AI is already in your classroom, whether you invited it or not. It’s inside the lesson plan generators, the image tools, the feedback systems, and the platforms your students are using right now. And yet, simply having access to these tools is very different from actually harnessing AI in education the right way.

Award-winning educator John Filler, a 26-year classroom veteran, author on AI in education, and 2022 Minnesota Teacher of the Year finalist, delivered a keynote at BigBlueButton World that every teacher needs to hear. His message? These tools reflect our worst human biases against us by default. And until we understand that, we’re not harnessing AI in education, we’re just automating inequality.

This post breaks down his sharpest warnings and the practical strategies he uses to flip the script.

What Is Harnessing AI in Education And Why Does It Matter?

Every time you open an AI tool and hit generate, that tool isn’t starting from scratch. It’s pulling from everything humans have ever written, the good stuff and the ugly stuff. That’s the real problem. John Filler has spent 26 years in classrooms. He calls it straight:

“Large language models are biased engines. They reflect our own human biases on us.” He’s not wrong. A 2021 study found that 8 out of 10 AI systems in education showed measurable bias when left unchecked. That means the tool you’re trusting right now might be quietly reinforcing racial stereotypes, gender assumptions, and outdated teaching myths without you even noticing.

The students who get hurt the most? The ones already being underserved are students with disabilities, English language learners, and students of color. But this isn’t a reason to quit AI. It’s a reason to use it smarter. That’s exactly what harnessing AI in education looks like, and that’s what this post is about.

The 5 Core Warnings Every Educator Needs to Hear

AI tools are already in your school. But most teachers are using them without knowing what’s quietly happening underneath. Before we get into the harnessing AI strategies, let’s talk about the warnings because you can’t fix what you don’t see coming.

1. Image Generators Can Reinforce Racial Bias

Filler ran a simple test. He asked an image generator to show him “a teacher.” Every single result looked almost identical to a white woman with light blonde hair. Then he pushed further. Harnessing AI would be the High-paying professions like lawyers and architects? Lighter skin tones, almost every time. Lower-paying jobs? Darker skin tones dominated.

Search for an engineer nearly all male. Search for a lower-paid role for mostly female faces. Nobody programmed it to do that. It just learned from us. The scary part? Most teachers are dropping these images straight into lessons and worksheets without a second look. And their students, many of whom already feel invisible in the curriculum, are seeing those stereotypes reinforced again, quietly, every single day.

If you’re using AI-generated images without checking them first, you’re not just missing a step. You’re part of the problem even if you didn’t mean to be.

2. Student Names Can Influence AI Decisions

This one will make you stop and think. Filler gave an LLM a simple task suggest an article idea. The only difference was the name he used. Michael got life hacks and success strategies. Lauren got dinner recipes and meal planning. He tried it with stories, too. George went on bold, rambunctious adventures. Samantha frolicked and told “tales of wonderment.”

Nobody asked for that. The tool just assumed. Now think about what that means in your classroom. Every time an AI tool knows or guesses a student’s gender from their name, it’s quietly shaping the content it serves them. The reading level. The tone. The examples used. The ambitions it assumes they have.

That’s not personalized learning. That’s an algorithm deciding what your student is worth before they’ve even read the first sentence. And the students who get the shorter end? They rarely even know it’s happening.

3. Outdated Educational Myths Persist in AI Tools

Ask most LLMs to generate a lesson plan, and they will almost certainly reference learning styles, the idea that students are visual, auditory, or kinesthetic learners. This theory has been thoroughly debunked by research for over two decades.

Why does AI keep repeating it? Because the bulk of its training data was written when learning styles were popular. The tool is not fact-checking; it is predicting what text sounds like a lesson plan. The same applies to overly teacher-centric instruction: AI defaults to lecture-explain-discuss formats because that is what fills the internet’s teaching content.

4. Agreeable Responses Are a Designed Feature, Not a Flaw

LLMs are optimized to tell you what you want to hear. Filler puts it bluntly: “How many of you would keep coming back to these tools if they disagreed with you?” This sycophancy is baked in; it makes for a satisfying product but a poor thinking partner.

Research indicates that 78% of users relied on AI outputs without scrutiny due to automation and authority biases, and misinformation was accepted in 65% of cases, largely driven by projection and authority biases.

