Future of Learning or a Dystopian Distraction?
Exploring the *real* impact of AI avatars on human learning
Hyper-real AI-generated avatars created on platforms like Synthesia are now capable of producing so-called “digital twins” of human instructors with hyper-realistic facial expressions, lip-syncing, contextual hand gestures, and even cloned voices.
These tools are already changing the face of online education, used at increasing volume to scale corporate training, MOOCs, customer onboarding and university instruction.
This rapid adoption of AI-avatars in education raises important questions about the pedagogical implications of synthetic instructors in learning environments. Yes, amidst all of this innovation and adoption, we rarely have time to stop to think about the pedagogical impact that lies (or, perhaps, doesn’t lie) beyond the more immediate and measurable practical benefits of AI-instructors.
As this specific form of “educational technology” continues to evolve rapidly, it becomes increasingly important to examine not just what is technically possible, but what is pedagogically sound and ethically responsible in the use of synthetic instructors.
In this week’s blog post, I analyse and synthesise academic research on multimedia learning, cognitive psychology and ethical design to explore the opportunities, risks, and best practices for using AI-generated avatar instructors in education content.
The Educational Promise of AI Avatars
Let’s start by summarising the key benefits of AI-generated avatars in the education context.
1. Scalability & Efficiency
AI avatars significantly reduce the production time and cost of video content. For example, Lind (2024) found that using synthetic presenters in a training context cut development time by over 60% while maintaining message clarity and brand consistency.
2. Learner Satisfaction & Social Presence
Studies show that learners report higher satisfaction and stronger perceived learning when videos include visible instructors—whether human or synthetic (Garcia & Yousef, 2022; Sondermann & Merkt, 2022).
The presence of a visual instructor enhances social presence, especially in online environments where learners may otherwise feel disconnected. Yet these subjective measures of satisfaction do not always correlate with objective learning gains, raising questions about whether perceived social presence translates to deeper cognitive processing.
3. Localisation & Personalisation
AI avatars can be easily customised to sign, speak, or present in multiple languages, addressing accessibility needs for d/Deaf learners and global audiences. Research shows these customisations foster inclusivity and engagement when designed appropriately (Chen et al., 2024). Nevertheless, there remains the challenge of ensuring that these tools do not introduce new forms of exclusion through representation biases or technical limitations.
Digital presenters maintain consistent tone, pronunciation, and pacing—ideal for standardised content delivery across regions. This is particularly valuable for organisations needing to train large, dispersed workforces or students in multiple countries. This consistency may come at the cost of authentic cultural nuance and instructional adaptability that human instructors provide through context-responsive teaching.
The Risks & Limitations of AI Avatars in Education
From a pedagogical perspective, what risks do we face if and when we replace the human instructor with an AI equivalent?
1. Cognitive Distraction
While AI instructor presence adds social value, it can also divert attention from key visuals. In an eye-tracking study with 87 undergraduate participants, learners focused disproportionately on the avatar's face, reducing attention to instructional graphics and charts (Sondermann et al., 2024). This visual competition may impair cognitive processing in visually rich content, suggesting that the mere presence of an instructor image may sometimes hinder rather than help comprehension.
2. The Illusion of Learning
Many learners report high satisfaction and confidence when viewing presenter-based videos—but these subjective measures don't always align with actual learning outcomes.
In a controlled experiment with 124 adult learners, Ayres & Ackermans (2025) found that enhancements like emotional design and subtitles increased engagement but not test scores unless paired with deeper cognitive strategies like annotation or retrieval practice. This disconnect between engagement and learning highlights the need for instructional designs that promote active processing beyond passive viewing.
3. Trust Drop-Off with “Synthetic Disclosure”
In a key randomised controlled study of 240 corporate learners, Lind (2024) found that participants rated AI and human instructors equally if unaware of the avatar's synthetic nature. But when told partway through that the presenter was AI, trust and engagement dropped sharply—despite no change in objective learning outcomes. This finding suggests that transparency timing and framing significantly impact learner receptivity to synthetic instructors.
4. Ethical Concerns Around Deception
As avatars become increasingly indistinguishable from real people, failing to disclose their synthetic nature risks violating learner trust and institutional credibility. Transparency is not just ethical—it's pedagogically prudent. This concern extends beyond simple disclosure to questions of representation, algorithmic bias in avatar appearances, and the potential shift in how learners conceptualise educational authority when instruction comes from synthetic sources.
