SAIL: Online AI & Learning Conference (free). Also, Intentionally Play with AI Everyday
Welcome to Sensemaking, AI, and Learning (SAIL), a regular review of things related to AI and learning.
Our 4th annual "Empowering Learners for the Age of AI" is being held next week. It's online. It's free. Register here and join 1400 friends and kind souls. While you're at it, register for our in-person conference at ASU in December.
In a few recent keynotes, questions arose about the adoption rates of AI tools like ChatGPT, Claude, Bard, and Bing Chat. Adoption rates vary - K-12 surveys suggest up to 63% of teachers have used ChatGPT, but don't really detail daily use. What I've found interesting during discussions with faculty, and questions during conferences, is that daily use and adoption of AI tools remains low. This is obviously anecdotal, but the hype of AI exceeds actual use. I'm surprised. If you haven't used ChatGPT recently (or Pi), you're in for a treat. New vision capabilities (with the paid version) are shocking. It can make suggestions of how to decorate a room. Read and explain formula. Or describe a picture you upload. Similarly, if you haven't played with Pi lately, you'll be surprised at ease of use and performance. Both still have a lag when you use audio (in and out), but that's something that will be resolved. Point is: Intentionally play with AI everyday.
AI in Learning
While AI proliferates, humans aren't very good at using it.
Intelligent tutoring systems are well established areas of research. AI (of the LLM variety) throws some confusion/questions into the mix. How do traditional conversational tutoring systems overlap with LLMs? Here's one attempt to examine that intersection: "users expressed higher ratings of understanding and remembering and further perceivedthe offered support as more helpful and the conversation as coherent. Our study provides insights for a new generation of scalable CTS technologies."
This is an excellent review, lead by Melissa Bond, on a "review of reviews" regarding AI use in higher education. 18 months in the making, but as complete an analysis as you'll find.
General AI:
Want a run through of the state of AI? This is your best one-stop report of where we are. Education doesn't receive great coverage.
Kaggle also offers its 2023 AI report. A collection of essays for their community.
A survey of experts on trends in AI. It's presented as an embedded Scribd which makes it easy to overlook the 13 pages of charts. Most respondents feel that a Department of AI should regulate AI, not an international consortium.
LLMs aren't ready for prime time. A summary of this article in Nature about LLMs in psychology.
Evaluating LLMs is hard notably in the society level impact.
We're not hearing as much about hallucinations as we did earlier this year. Some progress is being made, but the issue remains: "The main cause of AI hallucinations is training data issues. With black box AI being the LLM norm, it is also not possible to have direct insight into how the model has arrived at its predictions. However, one startup believes it has found the answer to making GenAI stop tripping." I've used Iris.AI over the last few years - it's a good tool that keeps getting better. This article impacts how the company thinks making progress on the problem: "By using the knowledge graph structure, transparency and explainability can be added to AI."
LLMs (in 2023). Great presentation. Detailed and technical.
AI & Work
How will robots impact jobs? "The study seems less concerned with actual job numbers, and more with how human employees and the public feel about the inevitable increase of robotics and AI in warehouses, manufacturing facilities and other industrial settings." Amazon says robots will speed up delivery, but not replace humans. A robot wrote that sentence.