eLearning and Technology

It's so tiring reading stuff like this: Somewhere between the death of our favorite aggregator websites and the world surviving a pandemic, the modern internet was reduced to four companies in a trench coat. You don't have to visit those websites. RSS still exists, aggregators still work. Blogs are still being published by thousands of people around the world. The only person stopping you from stepping outside a paywall-defined and/or toxic meta-site is you. That other internet exists - you have just conditioned yourself to pretend it's not there. Via Dan Gillmor.

"Meta is searching for a killer app for its Quest headsets, which today are primarily used for gaming," writes Jennifer Kingson. "Could it be education?" Um, no. At least, not as the headsets are currently designed, and not for (as illustrated) simulating a meeting across a table (we really don't need more than Zoom for that, or possibly, just a table). As this story makes clear, the real problem being addressed here is Meta's desire to sell more units. That's not a problem education needs to solve. Via Doug Levin, who comments, "it is apparently silly season in edtech."

Without open data there is no open AI (and no open science generally). "However," the authors write (9 page PDF), "the challenge of producing and openly disseminating data that are easily discoverable, accessible, interoperable, and reusable (FAIR) has emerged as a significant concern for policymakers." This article attempts to explore the reasons for that. To make a long story short, even if the researcher intends to abide by FAIR principles, institutions and repositories often have policies and regulations that make compliance difficult. "While some institutional repositories enforce policies like restricted open access, which reduces the visibility and reusability of research outputs, others impose eligibility restrictions on deposits, raise copyright concerns, and require funding for maintenance." I've argued with people that the default should be 'open', while restrictions should be the exception that requires justification. But that's the opposite of most institutions today.

I wish I could say it was just Missouri that is blocking all these websites, but I'm quite sure it isn't. There are two aspects: first, the deliberate blocking of websites, which is a political issue, which I will leave to the particular societies concerned; and second, the accidental blocking of websites, which I'm sure is why a site like NASA would be blocked (as well as, for me, at the office, the Open Education Conference), which results from (a) bad categorization, and (b) category-based content filtering. This is an education technology issue, and it is having a widespread impact, not simply because it makes the sites more difficult to access (not impossible, unless you're poor; you can just use your own phone or home computer) but also because it undermines our respect for the institutions that are managing access to information in such a sloppy manner. Via Doug Levin.

According to this article, "online mathematics programs may benefit most the kids who need it least." It's a catchy way to introduce the main argument, which is to say, the program works "only if used as intended", however, only five percent of students use it as intended, and these are already likely to be high achievers. I think the point is well made, but what's the solution? Enforce proper use? Not practical. Return all math instruction to in-person instruction? Also not practical, and it's also not clear that this would solve the problem (how many teachers teach 'as intended')? Getting the research right would be a start.

The scoring was applied to the State of Texas Assessment of Academic Readiness (STAAR) for Science, Social Studies, Reading, and Language Arts. As George Siemens reported, "the state hired around 2,000 human scorers this year, a small margin compared to the 6,000 employed in 2023."

I was going to build my own tool (a simple link to a website that says "No!") but while I get that ready (I've applied for funding) you can rely on this tool that "offers a curated list of questions designed to provide visibility into factors related to the institution's readiness for identification and adoption of generative AI solutions across three core focus areas: strategy and governance, capacity and expertise, and infrastructure." Here it is: 12 page PDF.

We want to make decentralized, but if we start there, we break a key principle of software design, which is that we must start with the user experience. Which sounds great, but when you start with the user experience it's really easy to make software that is very hard to decentralize later. So what, asks Gordon Brander, is the minimum we need to build to make our user-focused software decentralizable? His answer: Immutable data, universal IDs, user-controlled keys. "If you have all three," he says, "you have a fair shot at building first, decentralizing later." Sounds easy - but how to get even this far is hard.

Gobal Sustainable Development Goal (SDG) indicator 4.1.1a measures the proportion of kids in grades 2 and 3 "achieving at least a minimum proficiency level in (i) reading and (ii) mathematics, by sex." As reported here, data for this indicator has not been forthcoming and it faces deletion "because of a technical debate about measurement and a turf war between testing agencies." This UNESCO article "attempts to sketch the outline of a sustainable solution." It's an important indicator, because it measures where children "pass from the 'learning to read' to the 'reading to learn' stage". But it involves a series of tests, some of which only measure foundational skills (like phonological awareness) and not the overall objective (reading comprehension). This article recommends "a reporting scheme that relies on disaggregating or 'unpacking' the reporting of the minimum proficiency level by skill or subskill and to allow partial reporting if the assessment is not measuring the minimum proficiency level."

Consider the following research: I wonder what makes cars go forward. I get into 100 cars, press the accelerator in each car, and nothing happens. I conclude that pressing the accelerator does not make the car go. So what do you think? If you think this is pretty poor research, then consider the current study (65 page PDF), which looks at changing one variable - lower tuition fees - to see whether things move forward. They do not. "Relative to a synthetic match constructed from public universities in the Southeast, NC Promise did not affect first-year enrollment at the Promise institutions." There's some backpedaling in the discussion - "a need for comprehensive strategies that not only make college more affordable but also address other barriers" - but it feels to me that the five years of the program were really time wasted rather than an actual effort to effect change.

I saw this mentioned in the SoLiD discussion Matrix, and there are obvious links. This short post links to a longer document (66 page PDF) describing a distributed trust-based network. Now we've been down this path before, with things like Web of Trust (WoT). This is a bit different, with a basis in modern technologies like Distributed Identity (DID). And unlike Blockchain, it proposes a network of micro-ledgers, instead of a centralized ledger. I think, though, that schemes like this are putting the commercial cart before the relationship horse. We're still figuring how how to make distributed social networks work. We can't make the leap to a distributed contract network before we've even figured out how to talk to each other (without spam, harassment, disinformation, etc.).

