Thinking Forward: Anniversary Edition

Happy one year to my newsletter, plus more on new edtech tools, the paradox of AI advancements, and some notable new research

I was shocked to learn via Substack that it has been a year since I started the Think Forward: Learning with AI newsletter. Time really does fly when you are having fun.

I started this newsletter simply to share what I was learning about Gen AI and how it might intersect with education. Now that we are a year in, with more than 10,000 views in the last month alone and subscribers in every state except North Dakota (!), I'd love to know what you enjoy most about this newsletter and what we could do better. Based on the Substack stats, many of you seem to appreciate deep dives into specific AI education topics like math and assessment, timely responses to current events in AI and ed policy, and interesting research and data summaries. What would you like to see us cover over the next year? Leave a comment with your thoughts. (On mobile, the comment button should be on the bottom of the page; on desktop, it should be at the top.)

---

New Ed Tech Tools, Debates, and Discussions

Andrew Ng (of Coursera fame) is launching Kira Learning. According to one report:

The platform embeds AI directly into lesson planning, instruction, grading and reporting. Teachers can instantly generate standards-aligned lesson plans, monitor student progress in real time and receive automated intervention strategies when a student falls behind.

Students, in turn, receive on-demand tutoring tailored to their learning styles. A.I. agents adapt to each student’s pace and mastery level, while grading is automated with instant feedback—giving educators time to focus on teaching.

I recently watched the terrific discussion below, hosted by Leanlab’s Katie Boody Adorno with panelists Malvika Bhagwat (Owl Ventures), Arman Jaffer (Brisk Teaching), and Peter Gault (Quill). They had a really great conversation about how ed tech developers can design more effectively and ethically with educators as partners. My time at the ASU-GSV conference made me think this is essential. I’d add, though, that we won’t see breakthrough ed tech tools unless they’re designed alongside people who understand what prevents schools and school systems from achieving breakthrough results.

The paradox of Gen AI advancements

Two years after the release of Chat GPT-4, Generative AI is advancing at an incredibly rapid pace. I often find myself updating newsletters mid-stream as powerful new tools are released every day.

On the one hand, the advancements are simply staggering. Just this week:

• FutureHouse announced the first scientific discovery made by a multi-agent AI system—named Robin (!). Robin proposed a new potential treatment for macular degeneration that no one had previously considered. I recommend you watch the explanatory video here.

• Researchers at the University of Pennsylvania used AI to identify a new antibiotic compound from the DNA of a woolly mammoth. This compound has demonstrated effectiveness against drug-resistant bacteria, performing comparably to existing potent antibiotics.

• A new study on AI and emotional response suggests that LLMs can now effectively understand and replicate complex human emotional processes.

On the other…

As LLMs become more powerful, they are more prone to hallucination. According to this article, AI errors seem to be increasing, even compounding, with improved data access and feedback loops. The article also reminds us that “we don’t really understand how these models work.” Of course, educators should proceed with caution, but this is also a good opportunity to encourage students to fact-check their work and to be skeptical consumers of all information.

There may also be growing reason to be concerned about metacognitive laziness, i.e., people becoming overly reliant on AI for homework and work assignments.

And finally, there’s the always-concerning digital divide. Of note, President Trump recently announced the termination of the Digital Equity Act, a $2.75 billion federal program to support digital access for low-income communities, rural populations, racial and ethnic minorities, seniors, and veterans.

All of this points to something of an inherent paradox with AI that I’ve noted repeatedly over the past year: AI brings both opportunity and risk, wicked problems but also wicked opportunities.

---

Notable New Research

From Chalkboards to Chatbots: Evaluating the Impact of Generative AI on Learning Outcomes in Nigeria

A recent study explored the impact of AI in classrooms by introducing Microsoft Copilot, powered by GPT-4, into after-school English programs for high school students.

Students who used the AI tutor showed significant improvements in English, digital skills, and basic AI knowledge. The gains were especially strong for girls and students who were already doing well academically.

The findings show that when used thoughtfully, AI could help bridge learning gaps and expand educational opportunities, even in settings with limited resources. However, thoughtful use and reaching the students most in need can be an elusive goal.

ChatGPT’s Impact on Student Learning Performance, Learning Perception, and Higher-Order Thinking: Insights from a Meta-Analysis

A meta-analysis of 51 studies conducted over the past two years found that ChatGPT is having a large, positive impact on student learning performance and moderate positive impact on how students feel about learning and in their critical thinking skills.

The key takeaway, though, is that ChatGPT works best when it’s integrated thoughtfully, tailored to the subject matter, and used consistently over time. Once again, it’s not just about using AI, but using it wisely.

A note regarding both studies: Every study, even those using experimental designs, has its own limitations. Many such short-term studies are underway, in diverse contexts with varied results. Researchers in both studies raise concerns about over-reliance on AI-powered learning tools despite seeing positive results and point to other limitations. It will be critical to have ongoing and careful syntheses of new research as we continue to make sense of the interaction between Gen AI and learning outcomes.

---

Final Thoughts

It’s a shame, because some of these sound like really good reads. (404 Media)

A reminder to always, always fact-check. A few major US newspapers, including the Chicago Sun-Times, published an AI-generated "summer reading list"—but only five of the 15 books listed were real. Yikes! If you want a list of summer reads that aren’t made up, my editor recommends this one from The Atlantic.