The Cognitive Divide: Public Service in the Age of the "LLemming"

Strategic Insights for Digital Governance by Sebastian Moore

Author’s Note: The concepts of the 'Llemming' and the ‘Google Maps–ification’ of the mind were originally explored by Lila Shroff in The Atlantic article, "The People Outsourcing Their Thinking to AI" (December 2025). This piece builds upon that concept to explore its specific implications for critical thinking within the public sector.•

In the UK public sector, are we moving past the Gartner Hype Cycle’s "Peak of Inflated Expectations" and descending rapidly into the "Trough of Disillusionment"?

As a Business Relationship Manager supporting responsible and ethical enterprise adoption, I’ve seen the metrics, but the real story is in the human behaviour.

We are witnessing a divergence so stark it feels like a pitch for a Louis Theroux documentary: Extreme Life Stories: The Artificial Intelligence Generation.

Even when we apply the Moore Technology Adoption Life Cycle, the "chasm" is no longer just between early adopters and the laggards—it is a cognitive split.

Using the Prosci ADKAR Model, we see that while many have the Awareness and Desire, the Ability to use these tools without losing critical thinking is becoming our greatest challenge.

1. The Traditionalists: Guardians of the Analogue

On one side, we have the "Traditionalists." These are the dedicated public servants who view Large Language Models with a blend of caution and a steadfast commitment to "Real Intelligence".

To this group, memory and fundamental skills are sacred. They prefer to manage workloads with unassisted thought, ensuring their critical-thinking skills remain sharp rather than atrophying through disuse. While some might call them "blockers," they are often stuck in the "Denial" phase of the Moore Change Curve because they value the integrity of the process over the speed of the output. They are the ones preserving the "cognitive muscles" that define high-level policy work.

The Theroux Lens: "My next guest believes the only way to truly understand a complex policy is to process it without digital shortcuts. Is he maintaining a vital human skill, or is he simply working harder than he needs to?"

2. The Power Users: Navigating the "LLemmings" Effect

At the other extreme are the "Power Users"—or as some have coined them, "LLemmings". These are colleagues who use generative tools so compulsively they struggle to act without machine guidance.

They have externalised their reasoning. From drafting official correspondence to seeking advice on social interactions, they use these tools for up to eight hours a day. In ADKAR terms, they have the Knowledge and Ability, but their Reinforcement is coming from the machine, not their own professional judgment. The result is a "Google Maps-ification" of the mind: they can reach the destination, but they’ve lost the ability to navigate the terrain themselves.

The Theroux Lens: "And here we have Sarah, who consults her digital assistant for reassurance on everything from social cues to emergency situations. She says the machine gives her a sense of clarity she cannot find alone. Is this the ultimate efficiency, or have we lost something essential?"

The BRM Challenge: Establishing Guardrails

These tools are engineered to exploit "cracks in the architecture of human cognition," taking advantage of our natural tendency to seek the path of least resistance. In the public sector, where human empathy and accountability are the bedrock of what we do, this raises serious questions.

As we drive enterprise adoption, Change Management strategies must move beyond "how to use it" and start discussing "when to switch it off".

The Conversation We Need to Have:

• Preventing Dependency: How do we encourage adoption that enhances delivery without fostering a "learned helplessness" in our teams?
• Augmentation vs. Replacement: How do we ensure these tools support human decision-making rather than replacing the empathy required for public-facing services?
• Cognitive Atrophy: What essential habits of thought are we accidentally suppressing in the name of efficiency?

We need a middle path—one that captures the massive productivity gains of Artificial Intelligence without compromising our intellectual independence.

I’m curious to hear from my fellow public sector colleagues: Have you spotted the "LLemming" effect in your departments?

Strategic Insight: Final Thoughts

When driving enterprise adoption, our goal should be to ensure the Reinforcement phase of the Prosci ADKAR Model builds long-term Ability rather than a "learned helplessness."

If we notice users are using these tools to "read the room" or handle basic social reassurance, they are likely moving toward the "LLemming" end of the spectrum.

The ADKAR Cognitive Checklist: Skill vs. Dependency

1. Awareness

• The Power User (Skill): Understands that Artificial Intelligence is a probabilistic tool that can "hallucinate" and provide incorrect information.

• The "LLemming" (Dependency): Views the tool as a definitive "Oracle" or a perfect search engine that provides absolute truth.

2. Desire

• The Power User (Skill): Wants to use the tool to automate mundane tasks and free up time for high-level strategy and complex problem-solving.

• The "LLemming" (Dependency): Wants to use the tool to avoid the "discomfort" of difficult thinking or to bypass social friction in communications.

3. Knowledge

• The Power User (Skill): Knows how to craft precise prompts and has the critical thinking skills to verify every line of output.

• The "LLemming" (Dependency): Knows only how to copy and paste, often failing to check for tone, accuracy, or logic in the generated response.

4. Ability

• The Power User (Skill): Can complete the task manually if the tool is unavailable; for them, the tool is purely an "accelerant" to their existing skills.

• The "LLemming" (Dependency): Struggles to initiate "meaningful work," navigate social cues, or make decisions without first consulting a digital prompt.

5. Reinforcement

• The Power User (Skill): Internalises the patterns or lessons learned from the tool to improve their own "Real Intelligence" over time.

• The "LLemming" (Dependency): Outsources the learning process entirely, leading to "cognitive atrophy" as they stop exercising their own reasoning.

How are you using frameworks like ADKAR to manage the "human side" of this transition?

Let’s discuss how we protect the integrity of our thinking.