Stephen Downes

Donald Clark has already gotten a couple posts out of Google's EdAI announcements. This paper (86 page PDF) presents Google's efforts to "translate high level principles from learning science into a pragmatic set of seven diverse educational benchmarks" and to "develop a new set of fine-tuning datasets to improve the pedagogical capabilities of Gemini." To develop, in other words, a generative AI tutor. It deserves a careful reading, and your impressions of the 35 page main text and 50 pages of supplementary material may vary from mine. While the paper is very up-to-date with respect to AI, it reveals (to me) a dated and psychology-heavy understanding of 'learning science' that seems limited to work on intelligent tutoring systems - but you can view their 'Pedagogy rubrics' in section 4.3.1. Saying, "the gen AI models that power most of the latest EdTech systems are not explicitly optimised for pedagogy" the authors state that they "focus on conversational tutoring because we believe that it is one of the most impactful and general use cases." This forms the basis for their LearnLM-Tutor, introduced in this paper. There's a ton of information in this paper, but a lot that is not said - the model was "trained on an offline dataset," whatever that means (they say they tried several), and no stats are reported.

This is another effort to find some human skill or capacity that is at least different, if not better, than AI. "There are compelling reasons why human curiosity is needed more than ever in the age of AI," writes Anne-Laure Le Cunff, "and they stem from the fundamental differences between human and AI curiosity." The post offers a framework that identifies three key aspects of curiosity - processing, perspective, purpose - and tries to show how different they are for humans as compared to AI (not that people would really think of AI as 'curious' in the first place). I don't think it works because the descriptions are too surface-level. How do we know 'hunches and instincts to produce serendipitous discoveries' are actually different from 'processes vast amounts of data to uncover patterns'? The same mechanism could produce both results.

This preprint is tantalizing but far too short for what the title promises. Basically, the idea is to use generative AI to automate the process of giving multi-modal feedback on assignments. The instructor works with the AI, which may offer correction (for example, recommending the instructor focus on effort rather than outcome). And what's key is the idea of offering feedback in multiple modes, such as 'talking heads', rather than just text. I can't comment on the implementation because we don't really get anything in the way of details (this paper is a classic example of why we shouldn't simply count publications, because things like this can be pumped out in a day; we should actually consider what the publications contribute to the literature).

Just a note on what the reform crowd is up to. The Mastery Transcript Consortium "provides a mastery transcript that replaces traditional high school transcripts. Instead of a list of courses and grades, learners share competencies with colleges and employers." ETS (formerly Educational Testing Service) companies "deliver learning and assessment solutions around the world." The real big deal in this space will be when AI performs skills and competencies assessments based on real-world data and nobody is needed in the middle to own the process.

"We're experiencing an epochal shift in technology, on the order of magnitude of the Internet's commercialization," writes Eric Goldman (19 page PDF), "but this time, the regulators are intervening early, in a massive and unrelenting way." Hence the prediction in the title. Regulators will misunderstand the technology while seeing to control and censor the outputs. It doesn't help that incumbent technology companies are seeking to preserve their advantage. "Misdirected or malicious Generative AI regulations jeopardize all kinds of algorithmic activities, from personalized content to algorithmically sorted search results - things that we rely upon many times a day." Related: audio interview with Goldman about the talk.

During the playoffs I like to keep track of the scores, so I do a search. P{roblem is, there are four games, but Google only displays three results. It's so frustrating. I have to scroll way down to get a link to the actual website that displays all the scores. If Google has its way, I might never get to that website at all. As Kottke summarizes, "it's better/cheaper to provide potentially wrong answers to keep you clicking within Google than it is to send you away for the right answers." And an answer doesn't have to be incorrect to be wrong. It just has to be something I don't want. Like incomplete scores. Or summaries instead of sources.