The Case for Constraints: Why AI Needs a Bounded Universe in Education

By: Steve Behrendt, PhD, and Martin Castro, PhD, Germantown School District

As Scott Galloway puts it, “AI is not going to take your job… someone who understands AI will.” I’m not sharing this to suggest anyone is lining up to replace you as a teacher. Rather, I share it to highlight that we’re in one of the most human-centered professions, and AI has now entered our world. If you understand how to use it well, it can’t replace the human side of your work. It will free you up to focus more on it. By putting AI to work on the administrative tasks that drain our time and energy, we can show up more fully for the parts of the job that matter most.

As artificial intelligence becomes more integrated into our daily workflows, whether it be planning, communicating, grading, or content creation, it’s essential to understand the underlying concepts that shape how these systems function. If we want to truly want to maximize what we have at our fingertips, it is important to know what they do…and importantly, don’t do. Knowing the differences between deterministic, probabilistic, non-deterministic, generative AI, and large language models (LLMs) is a foundational understanding for advancing your intentional use of AI. 

Large Language Models are a type of generative AI, meaning they don’t retrieve or copy existing text; rather, they synthesize new content based on patterns they’ve learned from data. LLMs are trained on massive amounts of text to predict the next character in a sentence. Unlike traditional software that follows fixed rules, LLMs rely on probability and statistical patterns (essentially using averages) to generate content or responses. For the math folks out there, they’re probabilistic systems: they don’t produce fixed outputs, they calculate the most likely next word or phrase based on what they’ve learned. Without clear parameters or guardrails, they may generate different results, even when given the same input. It’s like rolling a set of weighted dice…some outcomes are more likely than others, but you can never be totally sure what you’ll get. 

Deterministic vs. Non-Deterministic AI - In simple terms, a deterministic process always produces the same outcome from the same starting point. It’s predictable and repeatable: 2 + 2 is always 4. A non-deterministic process, by contrast, can lead to different outcomes even when the starting conditions are the same, often due to randomness, variability, or probability.

In the context of AI, deterministic models operate on fixed rules and logic, consistently delivering the same result for the same input. Non-deterministic AI systems, such as LLMs, introduce randomness into how they generate responses, relying on probabilities and patterns learned from data. This means the same prompt might produce different, yet still reasonable or contextually appropriate outputs. It’s also important to note that these systems can introduce bias based on the data they were trained on, or generate hallucinations: responses that sound confident but are factually incorrect or entirely made up. (For more detail, here’s the full piece -  Gen AI is Non-Deterministic: Why it Matters and How it Changes the Way We Work With Software).

Frequent users of AI are well aware of the challenges that come with non-deterministic, probabilistic systems. In education, where the integrity of our content and communications matters, we must be intentional in how we design and deploy AI tools and pair them with deterministic guardrails. Structured prompts, content filters, human-in-the-loop systems, custom GPTs, and well-designed task parameters help create a “bounded universe” in which AI can operate more reliably. These constraints don’t limit creativity, just as systems and processes don’t diminish our autonomy. Instead, they help us focus our energy on what matters, freeing us from overthinking in areas where strong frameworks already exist. In the same way, bounding how we use AI doesn’t restrict its potential. It steers it in the right direction, allowing us to harness the power of LLMs with integrity, as our streamlined workflows free up our time and energy for what matters most: our kids.

Join us at SLATE this December, where we’ll show you how to leverage the tools already at your fingertips–your curriculum, instructional frameworks, lesson plans, assessments and rubrics, instructional design models, handbooks, policies, standards, and more. Together, we’ll build a bounded universe for AI that streamlines your workflow and helps you focus your energy on what matters most.