Stephen Downes

'Congress' is a large annual gathering in Canada of thousands of reserachers and dozens of separate conferences in the social sciences and humanities. One of my earliest papers, "Could Hume Play Billiards?" was accepted at Congress in Calgary in 1994, but I didn't make it because my car broke down in Edmonton. I've been once since (2012, Waterloo) and I'll be in Toronto this June to present three times (once at CNIE, twice at OTESSA). But this may be the last in-person Congress, because it just doesn't make financial sense any more. What's really interesting to me, though, is that academics will now learn that simply replicating the in-class experience, the way an academic conference does, is not such a great online experience. What works better? Well, that has been the subject of this newsletter, and my work, for the last 30 years. Maybe come see me while you can. :)

This site describes Boltzmann machines and demonstrates a tiny Boltzmann machine you can run in your browser. "Boltzmann Machines are used for unsupervised learning, which means they can learn from data without being told what to look for. They can be used for generating new data that is similar to the data they were trained on, also known as generative AI." As the site says, it mimics how energy works in physics, and more specifically, 'anneals' the connection weights by varying sensitivity, thus settling into an overall lowest-energy state, rather than into local minima. See more simulations from the same author and also some primers on key concepts.

This article may provide you with a little bit more information about the Model Context Protocol (MCP), a standard used by applications to expose services to AI systems. It includes an example of an MCP directive (though in a syntax that is unfamiliar to me (though it may just be comments)). The main reason I'm running this post today is to restate a point I heard on TWIT this week, that MCP is likely a transitional technology, and that future AI will probably not need a standardized way to access application services.

Yes, it's another AI literacy framework (43 page PDF). This one is a joint EC and OECD initiative. Here's the definition: "AI literacy represents the technical knowledge, durable skills, and future-ready attitudes required to thrive in a world influenced by AI. It enables learners to engage, create with, manage, and design AI, while critically evaluating its benefits, risks, and ethical implications." There are three major themes: how AI and machine learning work; human skills for AI collaboration; and AI's effects on individuals, society and the environment. And there are four major domains: engaging with AI, creating with AI, managing AI, and designing AI.

There are two parts to Ben Werdmuller's argument. The first is that people who develop new platforms should be able to earn a living. The second is the outline of a proposal for pro-social funding. I don't really disagree with either part; everybody needs an income of some sort, and the world does need things like open social software. But is the 'founder' approach really the way to go? I mean, even Werdmuller's admittedly progressive vision proposes, "Fund the founders, not the project... long-term success is tied more to founder mindset than to initial product ideas." In other words, "fund people like us". People who believe their service should make money. I don't think you can create essential social services on the VC model. The benefits are not realized by the providers of the service, but by the people using them, and there's no way to charge for the service without undermining access and renting it out to private interests. That's why they must be provided (and paid for) as public services.

According to this report (29 page PDF), "students benefit when institutions and instructors take a proactive approach to AI use." Nonetheless, "institutional support for student use of AI remains a challenge" while most research focuses only on writing tasks. This report broadens the question, asking how "student use and perceptions of AI in higher education impact student engagement." It's interesting, because while some students do indeed use AI to write assignments, we see more use dedicated to things like idea generation, time-saving on routine tasks (like grammar and spell check), tinkering with ideas, and summarizing articles. "Participants turned to AI when they believed assignments to be repetitive, redundant, or busy work." On engagement, the paper argues that "AI deepened emotional engagement, such as interest, meaningfulness, and belonging." And they note that "The transcripts also make clear that students are going to use AI whether it is allowed or not. This tension is not going to resolve itself."