Can AI instructional coaching end the achievement plateau?
Doubtful, but it's sure gonna do something.
Many years ago, during a jog, a buddy and I were discussing the importance of coaching for running form. We’d experienced our share of common running ailments — IT band syndrome, sore achilles, patellar tendonitis — that we were considering hiring a running coach to help our form. Coaches were expensive and inconvenient, though, and many of the best coaches were located in running meccas — inaccessible to many runners. “Wouldn’t it be great if we could just record our running form and have a coach virtually analyze it?” we mused.
Some time after, that buddy, Dave, started a company, Sibme, that aimed to apply that concept to instructional coaching. He’s been incredibly successful at providing a platform that allows teachers to record themselves and instructional coaches to provide timestamped feedback that teachers can see as they watch themselves teach.
Now, though, he and Sibme are taking instructional coaching to an even new level with Artificial Intelligence (AI). To be clear, I have no stake in Dave’s company, and even though we’ve been friends since we started teaching together in Houston in 2004, he did not ask me to write this post as free marketing, nor have I run a draft by him for vetting. If anything, Dave still owes me for stealing a box of Kashi cereal in 2007 when we cohabitated as underpaid teachers.
Dave recently demo’d Sibme’s AI tool, powered by ChatGPT, for me using that lesson that I taught many years ago, and I was blown away at what it can do.
First, it can transcribe everything that’s said in the video. Much of that depends on audio quality, but even for this video, which was recorded in 2014 with an older generation iPhone, its transcription was remarkably accurate. Transcribing audio itself is fairly unremarkable, but the transcription enables it to immediately generate data and insights that are quite remarkable, like:
A lesson summary. Aside from getting my name wrong (admittedly, B and V sound alike), this is pretty accurate. A short summary like this could be valuable in helping an instructional coach understand the “gist” of the lesson, freeing them up to provide more targeted feedback for the teacher.
Teacher-student talk time. Coaches uses this data point almost like a “temperature check” for a teacher’s instruction. If it’s abnormally high or low, it can indicate an underlying issue for the coach and teacher to address. Even the sharpest coach would need to take such careful notes to be able to accurately determine that ratio. Previously, a guess would suffice: the teacher did most of the talking, not the students. The AI bot does it in seconds.
Question analysis. The AI bot can immediately analyze the questions asked and determine if they’re open-ended or closed. Coaches use this metric to gauge the level of rigor in a classroom, but even experienced coaches can struggle to transcribe every question during an observation.
Alignment to lesson plan. The coach or teacher can upload a lesson plan template and ChatGPT can analyze how well the teacher executed the written lesson. My lesson didn’t have a formal lesson plan (this was me as a teacher in a nutshell), but Dave demo’d this for me with another lesson, and I was shocked at thoroughness and accuracy of the AI bot’s analysis.
Scoring the teacher’s lesson. I buried the lede. Most incredibly (and absolutely most controversially), the AI bot can rate the lesson when provided an instructional evaluation framework, like the Danielson Rubric. Not only can the AI bot score various subdomains on the rubric, it can add descriptive evidence for each portion of the rubric. To be clear, no districts are currently using AI to evaluate teachers, and I imagine this will be a thorny issue for policymakers, education leaders, and teacher unions if it does get piloted.
Will AI instructional coaches replace human coaches? Dave was skeptical about this, believing this will be more of a tool to enhance human coaches’ ability. This might be akin to NFL (see the NFL’s strict rules on tablets) or MLB coaches or players grabbing the tablet to review video of their last play or at-bat alongside immediate analysis of that play.
I’m less skeptical. As I’ve written before, coaching is an incredibly expensive (and as a result often inefficient) way to train and develop teachers en masse. The metrics it currently provides lend themselves to moving teachers from novice to proficient. I could see a world in which novice teachers record lessons early and often and get immediate feedback on their teaching from the AI bot, especially if it’s a fraction of the price of a human coach. Teachers would avoid the shame of having to share their imperfections with colleagues and potentially get more frequent feedback — as much as they can upload, in a way (at most coaches commonly work with teachers once a week). More advanced teachers would still benefit from the expertise of a human coach, who might be able to notice more complex elements in a teacher’s approach, like their explanation of a concept or their feedback to students.
At some point, though, as these AI models get more and more data on teaching and learning, it’s not unfathomable to think they could do what even the most expert coaches currently do. And if they can replace the coaches, at what point could AI replace the players?
Thanks for reading. Have a great weekend!