Three people walk out of the same AI training session. One is excited. One is anxious. One is quietly planning how to minimize their exposure to it.
They sat through identical content. They passed the same assessment. HR marks them all as "AI-ready." Then implementation starts, and everything falls apart.
The Problem Isn't Knowledge - It's Perception
Through studying how professionals actually think about AI, we noticed a persistent gap: AI-expertise research often examines one organizational level at a time. Far less work studies what happens when different parts of the organization think about AI in fundamentally different ways.
But that's exactly where implementations fail. Not from technical problems. From invisible conceptual divergence.
Parker & Grote (2022) found that work design - a human factor - matters more than technology capability for AI outcomes. What is work design if not a shared understanding of what AI should do? When that understanding fragments, so does implementation.
💡 The divergence problem:
Ask five people in the same room how they think about AI's role. You'll get five different mental models: one sees a tool to master, another a partner to collaborate with, someone else an encroachment on their professional territory, another a crutch they've become dependent on, and the manager defers to it as an authority whose recommendations shouldn't be questioned.
These aren't opinions. They're operating systems - and they're usually invisible.
What Divergence Looks Like: Four Real Patterns
Pattern 1: The Leadership Gap
Leadership sees: AI as a strategic tool or collaborative partner. Controllable, measurable, positive ROI (Davenport & Ronanki, 2018).
Frontline experiences: AI as an authority they defer to without thinking (Nagendran et al., 2020), a crutch eroding their confidence (Rinta-Kahila et al., 2023), or an encroachment on professional territory (Mirbabaie et al., 2021).
What happens: Leadership celebrates productivity gains. The team feels capability eroding. When leadership says "use AI more," the team hears five completely different things:
- "Become more dependent on something I don't trust"
- "My expertise is being automated away"
- "Stop thinking for yourself"
- "Let the machine do your job"
- "Accelerate your own obsolescence"
Same words. Completely different meanings. No amount of communication bridges this gap until you acknowledge the perceptual divergence exists.
Pattern 2: The Cross-Functional Clash
Data scientists see AI as a collaborative partner - something to onboard and calibrate (Cai et al., 2019). Operations sees it as a replacement strategy - which tasks can we automate? Legal sees it as an encroachment requiring jurisdictional defense. Product defers to it as an authority that just needs to deliver recommendations.
The meeting: Data science proposes expanding AI use in a new domain. Operations is enthusiastic about efficiency. Legal raises concerns about accountability. Product wants to ship fast.
What leadership hears: Normal cross-functional tension.
What's actually happening: Four incompatible mental models about AI's role colliding. The conflict feels political or personal - it's actually conceptual. But because it's invisible, it escalates instead of resolving.
Why traditional alignment strategies fail:
You can't align on implementation when you're operating from different conceptual frameworks. One team's "strategic partnership" is another team's "expertise erosion." Training on best practices doesn't resolve this - it talks past it.
Pattern 3: The Expert-Novice Split
Junior staff love AI. It accelerates their learning - Brynjolfsson et al. (2023) found novice customer support agents improved 34% with AI support. They see AI as a scaffold helping them reach expert performance faster.
Senior staff are more cautious. They've seen automation erode expertise before. They worry about juniors developing AI-dependent proficiency instead of independent capability.
The tension: Juniors think seniors are resistant to change. Seniors think juniors are building dependency. Both are partially right.
The actual problem: Nobody designed the scaffolding to come down. Permanent AI support creates what Dreyfus (2004) warned against - it prevents the embodied engagement that develops expert performance. But the organization interprets the disagreement as generational friction, not a design flaw.
Pattern 4: The Silent Schism
This is the most dangerous pattern: when divergence is completely invisible because nobody voices it.
Half the team has quietly entered delegator mode - override rates have dropped to near zero. They defer to AI recommendations without verification. Choudhury & Shamszare (2024) found this trust is self-reinforcing regardless of accuracy.
The other half is strategically minimizing AI exposure - Kronblad & Jensen (2023) documented how professionals redefine expertise around domains where AI can't compete. They're not resisting overtly; they're quietly carving out AI-free zones.
Leadership sees: Consistent adoption metrics. Everyone passed training.
Reality: Two completely different relationships with AI. One group is deskilling through over-reliance. The other is missing beneficial applications through defensive avoidance. Neither pattern is visible in standard metrics.
⚠️ The cascade effect:
Individual divergence cascades to team dysfunction, which cascades to organizational brittleness, which cascades to professional identity crisis. This is why single-level analysis can miss the actual failure mode.
