The Human-Centered Compass: Why AI in L&D is a Design Challenge, Not a Tech Trend

A conversation with Dr. Stella Lee—AI Literacy Architect, EdTech Strategist, and Founder of Paradox Learning—on why prompt engineering is only the tip of the iceberg, the danger of automating entry-level roles, and how to build truly AI-literate organizations.

By Louise Servoin · 2026-07-01 · 10 min read

## Dr. Stella Lee — AI Literacy Architect & Principal Consultant at Paradox Learning (Canada)

Expert in digital learning strategy, adaptive systems, and AI literacy. Alumna of the University of Hertfordshire (PhD in Computer Science). Has worked with the United Nations, UNICEF, and the Asian Development Bank. Based in Calgary, Alberta, Canada.

Some learning professionals understand pedagogy. Others understand computer science and data structures. Dr. Stella Lee is the rare expert who seamlessly navigates both worlds, bringing a visual artist's eye and an academic's rigor to the high-stakes frontier of global workforce development. Her unique trajectory, from studying fine arts in Bratislava and the US to earning a PhD in Computer Science in the UK, shapes her core belief that learning technologies must prioritize human agency, intuitive design, and ethical integrity.

We sat down with Stella to discuss why L&D must look "under the hood" of AI, how treating generative tools like "summer interns" can protect quality, and why the real disruption of AI lies in redefining how we think and learn.

## The Multidisciplinary Architect

You have a highly unique background that bridges fine arts, media communications, and a PhD in Computer Science. How does this artistic and technical duality shape how you advise organizations on AI and learning?

I think of the blend of my background as a Venn diagram – I always think and work in the intersection of art/design + education + computer science.

The arts trained me to hold ambiguity, to prototype before I have certainty. Computer science trained me to ask "what is actually happening under the hood, and what are the systemic consequences?" Most AI consultants come from one side or the other. I can translate between them.

The arts also shaped how I approach problems fundamentally: as design challenges, not engineering problems with a single correct answer. A painting isn't finished because you followed a formula; it's finished when it works for its purpose and audience. I bring that same sensibility to AI strategy. I'm not looking for the "right" solution, I'm looking for the most appropriate one for this organization, this workforce, this moment. And I'm comfortable iterating toward it rather than pretending certainty I don't have. A lot of advice in this space is presented as settled when it isn't. The willingness to say "we design, test, learn, and revise" is more honest and ultimately more useful to clients.

In practice, that means I can sit with a technical team and pressure-test an AI deployment on implementation logic, then turn around and help L&D leaders see what that same system will actually feel like to a learner or a frontline worker. That translation layer is where I add the most value.

## Beyond the Hype: The AI Literacy Blueprint

Your AI Literacy Framework has been implemented globally. Why do you argue that focusing strictly on "prompt engineering" is not enough?

Prompt engineering is a reasonable starting point, but it's already becoming less critical as a standalone skill. AI models are now designed to help you craft better prompts themselves. The skill is being automated. So if that's the ceiling of your AI literacy strategy, you're building on a foundation that's actively eroding.

What organizations actually need is the ability to engage with AI at a macro level, and this is where I think the framing has to shift. AI is categorically different from previous technologies. With most tools, you define the inputs, write instructions for how you want the systems to run, and predict the outputs. With large-scale AI systems, that relationship breaks down. As models grow larger and ingest more data, their behavior becomes less predictable, not more. We don't fully control what we've built, and we need to realize that.

That means AI literacy has to include the capacity to critically analyze outputs, not just generate them. To provide meaningful oversight. To assess risk, identify failure modes, and put appropriate guardrails in place. And increasingly, to understand how to co-design with AI systems, not just use them as tools.

The analogy I use is that we've moved from operating machinery to managing a collaborator whose reasoning we can't fully audit. That requires a fundamentally different skill set than knowing how to write a good prompt.

## The Danger of Outsourcing Thinking

You often use a very practical analogy, comparing generative AI to a "summer intern". What are the risks of using AI as a quick shortcut for workforce productivity?

The summer intern analogy served a useful purpose for a while. It gave people a mental model for why you need to check AI's work, why it can be confidently wrong, and why unsupervised deployment is risky. But I think we've moved past it. A summer intern has a ceiling. These systems don't, and that changes the nature of the relationship entirely.

What I'm more focused on now is a risk we're not talking about enough: what happens to human capability when we outsource cognition at scale. There's a growing body of research on this. A 2025 study published in Societies with 666 participants found a significant negative correlation between frequent AI tool use and critical thinking ability, mediated by cognitive offloading. MIT Media Lab researchers identified what they call "cognitive debt," where people who rely on AI assistance show measurably reduced performance when the AI is removed. A recent CMU, Oxford, MIT and UCLA study found that even ten minutes of AI use can impair subsequent independent problem-solving.

This is the productivity paradox no one wants to name. Organizations are measuring AI's efficiency gains but not auditing what's quietly eroding on the other side of the ledger.

