Stephen Downes

My first question on reading the title was whether this is the same concept as I have been talking about over the years? It seems to be. "To both companies (OpenAI and Google), personal AI is a personalized service—from them. It's not something you own and control. It's not about individual empowerment and agency," writes Doc Searls. "For individual empowerment and scale to happen, we need to be self-sovereign and independent. Personal AI can give that to us... by working as agents that represent us as human beings—rather than mere users." Let's call it, um, a Personal Living Environment. A PLE - for the AI age. P.S. if you read to the bottom you find that Searls's post is mostly an advertisement for Kwaai, which kind of makes me sign with disappointment.

I like the idea of a 'nutrition label' for publications, as we see described here about the Public Knowledge Project (PKP) example. However, I think there is room for debate about what is relevant, and what is not, in such a label. There's information about the reviewers, editorial and review process, and funding. But are the articles cited? Are they relevant, influential, or factual? What's the percentage of articles that had to be retracted? Does the journal publish replications of previous studies? Does it allow discussion? Are the papers used in courses? We need to start seeing these resources from a public, academic and scientific perspective, and not merely from a publication perspective.

"AI is going to be an enterprise transformation technology, just like digital," writes Josh Bersin. In this post he described a company called ServiceNow that pulls the pieces together to address a single problem: "Every role I have ever experienced in business, and I've been around for a long time, has to do with somebody trying to do aspiring work and finding themselves stymied because they can't find the tools or the information they need," he says. However, the transition is challenged by the cost of such systems and the complexity of implementing them. "You're now buying another complex enterprise platform on top of your existing complex enterprise platforms."

You've probably already read AI-generated articles without knowing it. This article reports on an investigation into AdVon Commerce, the AI contractor at the heart of scandals at USA Today and Sports Illustrated. Author Maggie Harrison Dupré outlines the company's strategy of using AI to author review articles to be published on high profile websites. The idea is that the articles contain affiliate links that generate revenue if a reader clicks through and buys something. The articles are sometimes attributed to AI-generated authors whos profiles are reused over and over. This is a lengthy and detailed article that presents a lot of evidence along with the back-and-forth between the author and AdVon as the company denies many of the allegations. For my own part, I see no problem with AI-generated content if it's accurate, well-written, and not intended to deceive. But as the publishing industry spirals downward we see more and more the influence of bad actors filling the web with dishonest pollution. Via TWIG.

What I like about this course is that I was learning things only two clicks in (specifically, I was manipulating 3D models of proteins). There's no login (unless I want to mark segments as complete) and the course is fully open access. AlphaFold, by the way, is a tool "for advanced modelling, such as predicting protein-protein interactions, modelling large protein complexes and alternative structural states." The course is offered by EMBL-EBI Training on the ELIXIR Training Platform.(see more). There's a lot of stuff here; it can all be overwhelming, but this course will guide you through the resources they offer.

This is a tool that allows you to search for relevant topics from among the 1763 papers presented at the ACM CHI Conference on Human Factors in Computing Systems (CHI 2024) conference. It returns a similarity graph plotting the most relevant matches to your search. You can filter for specific paper tracks and, interact with the scatterplot to explore the different papers. "The system works by computing sentence embeddings on the papers text, then applying dimensionality reduction on the embeddings to compute the scatterplot. You can even change the dimensionality reduction and embeddings algorithm and parameters." You may also enjoy the way the source code for this is presented. It calculates "CHI 2024 papers similarity using sentence embeddings via HuggingFace Transformer.js and dimensionality reduction methods using Druid.js."