Making Divergence Visible: The First Step
You can't manage what you can't see. Most organizations are flying blind on team-level mindset divergence because they've never made it visible.
The standard approach - assess technical AI knowledge - tells you nothing about perception. Two people with identical technical understanding will implement AI completely differently if they hold different mental models.
Our research identifies 6 distinct mindset profiles based on patterns in how people actually think. The Scotoma team assessment makes these visible:
- Each team member takes the 25-30 minute Scotoma Assessment
- You see the distribution - which mindsets dominate?
- Where's the leadership-team gap?
- Which orientation gaps need discussion? For example, leadership emphasizing Craftsperson control while the team experiences Naturalist-level integration
- What specific blind spots does your mix create?
Visibility doesn't solve the problem. But it's the prerequisite for solving it.
What Alignment Actually Looks Like
Alignment doesn't mean everyone thinks the same way. Different mindsets can coexist productively - if they're visible and explicitly managed.
It's appropriate for your data analysts to see AI as a partner (Cai et al., 2019) while your legal team sees it as a tool with defined boundaries. The problem is when:
- The difference is invisible
- Leadership assumes uniformity
- Training and policies are one-size-fits-all
- Metrics track adoption but not divergence
- Conflicts get interpreted as personal/political rather than conceptual
Healthy alignment practices:
1. Surface mindsets explicitly
Use team assessment to make mental models visible before conflict erupts
2. Design for divergence
Different roles may appropriately have different relationships with AI. Make that explicit.
3. Tailor interventions
Naturalist pattern → make invisible AI touchpoints visible. Sentinel pattern → clarify professional boundaries. Craftsperson pattern → watch for underestimated cognitive shift.
4. Track mindset-specific risks
Don't just measure adoption. Measure the blind spots your mix creates.
5. Create forums for conceptual alignment
Regular discussions focused not on "how to use AI" but "what role should AI play."
The Real Cost of Invisible Divergence
Teams waste months in implementation friction that gets blamed on "change resistance" or "communication problems." The real issue - incompatible mental models - never gets addressed because it never gets named.
Parasuraman & Manzey (2010) showed automation complacency develops slowly and reverses with difficulty. By the time leadership realizes half the team has entered delegator mode, undoing it requires structured intervention - not just better communication.
Most organizations are having the wrong conversation about AI. They debate tools, vendors, ROI, and training curricula - all downstream of the actual constraint. The invisible conceptual divergence is what sinks projects. The technical challenges are solvable once you align on what problem you're solving for.
See how your team thinks about AI
Free team assessment. Individual results + team distribution. Make the invisible visible.
The question isn't whether your team has divergent mindsets - every team does. The question is whether you can see them, talk about them, and design for them.
Make it visible. Then you can manage it.
References
Brynjolfsson, E., Li, D., & Raymond, L. R. (2023). Generative AI at work. NBER Working Paper No. 31161.
Cai, C. J. et al. (2019). Hello AI: Uncovering the onboarding needs of medical practitioners for human-AI collaborative decision-making. Proceedings of the ACM on Human-Computer Interaction, 3(CSCW), Article 104.
Choudhury, P., & Shamszare, M. (2024). Trust in ChatGPT: Self-reinforcing patterns in AI adoption. Harvard Business School Working Paper.
Davenport, T. H., & Ronanki, R. (2018). Artificial intelligence for the real world. Harvard Business Review, 96(1), 108-116.
Dreyfus, S. E. (2004). The five-stage model of adult skill acquisition. Bulletin of Science, Technology & Society, 24(3), 177-181.
Kronblad, C., & Jensen, T. (2023). Being and acting professional in the age of AI. Journal of Professions and Organization, 10(2), 158-176.
Mirbabaie, M. et al. (2021). The impact of AI on professional identity threat. Information Systems Research, 32(4), 1181-1200.
Nagendran, M. et al. (2020). Artificial intelligence versus clinicians: Systematic review of design, reporting standards, and claims of deep learning studies. BMJ, 368, m689.
Parasuraman, R., & Manzey, D. H. (2010). Complacency and bias in human use of automation: An attentional integration. Human Factors, 52(3), 381-410.
Parker, S. K., & Grote, G. (2022). Automation, algorithms, and beyond: Why work design matters more than ever in a digital world. Applied Psychology, 71(4), 1171-1204.
Rinta-Kahila, T. et al. (2023). Vicious circles of skill erosion: Understanding the self-reinforcing dynamics of AI dependency. MIS Quarterly, 47(4), 1643-1672.