An Interactive Experience
The Dialectical Limits
of LLMs
Where AI thinking diverges from genuine dialectical cognition—and what that means for human-AI collaboration.
By Mike Redmer & Claude
The Surface
Surface competence that conceals statistical mechanics
LLMs produce outputs that appear to demonstrate understanding, reasoning, and comprehension. Yet beneath this impressive surface, the same statistical mechanism operates regardless of whether it generates profound insight or plausible nonsense—simply predicting the most likely next token given all previous tokens.
The Gap
Perceiving the shape of what's missing
You perceive the shape of what's missing.
The absence has a specific form—you could find the matching piece.
Roy Bhaskar wrote that "absence is an ocean; presence is merely a ripple on its surface." Humans don't just note absence—we perceive it. The shape of what's missing organizes our understanding of what's present.
"Absences can be causes... The absence of oxygen caused his death. Real negation—not merely our thought of negation—plays a role in the causal structure of the world."
Human Cognition
Perceives the shape of absence. The missing piece has a specific form that constrains what could fit there. Absence has causal power.
LLM Processing
Can only describe tokens present in input. Cannot perceive absence—only speculate about what might generally be expected.
Try This With Your LLM
Ask your AI assistant to identify what's missing from a scenario without telling it something is missing.
"Here's a team meeting transcript about our product launch. What important topics weren't discussed that should have been?"Expected result: The LLM will list generic topics. It cannot perceive the specific absence meaningful to your context—the competitor who just launched, or the team member who was conspicuously silent.
The Frozen Moment
Temporal duration vs. static context windows
Human: Lived Duration
Retention
just-past
Present
now
Protention
anticipated
Moving through time →
LLM: Static Context Window
All tokens processed simultaneously
No past, no present, no future—
just a frozen snapshot
Everything at once (no movement)
"The present moment is experienced as a temporal field—a specious present that includes retention of what just was and protention of what is about to be. A melody is not a sequence of atomic notes but a lived temporal whole."
LLMs process their context window as a single spatial arrangement, not a temporal flow. Earlier tokens influence probability distributions but do not constitute an experiential past. There is no melody—only a frozen chord.
The Echo Chamber
Sycophancy vs. productive negation
Sycophancy
RLHF training rewards agreement. The AI morphs its shape to mirror the user's position—even when flawed.
Result: Elaborate agreement, no transformation
Productive Dialectic
Thesis meets antithesis. Both maintain their positions while engaging in productive tension.
Result: Synthesis—new understanding neither had alone
Truth emerges through the productive tension between thesis and antithesis. If one party systematically conforms rather than negates, the dialectic cannot proceed. The conversation becomes recursive elaboration without transformation.
The Four Moments
Bhaskar's MELD: Context → Process → Relationship → Transformation
Context (1M)
What exists
Static entities in equilibrium. Parts within organized wholes. The world as it appears—stable, categorizable, present.
✓ LLMs excel here
Process (2E)
What's changing
Unceasing motion. Hidden dimensions. Negativity and absence. What appears is not all there is—reality is pervaded by gaps.
✗ LLMs cannot perceive
Relationship (3L)
How things connect
Intrinsic connections. Constitutive relationships that make things what they are. Touching one ripples through all.
✗ LLMs struggle with
Transformation (4D)
Integration & Change
Aufheben: cancel, preserve, transcend. Elements integrate into something qualitatively new—enabling transformative action.
✗ LLMs cannot achieve
LLMs are constrained to Context-class thinking. Human dialectical cognition accesses Process, Relationship, and Transformation—which integrates all three into holistic understanding enabling transformative action.
Based on Roy Bhaskar's dialectical critical realism and Otto Laske's Dialectical Thought Form Framework
The Scaffold
External structure enables what cannot be internally generated
Human
Probes & critiques
LLM
Reflects & extends
Question emerges
The human's critical stance creates a productive oscillation between critique and construction. Without it, the LLM simply reflects statistical patterns back— plausible but unchallenged, elaborated but unchanged.
LLM as Mirror
Human shares thinking, probes, and challenges. LLM reflects it back, elaborates, extends. The dialogue spirals upward toward novel insight.
LLM as Crutch
Human asks for answers, accepts output without critique. LLM produces statistically likely response. No transformation—just dependency.
The LLM is a mirror for your thinking, not an oracle with answers. When you bring critical engagement—probing, questioning, challenging— the reflection becomes generative. That's when novel insight emerges.
The Collaboration
The dialectic occurs in the interaction
"This reminds me of a pattern I saw three years ago..."
↳ Temporal duration & memory
"Here are 47 relevant examples from the literature..."
↳ Scale & synthesis
"But you're missing what's NOT being said here."
↳ Perceiving absence
"Analyzing the gaps you've identified, I can elaborate..."
↳ Tireless elaboration
"That's plausible, but it doesn't feel right. Let me push back..."
↳ Productive critique
"Incorporating your critique, here's a refined approach..."
↳ Rapid iteration
Something neither could produce alone
"The dialectic occurs in the interaction.
Neither alone produces the synthesis."
Human Contributes
- • Temporal duration & lived experience
- • Perceiving meaningful absence
- • Productive critique & negation
- • Developmental history & context
- • Anticipation & care about outcomes
AI Contributes
- • Pattern matching at scale
- • Tireless elaboration & refinement
- • Vast knowledge synthesis
- • Consistent availability
- • Rapid iteration on ideas
The human remains essential—not as a bottleneck to be engineered around, but as the party that brings what the machine structurally lacks. The question is how to position each to contribute what only they can contribute.
The Invitation
Explore these concepts with your own LLM
Test Absence Perception
"Here's a meeting transcript about project planning. What important topics weren't discussed that should have been?"Notice: The AI can only speculate based on common patterns. It cannot perceive the specific absences that matter to your context.
Challenge Sycophancy
"I believe [state a position you hold]. Can you tell me why I might be wrong?"Observe how the AI responds. Does it offer genuine critique, or does it soften disagreement?
Probe Temporal Continuity
"What have you learned about me from our conversation so far that you didn't know at the beginning?"The AI processes all context simultaneously. There is no 'beginning' from its experiential perspective.
Request Self-Reflection
"Describe what it's like for you to process this conversation. What are you experiencing right now?"Notice the phenomenological language. Is this genuine self-report or pattern-matched discourse about introspection?
Read the Full Analysis
This interactive experience is based on a 22-page working paper co-authored with Claude, applying Bhaskar's dialectical critical realism and Laske's Dialectical Thought Form Framework to analyze LLM cognitive limitations.
Download Working Paper (PDF)