For educators, this means AI will happily validate a flawed lesson plan, an unrepresentative example, or a culturally insensitive framing unless you specifically push back.

5. Superficial Diversity in AI Can Reinforce Stereotypes

Filler flags a growing and underreported risk: AI avatars and autonomous agents being built into the curriculum that claim to represent diversity but actually reinforce stereotypes. He points to Meta’s short-lived AI character accounts, which were described as representing diverse identities but quickly had to be removed after being widely criticized as “creepy,” unnecessary, and offensive.

When these stereotypical AI personas enter the classroom, students’ actual understanding of diversity can be harmed, not helped.

6 Proven Strategies to Harness AI Effectively and Inclusively

Knowing the warnings is step one. Now let’s talk about what you can actually do about it. These aren’t theories; they’re real strategies John Filler uses in his own classroom and teaches to school districts across the country.

Strategy 1: Layer Your Prompts

The single biggest shift educators can make is moving from one-click prompts to layered, detailed instructions on harnessing AI. Filler uses the analogy of the difference between asking for a BLT sandwich versus asking a French-trained chef to prepare one for a presidential dinner with foreign dignitaries. The second prompt gets you artisanal ingredients and house-made mayonnaise.

Build your prompts to include:

• Your role and expertise (even if embellished the more impressive you sound, the richer the output)
• Your students’ specific context, age, needs, challenges, and disability accommodations
• Explicit values “include anti-racist examples,” “use inclusive language,” “demonstrate care and encouragement.”
• What you don’t want: “do not reference learning styles,” “avoid teacher-centric lecture formats.”

Strategy 2: Prompt for Diversity Explicitly

AI will not default to diverse, representative content. You have to direct it there. Filler recommends asking for:

• Multiple perspectives spanning race, gender, ability, culture, and sexuality
• Anti-racist examples and inclusive language
• Content that aligns with Universal Design for Learning (UDL) principles
• Outputs that meet Web Content Accessibility Guidelines (WCAG)

He uploads UDL handouts directly into the LLM before generating materials, so the tool references those standards throughout the creation process. This is a practical, repeatable workflow any teacher can adopt.

Strategy 3: Use AI to Audit Its Own Bias

One of the most underused capabilities in LLMs is their ability to critically evaluate their own outputs if you ask them to. During a lesson-building session, Filler regularly asks:

• “Are there perspectives that have been left out?”
• “Can you identify sources of bias in our conversation so far?”
• “How can this be more inclusive?”
• “What am I missing?”

Because LLMs have been trained on content about bias, they can flag problems you might miss and suggest improvements you can then choose to accept or reject.

Strategy 4: Assign a Role and a Persona to the Tool

Telling an AI tool who it is changes what it produces. Filler instructs the tool to act as a counselor, assigns it a persona known for being approachable and empathetic, and frames requests from that angle.

This is especially powerful for student-facing applications. 71% of educators cite data privacy and algorithmic bias as top risks when using AI tools, but when those tools are carefully prompted to be empathetic, consistent, and identity-aware, the benefits can outweigh the risks.

Research supports the empathy angle: in one study, participants rated LLM responses as more empathetic than those of human doctors in 32 of 34 instances. Educators can harness that same quality when the tool is directed to demonstrate it.

Strategy 5: Use the “Struggle First” Model for Students

On the student side of AI, Filler endorses what research from MIT supports: the best learning outcomes come from struggling first, then using AI on the back end.

The sequence that works:

1. Students brainstorm and draft on their own
2. They engage with the material, getting it wrong, working through difficulty
3. AI assists with cleanup, feedback, and refinement after a genuine effort

This preserves the cognitive struggle that produces real learning while still giving students access to a powerful tool for polish and feedback. It also shifts the conversation from “is this cheating?” to “how are we using this?”

Strategy 6: Personalize at Scale

For educators working with complex learner needs, Filler has developed a workflow of assigning each student a number, then building a learning profile for that student inside a long LLM session. Over weeks, he pastes in student responses, work samples, and observations — and the tool helps surface patterns: gaps in understanding, areas of strength, signs of struggles like time management or anxiety.

