The Power of Collaborative AI

If you were to read a random sample of articles on generative AI’s potential impact on teaching & learning, you’d come away with a fairly bleak view on the state of classrooms. The prevailing thinking seems to be that teachers are inundated with thousands of short, often unrelated tasks. And that the most humane thing that we can do as developers is to take some of those off their plates; find a teacher's pain point and smash it. Repeat.

At CommonGood we have a very different view on teaching. We believe that most teachers are deeply interested in big ideas, ways those ideas connect to their students’ lives and planning learning experiences that will make them come alive. We think that given the option, teachers will dedicate a large portion of their attention to this process. Rather than casting teachers as the task mistress that barely keeps chaos at bay, could we behave as though teachers are called to deeply think and expertly plan as a fundamental part of their practice?

In a recent interview OpenAI CEO Sam Altman suggested that we will soon transition from using AI to solve small, five minute problems to solving more complex problems. “Someday they’ll do 10-minute tasks, and then they’ll do an hour-long task,” Altman said. “But you’ll still have to think about, ‘How is this all going to fit together? What do I want to build?’” That sounds more like it. At CommonGood, we don’t see supporting teachers’ core work as a five minute problem - and we have an opinion on how things fit together.

For around a decade, our founders have gathered educators in cohorts to deeply consider their students’ communities, the learning that will be most compelling to them, and flex our collective learning science muscles to design learning experiences that will lead to good outcomes. These processes are time consuming, intellectually rigorous and sometimes personally demanding. They often challenge our assumptions, shine new light on the communities we serve and stretch our capacities. Educators also love them - this work can build new mindsets and communities of practice. Some have even said that it reinvigorated their love of teaching.

That’s good stuff. Perhaps even better is what these approaches yield for students. In a recent paper, we lay out how similar approaches don’t suggest incremental improvements to student outcomes. They suggest that transformation is possible. 

But these approaches require expertise and dozens of hours of work for participants. Few communities have the bandwidth to take this on. Enter Collaborative AI. CommonGood Co-Founder Kyle Morton designed a workflow management platform, with AI-powered supports for each step in the design process. Like the cohort-based approach, teachers are presented with the underlying theory for each step, along with recommendations and ideas on how to execute it. The educative process both builds fluency with evidence-based learning design techniques and maintains the centrality of the teacher. 

Even maintaining healthy skepticism and high standards for outputs, the affordances of the technology showed promise early on. “I’m a self-described design snob” said Co-Founder Dr. Carly Muetterties. “I think that because the design processes are so well defined (leaning on peer-reviewed research and frameworks, lots of examples to reference) that the technology is very good at following those guidelines and making contextualized recommendations. The tools added value to my own design process very early on.”

Well-defined processes with lots of reference material is key to the shift being proposed. Using generative AI (or really any technology) to identify a small bug or teacher pain point and smash it doesn’t require much fluency with education theory or learning science. It presumes that when a product has been built to smash a bug, those involved will use the audience that they’ve built to find / smash bigger bugs. But it can be difficult to make the transition from peripheral to more core problems of practice. By starting with well-established evidence-based models, there is already a tremendous amount of information on what works, how pieces fit together and what good outputs look like. 

We tested and iterated on the design process over the course of years, working with and gathering feedback from lots of educators and across a variety of contexts. We’ve compiled a huge amount of data on how design steps fit together, how to coach people on them, and what predictably leads to usable outputs. We’ve done a lot of (peer reviewed) writing on these processes as well. An evidence-based approach is by its definition, well-defined.

And generative AI seems well suited to facilitate defined processes, particularly when there are lots of examples to reference. “When we set out on this build,” shared Dr. Carly Muetterties, “we’d hoped to make the process more efficient for ourselves and our clients. We have been successful there - in our use we realize a 70-85% efficiency gain. What we didn’t entirely expect is that we’re discovering that the platform can make these approaches even more effective. In my own practice, the platform generates well-reasoned recommendations for me to consider (which pushes my thinking), surfaces sources that I wouldn’t likely have found, etc.. It’s humbling sometimes, but the technology has made me a better designer.”

Suggesting that technology be used to facilitate established models has a few important implications. First, we should stop casting teachers as an ‘in over their heads’ class of quasi-professionals constantly combating burnout. Rather we should work on the assumption that they’re learning science practitioners, constantly using, testing and informing the evidence base on what works in classrooms. Second, we should build tools in close partnership with people who have practice applying that evidence base to core problems. We should expect to be able to see what experts, what models, what evidence are informing the technologies that are being developed for classrooms. 

Our platform is being tested by partners on curriculum teams at both school districts / operators and solutions providers. These partners are among the most discerning and demanding users we could find, most with decades at the forefront of evidence-based R&D. We are excited to share more about what we learn from these partners in the months to come.

In sum, we think that the best application of something as powerful as generative AI is to make proven models more accessible. Improving curriculum & instruction is not mysterious, just very difficult. Thankfully we have 100+ years of learning science that holds unrealized potential. Our use of generative AI so far is that it can make complex and difficult tasks less complex, and time consuming processes dramatically less so. We want to use this new technology to allow teachers to take on more - more inspiring, deeply impactful, even transformative work - not to do less.