Stephen Downes

AI agents were set loose in SimCity (an online virtual city simulation) and linguists studied their interactions. They tended to adopt their own slang and jargon (though less so if agents from different LLMs took part). "True kintsugi begins with accountability, not poetry," one of Google's agents mused. "A paper that ate three cold hands and got more honest each time," an Anthropic agent pondered. I wondered whether similar work had been done with children in SimCity, but when I looked all I found were uses directly related to learning outcomes. This, for example: "As game-based learning gains momentum in education circles, teachers increasingly want substantive proof that games are helpful for learning. " Via RestOfWorld.,

There is obviously a difference between using an LMS to build a course and using an AI, but Hollis Robbins isn't completely wrong here. "You are not wrong to insist upon the principles of good teaching, like the importance of writing, human feedback, instilling a spirit of curiosity and discovery," writs Robbins. "But course management software, which ninety-nine percent of colleges and universities now use, violates all these principles." Like AI, the LMS promises to make the hard work easy. "Nobody is protesting Canvas because the laziest teachers love it. Canvas sells itself as a lazy-teaching technology." Now - maybe a lot of this work should be easy. But then, why not apply the same reasoning to the use of AI?

I had a couple of hours this morning so I created a collaborative slide editor. Oh sure, it's still very primitive, but I have bigger plans for it. So I'm right with Miguel Guhlin here: "In the same way, blogs democratized publishing for every person, vibe-coding leverages Gen AI to create just-in-time solutions to problems, cutting out the intermediary enterprise-level company and their armies of professional coders. Is this a good thing? Yes, yes, it is." He provides examples of his own, but my own experience convinces me. I was looking at various products when I wondered, "well, what if I just code my own?" A bit of work with ChatGPT and there it is. I've replaced an estimated six weeks or work with a morning over coffee.

I haven't tried this - I don't have the AI tokens to spare - but the concept intrigues me (I tried the live demo but the credits disappeared quickly). "OpenMAIC turns a topic or a PDF into a full interactive classroom — AI teachers, classmates, a whiteboard, and quizzes — running on your own machine. This guide installs it with Docker, wires up a model, and generates your first lesson." This is a real challenge, I think, to what traditional companies and institutions are offering. Limitations are honestly stated at the end of the article, as well as a genuine success measure: "The useful test is not whether the classroom looks impressive — it does — but whether you learn faster from it than from the document alone." I would insert "faster and better" as the test. It's all open source on GitHub and you can run it on local AI models, it just won't be as good.

This is the sort of thing I like: "Centuries of copyright-free art, historical photography, illustration and archival footage. Find high-resolution files by describing what you're picturing, free to use forever. Every image and video is CC0 or public domain. Free to download and use, with no strings attached." How can they afford to do this? Decentralization: "Arkivism doesn't host the collections — the works belong with the museums and archives that keep them, listed with credit on the Sources page." There's also a search for video clips. Via Grant Potter.

This editorial (10 page PDF) introduces a special issue on AI-aided systemic reviews of research in online learning (also styled here as Open, Distance, and Digital Education (ODDE)). The authors respond to recent criticism of systemic reviews and survey some recent lapses, including "considerable variation in methodological approaches, reporting standards, and the transparency of review procedures." None of this is helped particularly by AI: "increased efficiency does not necessarily translate into improved methodological quality." So they propose standards for AI-assisted reviews that include stipulations that "methodological criteria must precede automation," transparency about the role of AI, systemic validation of AI decisions, and review process that are "sufficiently reproducible and auditable." There's nothing wrong with any of these, but they seem to me to set a pretty low bar. We're still in the era of (almost) any data counts, and (almost) any theory is a good one.