Stephen Downes

Videos and notes from the FediForum conference I attended last week. Most recommended is Anil Dash's keynote address in which he laid down some truth on app developers hoping to jump on the VC gravy train (sadly there's no transcript). There's also a recording of the presentation on ATmosphere Spaces and a number of live demos, including the IndieBeat radio stream, Acorn Bluesky communities, the FediDeck Tweetdeck alternative for Mastodon and Bluesky, and the AllPeeP community app.

Some realistic thinking about the role of AI in philosophy: "With a little human guidance, you can crank out the kind of papers that make up 95% of what we publish in a few hours. Soon (already?), human guidance may not even be necessary. Do I find this depressing? Just the opposite. Sure, I built my career on this skill, honed it for decades, and now a computer can do it in a hundredth of the time. In some ways, it even does better work than I ever could. But this was always a goofy way of making a living. And, in truth, it's always been one of the worst things about the job." It really depends, I think, on what you valued about the work. If it was for being recognized as deep and erudite, then AI isn't for you. But if what matters to you are the questions themselves, AI might be what you need. Imagine, for example, AI solves the problem of consciousness. It's a loss only if your job was based on perpetuating the question. "What do we lose if LLMs kill the journal system? Nothing but some heavy ass chains, and a few weak heuristics for assessing CVs quickly." Via Justin Weinberg, who offers a feeble response defending the continued counting of beans.

I can accept some version of the evolutionary description of human cognition (illustrated). As a map it probably suffers from some overgeneralization and misattribution, but I wouldn't think it's in any way fundamentally wrong. Human cognition is complex; nobody denies that. Where I disagree with Benjamin Riley is in the characterization of computational thinking, described as 'cognition' in another diagram, and composed of attention, reasoning, decision making, working memory and long-term memory. It's the usual use of the 'mind as computer' metaphor so many cognitive scientists love. But modern AI-based computation doesn't work like that at all. It's complex, so complex that explaining how it works is a major challenge. Is there any reason an AI could not function as described in the illustration? Probably not, though we would ask, why would it? It's the human brain that struggles against evolutionary limitations, not AI.

Here's the argument: "Skilled teachers do more than deliver information and feedback. They exercise situated professional judgement, recognise learners' changing needs, support intellectual independence and create the social and knowledge relationships through which disciplinary understanding and professional identity develop. AI can augment these activities, but automating instructional functions does not replace teachers' educational role." First of all, I think that overstates what teachers do - people might be assisted in  creating relationships, but this is done by the person, not the teacher. And it understates what is very likely to be within the grasp of AI: judgement, recognition, and support. More to the point: I think that in the future we'll skip the 'teaching' step altogether. We'll go from wanting to do something to doing it, with the knowledge and support we need available to us on the fly.

David Hopkins asks the question and comes back with a number of different answers, everything from "concentrat(ing) on the structure of a learning experience, working with subject specialists to consider activities, sequencing, engagement and assessment" to "work(ing) more directly with platforms, implementation, staff development and support." I would imagine there are more roles than listed here. "They need to understand accessibility, inclusion, assessment, copyright, data protection and the policies or regulations within which their institution operates. They may need skills in communication, training, project management, evaluation, procurement and organisational change. They must also be able to work with people whose needs, confidence and priorities vary enormously."

There's a lot of interesting thinking going on here, though the implementation is a little but clunky - the design is fine, but scrolling is not smooth and there's a short lag every time you click a link. The site is called Interpoll and the idea is twofold: first, there's no central server at all, everything runs peer-to-peer, and content is replicated as it propagates; and second, "a key lives on your device and signs every action you take. No relay or server can forge a vote in your name." It's based on a "a peer-to-peer database (GunDB). Every connected device holds a copy" (I'm totally not on board with that branding). There are some complaints on Reddit about the use of AI, but on the other hand, the creator is a 14 year old who enjoys coding. Similar projects include Nostr and Scuttlebutt (more).