Stephen Downes

I made a comment during a talk earlier this week (slides, audio coming soon) that we cannot become ethical people merely by avoiding risk. That's why I challenge risk-based definitions of ethics in AI research and practice. We have to hope for and work toward something better. That's at the core of a lot of punk-based rebellion, and I will confess, if it has 'punk' in the name, I'm probably on board, because it implies building and doing and making for yourself. With that preamble (emphasis on the 'amble') I introduce this paper, which tries to do better: "cultivating hopepunk and solarpunk attitudes within the field of higher education and educational technology, as well as rewidening and rewilding higher education using utopian imagination, the article (20 page PDF) points towards more hopeful, preferable futures for the people and the planet."

There are two stories here, the main story, and the story I think is written between the lines. The main story is that the small publisher is shutting down because the deluge of AI-generated content submitted is too much to wade through. Also, the content is poorly written: "It is soulless. There is no personality to it. There is no voice. Read a bunch of dialogue in an AI generated story and all the dialogue reads the same." Between the lines, though, is the likelihood that it's taking more and more time to distinguish between the AI-written and human-written content. And then what? How does a publisher decide what's worth publishing? Via Paul R. Pival.

This paper describes "MetaMate, an open-access web-based tool leveraging large language models (LLMs) for automated data extraction in educational systematic reviews and meta-analyses." While some may reasonable express scepticism about the use of AI to summarize papers for metastudies, I think it will be a positive to spend more time reading the papers and less time searching for them. Anyhow, the paper evaluated the software's precision and finds it compares well with human coders in most areas, though it's weaker in subjects such as Delivery Mode (81.25%), Intervention Duration (87.5%), and Academic Subject (87.88%).

As usual, we could replace the term 'journalism' with 'education' and arrive at pretty much the same conclusion. "Twitch is a major player among live video platforms with 1.6 billion hours of content produced monthly, much of it by users age 25-34. That content is mostly livestreamed gameplay, but the app is an increasingly common distributor of news and information." Here's a livestream we did just last week (I livestream using YouTube instead of Twitch). Here's a playlist of 65 videos where I livestream my own experiences learning new software. More and more, I think, we'll just livestream learning experiences directly, and more and more, they won't be in the classroom. (p.s. I often livestream my gaming sessions as well).

This article effectively draws a link between what it means to learn a discipline and the concept of epistemic justice. "In mastering a discipline, learners need to master the 'underlying game' associated with disciplinary epistemes reflecting ways of thinking and practicing within a particular discipline." Quite right. Consequently, "If these disciplinary epistemes are based on epistemic hegemonies from the global north, then they are potentially exclusionary by definition and will ensure that certain learners either never grasp the 'underlying game' or have significant difficulty in doing so." To oversimplify (only a bit), there are two approaches. One is to change the learner. That's colonization (of the person from the South by the values of the North). The other is to change the game. That's decolonization. The challenge, though, lies in how to decolonize a discipline without undermining its factual basis, utility or relevance. The discipline may, for example, devalue "knowledge that is derived from everyday experiences or common-sense ways of thinking," but it may do so for a very good reason. Image: Loring.

This is an interesting question. Companies collect information about you, and you can demand to see what they've collected and correct it if the information is wrong. But what about what a company believes about you? For example, the data might be 'Stephen missed a payment on May 18', while the corresponding belief might be 'Stephen is a bad credit risk'. My rights with respect to the first seem to be different than my rights with respect to the second. And to the extent that these beliefs are increasing generated by AI, we are developing a new class of information - 'generated information' - what we might not even be allowed to see, much less correct. Image: ChatGPT 4, "a line drawing of an AI with 'opinions'".