with: the fastest way to build a national AI tutor may be to build the boring national course platform first. Without the plumbing, the intelligence has nowhere to flow.