How to Maximise the Value of AI Avatars in Education
Based on what we know from the research, my four top tips for optimising the impact and minimising the risk of AI instructor-avatars in education go as follows:
1. Pair Avatars with Purposeful Visual Content
Avatars alone are not enough. According to Mayer's Cognitive Theory of Multimedia Learning, learners process content through dual channels (visual and verbal) with limited capacity (Mayer, 2001). Visual instructor presence that doesn't accompany instructional visuals risks underutilising learners' cognitive bandwidth.
Best practices include:
Signalling: Use arrows or vocal emphasis to highlight key ideas (Noetel et al., 2021).
Temporal/spatial contiguity: Present visuals and narration at the same time and close together on screen.
Segmenting: Break content into pause-able sections to avoid overload.
2. Design for Function, Not Just Form
Avatars should serve instructional purposes, not just visual engagement. Sondermann & Merkt (2022) found that while learners enjoyed videos with avatars more, their factual recall decreased when the avatar was present throughout, suggesting visual competition hinders deeper processing.
Use avatars strategically:
Introductions & summaries
Motivational or reflective moments
Emotionally rich content
Avoid them during content-dense segments where infographics or animations should dominate the visual field.
3. Be Transparent About Synthetic Identity
Learners deserve to know when they are being taught by AI. Research shows that when learners discover mid-way that an avatar is synthetic, it creates a breach in trust (Lind, 2024). However, if avatars are introduced openly as AI agents, learners often accept them without issue.
A simple disclosure like, "I'm your AI learning assistant, designed to guide you through this topic." can prevent ethical concerns and preserve credibility. In Europe, this level of transparency isn’t just a “should have” — it’s a requirement under the EU AI Act.
4. Evaluate Learning, Not Just Perception
Don't confuse high engagement with high effectiveness. Use A/B/C testing to compare:
Group A: AI avatar + visuals
Group B: Human presenter + visuals
Group C: Voiceover only + visuals
Measure:
Learning outcomes (pre/post tests)
Affective responses (surveys on trust, satisfaction)
Behavioural data (completion rates, pause/replay behaviour)
Only by triangulating these metrics can educators determine whether avatars are truly adding instructional value or simply aesthetic polish.
Concluding Thoughts
AI-generated avatars are quickly becoming a core component of modern education and training ecosystems. Their rise is powered by remarkable advances in realism—fluid gestures, lifelike expressions, hyper-natural voice synthesis—and the promise of scalability, personalisation, and efficiency.
Platforms like Synthesia now make it possible to generate high-quality, multilingual video content at a fraction of the time and cost of traditional production.
Yet as this technological capability accelerates, it prompts a critical shift in our guiding question—from "Can avatars teach?" to "How should we design and use avatars to support real learning?"
To ensure that AI avatars contribute positively to learning, stakeholders must go beyond surface-level appeal and prioritise cognitive alignment, ethical transparency, and rigorous evaluation.
For Learning Designers:
Use avatars purposefully, at moments where human presence enhances clarity or motivation—not throughout every scene.
Apply multimedia learning principles, including signalling, temporal contiguity, and segmenting, to reduce extraneous cognitive load and enhance knowledge transfer.
Be transparent with learners. A simple, honest disclosure about the avatar's AI nature builds trust and sets accurate expectations.
Evaluate impact using controlled A/B/C studies, not just learner satisfaction surveys. Measure actual comprehension, engagement behaviour, and emotional responses.
For AI Avatar Developers:
Embed ethical defaults into your platforms—tools for disclosing AI identity should be standard, not optional.
Provide pedagogically-informed templates and features, such as adjustable gesture realism, guided scripting, and interactive segmentation.
Deliver analytics beyond vanity metrics—offer insight into how avatar content affects learning retention, not just click-through rates.
As the field continues to evolve, several critical questions remain unanswered that should guide future research:
How do learner characteristics (digital literacy, prior knowledge, cultural background) moderate the effectiveness of AI avatars?
What are the long-term impacts of synthetic instructor exposure on learner trust and relationship-building in educational contexts?
How can we develop empirically validated design guidelines that optimize the balance between visual instructor presence and cognitive processing?
What ethical frameworks should guide institutional policies on synthetic instructor use and disclosure?
The promise of AI avatars is not in their ability to mimic human appearance, but in their potential to extend human learning capacity—to make education more accessible, more scalable, and more personalised.
But this promise can only be fulfilled if we shift our design priorities from how real AI avatars look to how well they support thinking, understanding, and ethical communication.
In the end, the best avatars won't be those that fool us into thinking they're human—but those that help us become better learners.
Happy innovating!
Phil 👋
PS: If you want to dive deep into the cutting edge of AI-powered L&D, apply for a place on my bootcamp.