In AI We Trust?

There are a few genuinely useful operational ideas - the 'narrow circle' privacy approach to adaptive learning and deliberately simulating AI errors in curriculum - but they are surrounded by extended motivational framing, platitudes about curiosity and agency, and podcast throat-clearing that dilutes the substance across 43 minutes.

“we look at, in the context of a particular lesson, how is that student interacting with that lesson in that moment. So their behavior right there affects how the AI will interact with them. So it's very localized. It does not include broader context”

“We actually create mistakes on purpose because at the end of the day some of these models are not deterministic. We can't assume it's going to make the mistake we want it to make”

The reframing of digital/AI science as foundational like physical sciences (the 'dissecting a frog' analogy) is a modestly fresh way to make the case for AI literacy, and the distinction between computational thinking and the emerging need for systems-level thinking has some conceptual value, but the rest of the episode follows a predictable ed-tech nonprofit narrative arc.

“this shift from computer science as a vocational, uh, domain to a foundational science, I think is one that we don't talk about enough”

“we'll not only teach computational thinking skills, I think we'll go up a step and start to teach a bit more about systems thinking as we work with AI systems”

Karim Meghji is a genuine practitioner with 30 years of software engineering and real organisational scale (100M+ students, 3M teachers), and he speaks credibly from operational experience rather than as a thought-leader, but his domain - nonprofit K-12 education - has limited direct applicability to the B2B operators this index targets.

“I spent time pursuing a computer science education path. Went into software engineering, building technical products for the last 30 years”

“We actually began working with AI tools on the teacher side first”

The episode offers a reasonable set of concrete numbers (33M Hour of AI learners, 100M students, 3M teachers, 2018 AI curriculum start date, 18%/34% Gallup-Walton teacher guidance figures) and a couple of named examples (AI for Oceans vs AI for Farmers, Blockly library), but lacks outcome metrics, financial data, or granular evidence of learning impact.

“only 18% of K through 12 teachers have received formal guidance from their administrators on how to use AI in the classroom. 34% are getting no guidance at all”

“we have something called AI for Oceans... we've just released a version of that for the India context called AI for farmers”

The host asks a few pointed questions (age-differentiated critical thinking, the tension between personalised learning and privacy) but consistently validates and agrees rather than probing or pushing back; the interview ends with an apologetic pivot to the most newsworthy item and no follow-up challenge on the EU/OECD framework's enforceability or outcomes.

“I apologize. I have buried the lead”

“I'm hearing a theme from you today that I'm hearing from a lot of thoughtful people in this field where you're only using the data necessary”