Across instructional coaching and education, AI tools are emerging with the benefit of being able to read and identify facial expressions that teachers and students feel during a lesson. These companies will claim their systems can detect a range of emotions with a quick glance.

But we must ask ourselves, how accurate are these models when used in real learning environments? And what do instructional coaches, ESAs, and PD leaders know before utilizing them?

A recent publication, Am I as Effective at Identifying Emotions as Artificial Intelligence?, highlights a core challenge. Even the best emotion recognition models can mislabel and misinterpret what is actually happening in a learning conversation.

Emotion data can be extremely valuable for coaching, reflection, and growth, but it can be misleading if used without the correct context. This is why platforms like Vosaic focus on responsible AI that supports, not replaces, human expertise.

Humans vs AI in Real Learning Conversations

To test the accuracy of emotion recognition AI, researchers observed a reflective learning conversation between two people. Then, they compared interpretations from the three groups: An AI system (MorphCast), peer observers watching the video without audio (Vosaic), and the participant, who self-reported emotions with full context.

The goal was to see if AI could read emotions as accurately as humans who have access to tone, dialogue, and the natural flow of conversation.

For many educators and coaches, this mirrors a familiar challenge. Coaches regularly try to read how teachers feel when discussing classroom performance, reflecting on a lesson, or navigating difficult feedback. However, misreading an emotion can change the whole coaching cycle. This research showcases both the promise and limitations of AI in these moments.

Key Finding: AI Sees More Negative Emotions Than Humans

One big takeaway is the emotional bias that occurs when AI is present. AI systems frequently detected negative emotions such as disgust, fear, anger, or sadness. Comparatively, human observers saw mainly neutral or more positive expressions.

In fact, it appears the AI system used in this comparative study was skewed about 45% negative, and several “negative” emotions were flagged by the system that was never reported by the participant.

So what is causing the disconnect?

Well, AI systems are typically trained to be highly sensitive to micro expressions; therefore, these models often mislabel concentration or reflection as stress or frustration. Humans, on the other hand, can interpret those same moments through tone, context, and situational awareness.

For instructional coaching, this matters. A teacher who appears tense or nervous might be mislabeled and given unhelpful feedback if we are solely relying on AI models. Using emotion AI without context could lead to assumptions that damage trust rather than strengthen it.

Key Finding: Humans and AI See Different Emotional Patterns

It was noticeable in how humans and AI interpret emotional transitions. Humans reported steady emotional states with very few shifts. Whereas AI and peer observers detected more transitions, sometimes multiple within a single minute.

This implies that teachers may feel internally steady even while their faces are showing small shifts in their concentration or processing levels. AI can bring patterns to the surface, but not all of them bring significant meaning or should result in an action.

There is where coaches play a huge role in feedback; they are the ones who must decide which cues matter and which simply reflect normal cognitive engagement.

How Vosaic Supports Teachers Without Mislabeling Them

This comparative study highlights a big core truth. AI is an excellent tool at noticing the details, but humans are going to be excellent at interpreting meaning. To have the biggest impact on your educators and coaching teams, AI and humans should work together to form the strongest coaching experience.

The findings also reinforce why Vosaic avoids relying on facial expression only emotion AI. The researchers found that the AI model, MorphCast, frequently misidentified emotions, especially negative ones. Nearly 45 percent of expressions were labeled as anger, fear, sadness, or disgust, even though participants themselves reported mostly positive or neutral emotional states during the same conversation. Vosaic was part of the research process, but not as an emotion-detecting agent. Instead, the platform provided the contextualized video environment the researchers needed to interpret emotional cues with greater accuracy. This supports our belief that AI should support human judgment rather than attempt to replace it.

This is the same model Vosaic follows. Our AI-feedback video tool provides the context that facial expressions alone cannot capture. Our model highlights patterns, tags important moments, and surfaces cues that spark deeper reflection. But coaches are the ones who bring their expertise, empathy, and interpretation.

Teachers deserve feedback that reflects their intentions, not assumptions made by an algorithm. They deserve support that feels human and fair. Vosaic’s model ensures that teachers stay in control of their own narrative, while coaches gain better visibility into what actually happened in the lesson.

This human-centered design is intentional. It respects the complexity of teaching and the trust required for meaningful growth. It ensures that AI serves as an additional lens rather than a judge. And it protects teachers from the very mislabeling errors that facial-only systems tend to produce.

Why This Matters for Educators and Coaching Teams

Educators and coaching teams face increasing pressure. Coaches have larger caseloads, support multiple districts, and must provide consistent feedback across diverse environments. Remote and hybrid coaching means coaches may have fewer opportunities to observe in person. Nonverbal cues are harder to read. Teachers are balancing burnout, new initiatives, and shifting instructional expectations.

In this context, emotion-aware coaching is becoming even more important. Teachers want to feel seen, understood, and supported, not analyzed.

AI can help coaches notice patterns in practice, but coaches are the ones who give those patterns meaning. When used responsibly, AI helps reduce blind spots, strengthens calibration across districts, and gives coaches more evidence to work from. It doesn’t replace the human connection that makes coaching effective.

AI Should Help Coaches, Not Replace Them

Emotion AI can be a powerful partner in instructional coaching, when used carefully and ethically. It should provide insight, not conclusions. It should spark conversation, not dictate meaning. And it should always preserve the human relationships at the heart of teaching and learning.

When AI and human expertise work together, teachers receive coaching that is more consistent, more accurate, and more supportive. Video makes AI insights meaningful. Coaches make them actionable.

Explore how Vosaic blends AI with human expertise.