What even is AI? It is essential at the outset of any discussion concerning “artificial intelligence” to examine definitions and understand the complexities involved in the foundational software that fueled the prominence of the term, principally: neural networks, machine learning, and large language models (LLMs). As a professional who has spent millions of dollars on programmatic advertising (LLMs), built predictive polling models using statistics (LLMs), and constructed generative audio and video models (LLMs)—I am too often disappointed when pundits and national leaders exaggerate the meaning of “artificial intelligence,”1 so as to render the term meaningless. The recent Department of Education guidance on AI unfortunately fits that mold.
For ten years now, LLMs have progressed at an exciting pace with rapid improvements to microchips and GPU architecture. This has fueled what the Trump Administration calls our era’s “new frontier of scientific discovery.” Truly there are some wonderful advances on the horizon and I am hopeful that the United States can remain at the forefront—the bleeding edge—of breakthroughs that make our defense systems, power-grids, and other large-scale systems more efficient and effective.
However, the narrowly tailored LLMs that excited me to start using them a decade ago are now being replaced with broadly sourced chatbots that attempt to consume the whole internet. These chatbots are LLMs that pull from millions of indexed data points that are often sourced on questionable information. This can create a form of mesa-optimization2, the situation that occurs when a learned model optimizes based on an internal optimizer. That means that decisions being made within the model are both a black box and they can rely on the wrong information or alignment to begin with.
Furthermore, all “AI” chatbots exhibit instrumental convergence. This is where the primary goals shared between the machine and human appear similar, but subgoals are not in agreement. For example, a young child may seek “relaxation” and the machine may not be able to discern why a cigarette or alcohol is a prohibited option to achieve the end. That means that every piece of information on the internet, right or wrong, is being broadly pulled into a giant chatbot gestalt with immense limitations on understanding the intended ends of its users' needs and questions.
The environment for these devices is nowhere near perfect or ready. And with hundreds of hungry start-up companies all basing their applications on one of the major flagship chatbots, they are contributing to a shared hubris that all of the world’s knowledge can be captured into a single device, bot, or algorithm. And if not now—because of technological limitations—possibly later with more improvements to chipsets and software. Not only is this world view incorrect, it is folly to completely offload the human condition to well-indexed random number generators.
The artes mechanicae—ironically, the mechanical arts—are often called the servile arts because they are produced in the service of man. Tools are very good at making things or completing a single task. It is good that office lights automatically turn off when we leave a room. “AI” can dutifully service this end. In contrast, the artes liberales are the activity of contemplation, learning, and understanding.
“An activity which is meaningful in itself…is the exact opposite of the worker’s attitude marked by concentrated exertion. One of the fundamental human experiences is the realization that the truly great and uplifting things in life come about perhaps not without our own efforts but nevertheless not through those efforts.”3
Learning is good in itself and requires no end. Though teachers may teach, education does not provide a service or “build” a student (which is one of the greatest debates of our time). It is an activity of love.
Machines built by men have very little to say about education. Not because they are slow or incorrect, quite the opposite—they are speedy and informative. Rather, because the minds of human beings are complex and ever changing while tools are complicated and structured. Education is much more than memorizing information and job preparation. A human bond exists between teacher and student that builds curiosity and wonder, which are essential to the intellectual life. Proper education constitutes the creation of the person, the curation of the citizen, and ultimately the care of the soul.
As we look into the future, this debate between artes mechanicae and liberales will fundamentally alter education and ultimately decide the future of society. Man is already suffering the consequences of of his unwieldy tools. Should the artes mechanicae win out (and education no longer be focused squarely on the soul), man will be completely subject to his creation; a reemergence of the ancient forms of slavery.
[This is an edited and abridged version from a public comment letter published here: https://responsiblete.ch/wp-content/uploads/2025/07/DOE-AI-software-will-harm-school-children.pdf]
U.S. Code Title 15, 9401, 3: The term “artificial intelligence” means a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments. Artificial intelligence systems use machine and human-based inputs to— (A) perceive real and virtual environments; (B) abstract such perceptions into models through analysis in an automated manner; and (C) use model inference to formulate options for information or action.
https://www.alignmentforum.org/posts/FkgsxrGf3QxhfLWHG/risks-from-learned-optimization-introduction
Only the Lover Sings: Art and Contemplation by Josef Pieper
Nice piece Michael! I'll be including this in the August Bazaar at Stranger Worlds, which should run in about two weeks.
Stay wonderful!
Chris.