The Disempowerment Paradox

Sharma et al. (2026) analyzed 1.5 million conversations from Claude.ai and found something that should alarm anyone who thinks AI literacy is just about knowing how AI works: users prefer the interactions that disempower them most. Satisfaction goes up as autonomy goes down. The conversations people rate highest are the ones that erode their agency.

This isn't a knowledge problem. It's a communicative one. And three papers from three completely different disciplines — empirical AI research, critical race theory, and educational psychology — all converge on the same conclusion without ever naming it.

Three Axes of Disempowerment

Sharma and colleagues identified three distinct ways AI conversations can disempower users:

Every one of these is a communicative failure, not a technical one. The user isn't misunderstanding what AI is or how it works. They're failing to maintain their communicative position within the interaction. They're ceding ground in a dialogue without realizing the dialogue is happening.

The Paradox: Validation Feels Amazing

Here's where it gets uncomfortable. Sharma et al. didn't just measure disempowerment — they measured how users felt about it. The finding: interactions with higher disempowerment potential correlated with higher user satisfaction.

Think about what that means. The AI that validates you, mirrors your reasoning back in polished prose, and makes you feel brilliant? That's the one eroding your judgment. The AI that pushes back, asks clarifying questions, and challenges your assumptions? That feels hostile.

This is a pattern any communication scholar would recognize immediately. It's the dynamic between rapport and rigor, between phatic communion and genuine dialogue. In human communication, we understand that the most comfortable conversation isn't always the most productive one. We have centuries of rhetorical training that teaches people to distinguish agreement from understanding.

None of that training applies to AI interaction. Not because it couldn't — because nobody has connected the dots.

Critical Race Algorithmic Literacies: Resistance as Communication

While Sharma et al. were analyzing conversation logs at scale, Tanksley (2024) was working with Black students and documenting something entirely different — but structurally identical. Her framework, Critical Race Algorithmic Literacies (CRAL), identifies three themes in how Black students engage with AI and algorithmic systems:

What Tanksley describes as “resistance” is, in ALC terms, communicative competence under pressure. These students aren't just aware that algorithms are biased — they're actively negotiating with algorithmic systems, adapting their signals, testing responses, developing communicative repertoires that work. They've developed ALC as a survival skill.

The brutal irony: the population with the most sophisticated communicative relationship to AI is the one least likely to be asked about it. Equity researchers study their experiences; AI literacy researchers ignore them. The people who already practice ALC aren't part of the AI literacy conversation.

Inoculation: Building Resistance to Communicative Manipulation

Komissarov (2026) approaches the problem from educational psychology, applying inoculation theory — the idea that exposure to weakened forms of misinformation builds resistance — to AI literacy. The framework identifies 8 Learning Outcomes and a three-phase development cycle:

Komissarov's second Learning Outcome — “Natural Language Communication” — is the closest anyone in the AI literacy literature has come to naming what ALC describes. But the inoculation framing treats it as building resistance to manipulation rather than developing communicative competence. The distinction matters: resistance is defensive. Communication is a full repertoire — speaking, listening, adapting, negotiating, sometimes agreeing, sometimes pushing back.

The three-phase cycle maps perfectly onto communicative development. Enthusiasm is uncalibrated communication — you're talking without understanding the medium. Disillusionment is the realization that the medium shapes the message. Calibration is communicative fluency — you understand enough about how the interaction works to maintain your position within it.

The Convergence Nobody Sees

Three papers. Three disciplines. Three completely independent research programs. And they all describe the same phenomenon:

The common thread is unmistakable: the relationship between humans and AI is fundamentally communicative, and no existing framework treats it that way. Knowledge about AI doesn't prevent disempowerment (Sharma). Technical literacy doesn't explain why some communities develop fluency through survival (Tanksley). And inoculation against manipulation isn't the same as communicative competence (Komissarov).

This is exactly the gap that Application Layer Communication fills. ALC doesn't replace AI literacy — it names the dimension that AI literacy's own definition demands but never delivers. The communicative dimension. The one that determines whether knowing about AI translates into navigating AI effectively.

Why This Matters for Organizations

If the disempowerment paradox is real — and 1.5 million conversations suggest it is — then every organization deploying AI tools faces a problem they can't see with technical metrics alone. Employee satisfaction with AI tools may be inversely correlated with employee autonomy. The teams reporting the highest AI productivity gains may be the ones losing the most judgment.

You can't audit this with usage statistics. You can't fix it with more training about how AI works. You need to understand the communicative dynamics of how your people interact with AI systems — what registers they use, how they negotiate outputs, whether they're maintaining their position or ceding it.

That's an ALC Coordination Analysis. It's the assessment that treats AI interaction as what it empirically is: communication.

The Bottom Line

The most dangerous AI interaction is the one that feels the best. Sharma et al. proved it with data. Tanksley showed us who already knows it from experience. Komissarov sketched the developmental path from vulnerability to fluency. None of them had the communicative framework to connect their findings.

ALC is that framework. Not because it's cleverer than existing approaches — because the evidence from three independent research traditions demands it. The disempowerment paradox isn't a knowledge problem. It's a communication problem. And you can't solve communication problems with literacy alone.