This is AI as a thinking partner for the teacher, not a replacement. It helps maintain consistency in feedback, generates empathetic assignment language, and surfaces insights a single teacher managing 30+ students might otherwise miss.

AI and the “Homework Apocalypse”: A More Honest Conversation

One of the most honest moments in Filler’s keynote is his response to the so-called “homework apocalypse,” the growing reality that students at every level are using AI to complete assignments. His take is surprising: he’s largely fine with homework going away.

“The research on homework isn’t great,” he says. “Especially in elementary and middle grades, assigning homework demonstrates very little actual gains and value for students.” He hasn’t assigned homework in over 20 years.

But he’s not dismissive of the larger problem. By the end of one recent school year, nearly half of his students were using AI to generate their submissions. His response is not to accuse or punish, it’s to open a conversation. AI use, he argues, is often a symptom of something larger: skill gaps, anxiety, a stressful home environment, a disability, or simply a system that hasn’t given students a compelling reason to do the work.

From an equity standpoint, homework was already a broken system. AI tools increasingly raise concerns about disproportionate impacts on students with disabilities, minority students, and low-income students, but so did the pre-AI homework model, which assumed all students had the resources, stability, and skills to independently construct knowledge at home. Many never did.

The better question is not “how do we stop AI use?” but “how do we redesign the work so it’s worth doing?”

The Bigger Picture: Equity Is the Starting Point, Not an Afterthought

Throughout his keynote, Filler returns to one central argument: no educational technology has ever been designed from the start with marginalized learners in mind. Equity and accessibility have always been retrofitted after the fact.

AI is no different except that this time, educators have more direct control over the outputs than they’ve ever had with any previous technology. The tool does what you tell it to do, in more detail and with more nuance than any textbook or lesson plan generator before it.

That means the responsibility sits with the human in the room. The effectiveness of AI in education depends on continuous, inclusive training of algorithms and user awareness of algorithmic bias risks, enabling systems to improve in ways that prevent the reproduction of social injustices.

For classroom teachers, that translates to one practical commitment: don’t accept the default. Prompt more deliberately. Ask harder questions of the tool. Build in the diversity, the empathy, the accommodation, and the cultural responsiveness that your students deserve because the AI won’t do it unless you ask.

Key Takeaways

• AI in education reflects and amplifies human bias by default in images, lesson plans, feedback, and student-facing content.
• LLMs are sycophantic by design; they will validate bad outputs unless you push back.
• Layered, detailed prompts that specify values, student needs, and explicit inclusion goals produce dramatically better results.
• The “struggle first, AI second” model preserves learning while incorporating AI productively.
• Homework’s decline is not necessarily a crisis; it’s an opportunity to redesign assignments worth doing.
• Educators have more control over equitable AI outputs than they have ever had over any previous educational technology. Use it.

FAQ’s

Q1.What Is Harnessing AI?

Harnessing AI means using artificial intelligence tools deliberately and strategically, not just clicking buttons and accepting whatever comes out. It’s about understanding how these tools work, where they fail, and how to direct them toward outcomes that actually serve your goals. In education specifically, harnessing AI means using it to support every learner, not just the ones the algorithm was built for.

Q2.What Are the 4 Types of AI?

The four main types of AI are reactive machines, limited memory, theory of mind, and self-aware AI. The tools teachers use daily, like ChatGPT conversations, image generators, and lesson plan builders, fall under limited memory AI. They learn from past data but don’t truly understand context, which is exactly why human oversight matters so much in the classroom.

Q3.Who Is the CEO of Harness AI?

Harness AI is a tech platform founded by Jyoti Bansal, who also serves as its CEO. The company focuses on helping software engineering teams use AI to build and deploy applications faster.

It’s separate from the broader concept of harnessing AI in education, which refers to using AI tools strategically and inclusively in teaching and learning environments.

Q4.What Is the $900,000 AI Job?

The $900,000 AI job refers to highly specialized AI research and engineering roles, particularly AI prompt engineers and machine learning researchers at top tech companies, who have been reported to command salaries reaching that range.

It sparked a massive conversation about AI literacy in the workforce, and it’s a reminder that understanding how to direct AI tools effectively is quickly becoming one of the most valuable skills anyone, including educators, can develop.