SAIL: Systems Change, Head of AI, Robots, Hype
August 6, 2025
Welcome to Sensemaking, AI, and Learning (SAIL). I focus on AI and higher education.
My primary interest in AI is in how it impacts higher education. My instinct is to focus on the systems level effects. The AI landscape is uncertain at best - we’re seeing strong indications of maladaptive use, recognizing exploitation across the entire pipeline of data access through to labeling, and realizing the economic, environmental, and likely labor market impact of AI. Essentially, you can position yourself on a belief spectrum from “critical - burn it with fire” through to “hypester with clickbait headlines”. And you’d be right. The effects of AI are at least on par with previous general purpose technologies (the OG GPT) as it will impact all aspects of society.
Universities are starting to initiate new roles that focus specifically on AI leadership and strategy: Western, Utah, George Mason, SNHU. Some universities are repurposing IT (CIO, CTO) or similar roles to broaden them to include AI. That’s a mistake. A clear multi-domain skill set is needed for a chief AI officer with technical, organizational, learning sciences/psychology, product, and innovation background. Your CIO will not get you to systems level change with AI. Neither will your provost.
AI and Learning
Against the Uncritical Adoption of 'AI' Technologies in Academia I’m reminded of early adoption of online learning. There were some reports, and the odd meta analysis, out that suggested online and classroom learning had no significant difference in terms of outcomes. This wasn’t really the argument - online and digital learning was going to arrive regardless of effectiveness - largely due to the economic interests of those advancing technologies. We’re at the same stage now with AI. But on a bigger scale. The authors argue that “universities must take their role seriously to a) counter the technology industry’s marketing hype and harm and to b) safeguard higher education critical thinking expertise academic freedom and scientific integrity”. A key challenge here is the diminishing public support, at least in USA, for everyone to get a degree.
Ethical and legal guide: AI in Education. This document won’t contain much new for people tracking current trends. But it has solid practical examples of bias or need for human-in-the-loop. It’s a reasonable document that avoids the usual bluster in anti/pro camps.
Accuracy of model outputs are a significant concern in education. OpenAI just released a report that explains why hallucinations happen. Short answer: the reward structure during training and evaluation. No “points” are given for saying “I don’t know” so the reward function favors making things up when something isn’t known. GPT5 is better, though.
OpenAI has launched a jobs and certification program. What does it mean? Here’s a solid breakdown (i.e. they want to “own the window where learning happens”)
There is a fantastic section in this paper (Section 5 personal reflections) where mathematicians assess the capabilities of GPT5 for novel (not only recall) math work. This line should likely be tattooed on roughly every academic, with mild modifications for their domain: “If students rely too heavily on AI systems that can immediately generate technically correct but shallow arguments they may lose essential opportunities to develop these fundamental skills. The danger is not only a loss of originality but also a weakening of the very process of becoming a mathematician.”
Will AI replace teachers? Usual mixed bag assessment: basically, the answer is yes and no.
Have I mentioned that we have an excellent online conference on AI in education? With amazing keynotes and panels? You should sign up. Or not. I don’t care.
AI Technology
I feel for two camps: the AI hypesters and the AI Bubble Will Bursters. It must be so frustrating to always be right on the edge of AI nirvana or destruction - to see glimpses but not see either desired state fully realized. Here’s a pro/con assessment. The concluding two paragraphs sum it up nicely.
Robots are the next wave. China is winning here. And they’re spreading their manufacturing wings with $1b Middle East initiative.
Waymo continues to expand. This is a direct area where human labor will be impacted.
Working with AI agents in education is promising and challenging. For example, if you have an agent that interacts with a learner in dialogic learning, you’ll quickly come up against memory and function issues (i.e. moving from dialogic to Q&A). To address these challenges, we’ve adopted a multi-agent architecture where different agents play different roles. A dialogic agent supports deeper engagement on a topic and an assessment agents focuses on evaluating understanding. The big issue that arises is orchestration between agents so that the learning experience doesn’t feel stilted or inauthentic. AWS (and many others) have an approach to this orchestration problem.
Speaking of memory: Memary - an open source memory layer for agents.
Agentic Design Patterns. This is good. Ranging from basic overview to what makes an AI system an agent? through to complex implementation.