## The Truth Crisis and 'Process as Product'

In a world where LLMs can hallucinate false information with absolute confidence, how must L&D and educators adapt their evaluation metrics?

The conversation about AI and evaluation tends to start and stop at hallucination. And yes, that's a real problem. But I'd argue there's a more insidious failure mode that doesn't get enough attention: AI that is technically accurate but contextually incomplete.

These models are trained on vast datasets that skew toward dominant languages, dominant cultures, and majority contexts. The result is that they perform well in the center of the distribution and poorly at the edges, and the edges are where a lot of the people who matter most actually live. Marginalized communities, indigenous contexts, non-Western cultural frameworks, low-resource languages, edge cases in organizational practice. The AI doesn't hallucinate about these groups. It just quietly excludes them, or flattens their complexity into something that resembles the majority norm. That's harder to catch precisely because the output looks plausible. It passes a surface-level check. What's missing isn't visible.

So when we talk about how L&D and educators need to adapt their evaluation practices, I'd actually reframe the question. It's not just "is this accurate?" It's "accurate for whom, and in what context?"

In my AI Literacy Framework, I treat Critical Thinking and Sense-Making as a core domain for exactly this reason. At the foundational level it starts with recognizing that AI outputs require verification before use. But the competency has to develop toward something more systematic, such as interrogating citations, tracking claims to original sources, weighing outputs against known evidence, and asking whose knowledge and whose context is actually represented. That critical lens has to be integrated consistently into design practice, not treated as an occasional check.

And critically, this can't remain an individual habit. The question is whether the team, the workflow, and the culture are built to do it reliably. That's an organizational design challenge, not just a skills gap.

## Rethinking the Learning Ecosystem

You have been a strong critic of the traditional, "one-size-fits-all" LMS mindset. What is your vision for corporate learning in an AI-driven era?

The idea of an intelligent tutor, one that can deliver personalized instruction without a human teacher, is not new. It dates back to 1924 when Sidney Pressey at Ohio State University built a mechanical teaching machine designed to do exactly that. Personalized learning at scale has always been the Holy Grail, and every generation of technology has promised to finally deliver it. What's different now isn't just the sophistication of the technology. It's the scale, the speed, and the degree of personalization AI makes possible. But the more important question is whether we're designing it wisely.

My vision for corporate learning in an AI-driven era is a layered architecture where AI and humans each do what they do best, and L&D professionals design the intentional structure that connects both.

The first layer is where AI genuinely excels: personalized knowledge delivery, adaptive practice, immediate feedback, pacing that responds to the individual learner. The second layer is where humans and peers become irreplaceable: sense-making, debate, contextual judgment, and critically, interrogating AI outputs together. Collective sense-making is a necessary check on AI-generated content that no individual learner should be left to do alone. The third layer is L&D's core strategic role: designing the architecture that holds both layers together, ensuring neither cannibalizes the other.

Looking further ahead, I think AI will become far more embedded in the learning experience than a screen-based interaction. We're already seeing the early edges of this with wearable computing, for example, haptic feedback for skills training, biometric-informed pacing that responds to cognitive load and attention, and augmented reality overlays for performance support in the flow of work. The boundary between learning and doing will continue to blur.

## The Golden Thread

We often see a gap between what technology can do and what professionals actually adopt. How do we ensure AI becomes a 'co-pilot' that elevates human judgment rather than a tool that encourages intellectual passivity?

The adoption gap is real, but I think it's misdiagnosed most of the time. It gets framed as a resistance problem (e.g. professionals are slow to change, risk-averse, set in their ways), but I don't think that's what's actually happening.

AI is a genuinely confusing technology. It cuts across every job function and workflow simultaneously. It brings real opportunity but also significant risk. It's moving so fast that even people who are paying close attention feel like they're chasing a moving target. And critically, many AI implementations fail not because of the technology but because of a misalignment between what the technology can do and what the organization actually needs.

At the individual level, the sentiment I hear consistently from working professionals is that AI is being pushed onto them without a clear why. No solid use cases, no meaningful support, just pressure to use it. A perfect example is Microsoft 365 Copilot. Most large organizations already have it, and it gets switched on as part of an existing license. No change management, no workflow integration guidance, no real strategy.

Just because AI can automate something doesn't mean it should. The design question should never be "what can AI do?" It should be "what should AI do in this context, given what we know about how people learn, grow, and maintain capability over time?" And this is where I think L&D has a genuine and underutilized strategic role. We are the function that understands the human side of capability change, not just the technology, but the motivation, the identity threat, the skill atrophy that happens without deliberate practice, the organizational conditions that either support or undermine learning.

The opportunity in front of us is to design what co-working and co-creating with AI actually looks like in practice, and building the intentional architecture that keeps human judgment at the center.

Tags: AI Literacy, EdTech, L&D Strategy, Paradox Learning, Future of Work, Digital Ethics, Workforce Development