I have to say that I'm not really feeling threatened by AI-authored content. But as Ben Dickson reports, Medium - which charges subscription fees for 'Membership' content - is struggling. Though its writers are prohibited from uploading AI-generated content, they have no real incentive not to, and Medium has no real way to detect it. According to Dickson, "business models that will work in the age of AI-generated content are those that directly connect content creators to consumers." Maybe. But we will have to get past the idea that they are 'content creators' and think of them as something else. For while we may 'create' content, that's hardly the half of it.

This is OECD's 'live' live repository of AI strategies & policies from more than 70 countries, territories and the EU. I didn't find the interface very appealing but you can download all the policies as a .csv document. I wanted to link to the source article from IBM where I found this (it came in an email newsletter) but it seems to have disappeared, though there's a copy here. What was really weird was that as I tried to find the original by searching for extracts (this is my go-to method for finding sources) I found them scattered through other IBM documents, suggesting that this article was formed by combining these other articles by some means.

Campus Technology reports on the founding of the Digital Education Council (DEC) "to address ed tech issues worldwide, including the impact of AI on education and work" by SuperCharger Ventures and a dozen or so business and management schools. DEC came out in March with one of the more puzzling manifestos I've seen, and there are a few more vague articles on the website (here's news: university leaders are concerned about negative impacts from AI). No RSS, naturally, but there's plenty of Google and Facebook tracking on the website.

This is a good article and I appreciate Neil Mosley's desire to engage with the argument in the article 'Online learning has all but destroyed the university experience' (from the Telegraph, and paywalled). The key point appears near its end: "more online learning in on-campus courses may lead to, or risk, students becoming isolated in their accommodation, thus preventing interaction and socialisation with peers and educators." Mosley walks a reasonable path in response, but then there's a turn two thirds of the way through, where he asks, "Is online learning doing enough to support healthy study habits?" We could answer the question either way, but the question isn't relevant. That's not the sort of "interaction and socialisation" that proponents of a traditional university education at the Telegraph are looking for.

This is an important (if self-serving) article that looks at how the development of foundation models in artificial intelligence will (once again) change the way we look at data. It's the latest transformation in a series that began with documents, then relational databases, graphs, vector databases, and now, foundation models. Now it's true that "the latest advancements in knowledge representation really do usher in a steep increase in technical sophistication and complexity." But the greatest skill will be demonstrated by those not lost in the complexity but those able to make these concepts accessible - at least at some level - to the people who need to work with the data. As for me - well, the learning never stops. This article doesn't make things easier, but it lays out the territory. If you're with me, have a go at it, follow the links, but be OK with not understanding it all (he's not trying to make it easy).

I have a subscription to Duolingo and it gives me feedback and tries to get me to continue my long-languishing Spanish lessons (I have no planned visits to any Spanish speaking areas and this might explain my lack of motivation). It has a cutesy little character unimaginatively called 'Duo' to try to make it like a friend. I can see that it could be like a friend, if it were better designed. Anyhow, this article interviews all of seven people (so few it actually names them all) and interviews 32. Compare this to "data from over 500 million Duolingo learners around the globe." I don't want to criticize these researchers in particular, as they are just doing what is expected of them in academia. But I ask of all academia, especially in the field of education: what are you doing? How is this helping? Why not just write some speculative fiction: wouldn't that at least be more interesting? Journal editors: are you being responsible when you request and select this sort of study for publication? I wish I could force the entire discipline to stop and reflect on its methods for a bit.

Nobody needs to convince me of the importance of mental health, and therefore, of the importance of mental health education in general. But this post really missed the point. Here's the main point: "Now, more than ever, a large-scale, preventive mental health approach is needed to promote student mental health and well-being." So "A mandatory mental health training would teach students easy-to-implement strategies to manage stress and solidify "mental health literacy" to improve knowledge of mental health." So many problems with this. First, throwing mandatory training to solve a problem almost never solves the problem. Second, mental health is something that all of society needs, not just students. And third, universities aren't exactly the institutions I would trust to offer mental health education. Let's put this properly where it belongs: as a health issue, to be addressed by public health care, for everyone (not just those privileged enough to pay tuition), managed by doctors, with support where needed and requested from other institutions.

If you look at the source of, say, Ben Werdmuller's blog, you'll see a script of the form 'type="application/ld+json"'. This is a Javascript Object Notation (JSON) representation of the entire blog post. The idea is to make it possible to link the post into a wider linked data network (LDN). You won't find a plugin for this just by searching WordPress (I tried; what you see is mostly SEO). But Doug Belshaw links into a video and download of just such a plugin so you can try it on your own WordPress blog. On the surface, it will look like nothing happened. But if you view your blog source, you'll see the linked data. You don't need to do anything fancy like upload it into GPT-4, but if you did, GPT-4 would find it useful. (P.S. There are various other schemes for doing the same thing - schema, open graph (og), twitter, etc).

This is a pretty good post and will help readers hone their intuitions about what a large language model (LLM) like chatGPT actually does. But I want to focus on one tiny little statement: "ChatGPT doesn't really 'know' anything. It has no self-awareness or consciousness." Whoa, whoa, whoa, wait a minute now. Since when does 'knowledge' consist of 'self-awareness' or even 'consciousness'? When we look at current and historical accounts of knowledge (such as the widely discussed definition of knowledge as 'justified true belief') the critical elements seem to be (a) an assertion that some proposition is true or false (ie., a belief), that is is true (eg. according to Tarski's theorem), and that there are grounds for our assertion (based on evidence, confirmation, or any number of other proposals). If you want to say consciousness or self-awareness are necessary, that's fine, but you also need to tell us what consciousness or self-awareness bring to the table. Via Alan Levine.