Insights On The Role Of Data In EdTech Innovation

This new Stanford study might change how we think about AI in education. Everyone’s talking about AI that writes lessons. But what about AI that understands students? The study, from Stanford University and Carnegie Learning, found that just 2–5 hours of student interaction with an edtech tool can predict end-of-year test performance with surprising accuracy. In some cases, it matched the predictive power of full-year data or even a formal pre-test. AI’s real value in education might not be content generation (e.g. lesson planning or rubric generation). It might be early prediction—the ability to identify struggling students before any test is given. That’s the bet we’re making at Flint. We’re not just helping teachers generate materials. We’re helping them understand where students are, how they’re progressing, and what to do next. All in real time via an army of AI teaching assistants. The next generation of AI edtech tools will focus on what students need—and when. Full study (and overview) linked in the comments 👇 #ai #edtech #aiedtech #flint

I've always believed that assessment is the unlock for systemic education transformation. What you measure IS what matters. Healthcare was transformed by a diagnostic revolution and now we are about to enter a golden era of AI-powered diagnostics in education. BUT we have to figure out WHAT we are assessing! Ulrich Boser's article in Forbes points the way for math: rather than assessing right answer vs wrong answer, assessments can now drill down to the core misconceptions in a matter of 8-12 questions. Instead of educators teaching the curriculum or "to standards" we now have tools that allow them teach to and resolve foundational misunderstandings of the core building blocks of math. When a student misses an algebra question is it due to algebraic math skills or is it multiplying and dividing fractions? Now we will know! Leading the charge is |= Eedi - they have mapped millions of data points across thousands of questions to build the predictive model that can adaptively diagnose misconceptions (basically each question learns from the last question), and then Eedi suggests activities for the educator or tutor to do with the student to address that misconception. This is the same kind of big data strategy used by Duolingo, the leading adaptive language learning platform. It's exciting to see these theoretical breakthroughs applied in real classrooms with real students! Next time we should talk about the assessment breakthroughs happening in other subjects. Hint: performance assessment tasks - formative & summative - are finally practical to assess!! #ai #aieducation Edtech Insiders Alex Kumar Schmidt Futures Eric The Learning Agency Meg Tom Dan #math Laurence Norman Eric https://lnkd.in/gxjj_zMW

Now that I have had a chance to reflect on ASU+GSV Summit this year, and naturally AI dominated the conversation and showcases, I can’t help feeling that edtech is going in the wrong direction. Most new products focus on automating or replacing teaching itself. These bots often function merely as advanced information retrieval – a conversational alternative to searching online. But is easier information access the primary barrier to better learning outcomes? Or are real challenges student motivation, connection with teachers and classmates, or lack of personalization due to too high student-teacher ratios? AI's true potential lies not in content or Q&A, but in augmenting humans and fixing systemic issues. Focus areas: 1. Comprehensive Learner Profiles: Effective teaching is personalized. Most edtech fails to track student strengths, weaknesses, and preferences persistently from their own interactions, let alone across systems; go from one grade to the next and all the history is lost. Unifying millions of data points into a dynamic learner profile, informing teaching systems, is key to personalization and moving towards one-student-one-teacher. 2. Process Automation for Educators: Teachers spend ~46% of time teaching. AI should automate administrative, non-teaching tasks consuming the rest. Freeing educators for teaching and mentoring is more impactful than automating pedagogy. 3. Empowering the Learning Ecosystem: Especially important for young learners - AI should empower parents to be educators. Person story - my 5-year old needed help with “54-19” without using a number line…and I couldn’t come up with an approach she could use. ChatGPT, Gemini, and Claude all failed to produce anything better other than giving her 54 blocks to subtract 19. Googling I found a YouTube video that had great suggestions, and it also let me create a lot better prompts that eventually I could help my daughter. But most parents wouldn’t think to do this. This is where a better purpose built model would work great. I foresee several developers in the next few years: 1. Most standalone LLM wrappers will fade as districts standardize on integrated platforms from major providers. 2. Schools implementing comprehensive learner profiles (when they arise) will see most significant outcome improvements. 3. The gap between high- and underperforming schools will widen. Under-resourced schools will 'outsource' pedagogy to chatbots. Those chatbots will fail, get replaced with the next set. The next set will eventually fail, get replaced, cue “rinse and repeat”. But it's not all bleak (more in comments)