Stephen Downes

Victor Kumar posted: "You should never offload cognition. Always think for yourself. That's why I don't read." The reference is to Schopenhauer: "When we read, someone else is thinking for us: we merely repeat his mental process..." The point, says Martin Lenz, is that "we seem to have handled cognitive offloading through various means quite well in the past, suggesting that AI, and especially LLMs, might just turn out to be the next form of refined tool use." Ah, but there's a tipping point. We should be cautious. "We've reached a level of offloading in food production where we increasingly allow for outright self-harm by stuffing ourselves with fast food." The same could happen with AI. It's not cognitive offloading itself that is the problem; it's what happens if offloading turns into self-harm.

Just for fun: "Air Theremin - a browser theremin played in mid-air: tilt your phone, or wave both hands in front of the camera. Pitch, volume and timbre follow you, with note snap, cave reverb and recording." Like any musical instrument, it will take a lot of practice to learn how to play anything worth listening to. But yes, it is fun. Via dogtrax.

I like Julian Stodd though his work comes in fragments, making it hard to see the whole. This short post is one of few giving a broader perspective (and it helps to read the two or three posts preceding it). Here he outlines "a clear picture of a shift, away from ownership, structural technology, and control, towards a focus on collaboration, application, and more rapid experimentation." His recommendation: "develop specific expertise in certain new areas – like supporting experimentation, focusing on collective capability, and narrative aspects of learning – whilst letting go of a great deal more." I would use the term 'cooperation' instead of 'collaboration', but otherwise am more or less aligned with this perspective.

According to this summary article, " the CommonsDB pilot has proved that purpose-built, content-derived registries operate reliably at production scale." Importantly (to me, at least), "A recurring theme throughout the event was the rejection of centralized, single-source databases in favor of a distributed, federated network of community-governed registries." There are many moving parts in a system this complex, and the intent is to show "that open standards, cryptographic trust, and federated networks can deliver a scalable, public-interest copyright ecosystem." The full set of slides from the conference is available here.

The problem with traditional academic papers, according to the authors (18 page PDF), is that people need to read, understand and implement the data sets and code bases. They introduce a new product, Paper2Agent, that convers the paper into an AI agent that can be accessed by other AI systems and incorporated into other projects. Code (it's implemented as an AI skill that represents the paper as an MCP server) is availabel on GitHub, along with some examples. "Looking forward, just as many journals now require data and code availability sections, we anticipate the emergence of an 'agent availability' section that specifies whether and how the contribution has been embodied as an interactive agent." Related: my own article, The Present and Future Landscape of AI in Scholarly Publishing.

UNESCO has released a series of background papers, which can be accessed here, to support a consultation (responses due October 15) on these themes:  sustaining education as a common good, digital and AI sovereignty, age-appropriate AI, learner safety, AI procurement in education, total cost of ownership, and explainable AI. I haven't had a chance to review any of these, but I felt I should get the information out there, given the short comment deadline. I'll see what I can do to offer my own analysis.