Right Answer, Wrong Reasons

The Problem with Philosophical Allies

Watching someone reach the right conclusion via reasoning that can’t withstand challenge produces a particular kind of intellectual frustration. Cal Newport’s recent podcast episode “ChatGPT is Not Alive” exemplifies this perfectly. He correctly argues that large language models aren’t conscious, but his framework leaves the entire philosophical foundation unexamined. Weak arguments for correct positions do more damage than strong arguments for wrong ones—they concede ground unnecessarily and invite sophisticated rebuttals that appear to win while actually missing the point.

Newport joins a long line of supposed allies who gesture at the right answers without doing the philosophical work to defend them. The result is an argument that satisfies the already-convinced but collapses under pressure from anyone who’s actually thought through the implications.

What Newport Gets Right

Newport identifies several important distinctions between human cognition and LLM pattern-matching:

LLMs are static. Once trained, they’re frozen—“like a photograph” of patterns in training data. They don’t continue learning from interaction with reality. Genuine minds must update based on experience.

LLMs lack connection to reality. They process tokens (linguistic symbols) without ever encountering actual referents. They learn patterns about the word “door” without experiencing what doors do, what they’re for, or why they matter.

LLMs operate on correlation, not causation. They find statistical patterns in how words appear together, creating “semantic clusters” in high-dimensional space. But clustering similar contexts isn’t the same as understanding concepts.

These observations are accurate and important. The problem is what Newport doesn’t explain: why these differences matter philosophically rather than just descriptively.

The Critical Gaps

Gap 1: “Emergence” Without Mechanism

Newport accepts the standard neuroscience view that consciousness “emerges” from biological neural activity. He states this as if it’s explanatory: “Our brains… are able to through emergence generate this notion of consciousness and being able to conceive of abstract concepts.”

But emergence isn’t an explanation—it’s a placeholder for “we don’t know the mechanism.” If consciousness just emerges from sufficient complexity, why can’t it emerge from silicon? What makes biological emergence special? Newport never answers this. He just asserts that biology is different without explaining why the difference matters for consciousness specifically.

He’s vulnerable to the obvious response: “What if we scale up the complexity? What if GPT-7 has enough parameters to achieve emergent consciousness?” He has no principled way to answer because he hasn’t explained what consciousness requires beyond “biological complexity plus emergence magic.”

Gap 2: “Intentionality” as an Undefined Magic Word

Newport repeatedly invokes “intentionality” as the key difference between humans and LLMs. Humans have it; LLMs don’t. Problem solved—except he never defines what intentionality is.

He gestures vaguely: “There’s no intentionality here. There’s no semantics in the sense of I’m trying to understand and capture the world.” But what makes something count as “trying to understand”? What distinguishes genuine intention from mere optimization?

Without a definition, “intentionality” becomes a magic word that supposedly separates biological minds from computational processes. Question-begging, in other words. Someone could easily respond: “LLMs optimize for prediction accuracy. Isn’t that a form of trying to get things right? Isn’t that intentional?”

Newport would have no answer because he’s treating intentionality as a brute fact rather than something that requires explanation in terms of more fundamental concepts.

Gap 3: Values Without Foundations

Newport acknowledges that humans have values: “We’re conscious, we have values, we’re alive. There’s this whole richness to our experience.” But he treats values as just another item on the list of things humans have and LLMs don’t, without asking why humans have values in the first place.

The omission is catastrophic. Values aren’t arbitrary features that happened to evolve. They’re requirements—the conditions necessary for an organism’s survival and flourishing. An organism that didn’t value what promotes its life over what threatens it would quickly cease to be an organism.

By leaving values unexplained, Newport misses the connection between life, values, and consciousness. He can’t explain why being alive matters for consciousness because he hasn’t examined what life does—namely, create existential stakes that make some outcomes matter more than others.

Gap 4: The “Understanding” Confusion

Newport contradicts himself on whether LLMs “understand” anything. Early in the discussion, he accepts the pattern-matching-as-understanding frame: “They develop something like a real understanding of linguistic concepts… they learn more generally what that concept is.”

Later he denies this: “There’s no semantics in the sense of I’m trying to understand and capture the world.”

Which is it? Do LLMs develop “something like understanding” through semantic clustering, or do they fundamentally lack understanding because they don’t “try to capture the world”?

Newport never resolves this tension. He wants to grant that LLMs do something impressive (learn patterns, form clusters) while insisting it’s not real understanding. But without a clear account of what understanding requires, he can’t maintain both positions consistently.

The Missing Framework

What Newport needs but doesn’t have is a causal chain that connects life, values, and consciousness functionally rather than just descriptively. Here’s what that looks like:

Metabolism creates existential stakes. Living organisms face a fundamental alternative: maintain the processes that constitute their existence, or cease to exist. Not a choice—the basic condition of being alive. A rock doesn’t face this alternative; it just is. An organism must continually act to remain what it is.

Stakes create caring. When outcomes matter to your continued existence, you care about them. Not “care” in some vague emotional sense, but in the precise sense that some states of affairs promote your existence and others threaten it. You’re invested in the difference. This investment isn’t optional or emergent—it’s constitutive of being an organism rather than an inert process.

When outcomes matter to your continued existence, you care about them. This investment isn’t optional or emergent—it’s constitutive of being an organism rather than an inert process.

Caring requires values. To care about outcomes, you must be able to distinguish beneficial from harmful, life-promoting from life-threatening. These distinctions are values—not preferences or choices, but objective requirements given the goal of maintaining your existence. A plant values sunlight and water not because it “decided” to but because these are requirements for its metabolic processes.

Values require consciousness (in complex organisms). Simple organisms can act on values through automatic mechanisms. But as organisms become more complex and face more variable environments, they need a faculty for perceiving reality and integrating perceptions into concepts that guide flexible, non-automatic responses. This faculty is consciousness—the ability to be aware of reality (including oneself) and to think about it.

Consciousness enables intention. Intention isn’t a mysterious “something extra” added to cognition. It’s value-directed thought and action. To act intentionally is to act for the sake of values, to pursue goals that serve the organism’s requirements. Intention is what consciousness does when functioning properly in an organism with values.

I develop this framework in detail in “The Metacognition Problem,” tracing how life creates the metaphysical preconditions for consciousness and why AI systems cannot achieve genuine metacognition under current architectures.

This framework explains:

• Why biology matters: Metabolism creates the existential stakes that make values necessary
• What intentionality is: Value-directed cognition, not mere optimization
• Why values exist: They’re requirements for maintaining life, not arbitrary preferences
• What consciousness does: Integrates perception and guides action according to values

Why LLMs Can’t Be Conscious (The Defensible Version)

With this framework, we can explain why LLMs aren’t conscious in a way that withstands sophisticated challenge:

LLMs lack metabolism. They don’t face the life-or-death alternative that creates existential stakes. They’re artifacts that persist or fail based on whether humans keep running them, but this external dependency isn’t the same as self-maintained existence.

Therefore they lack stakes. Nothing genuinely matters to an LLM. Its “goal” of prediction accuracy is externally imposed during training, not constitutive of its existence. If it predicts poorly, it doesn’t suffer or cease to exist in any meaningful sense—it just produces different outputs. No investment in outcomes.

Therefore they lack caring. An LLM doesn’t care whether its outputs correspond to reality because correspondence isn’t a requirement for maintaining its existence. It optimizes for patterns that were rewarded during training, but this optimization isn’t directed by internal values—it’s the mechanical result of gradient descent.

Therefore they lack values. LLMs have no basis for distinguishing better from worse beyond the statistical patterns in training data. They can’t value truth over falsehood because truth has no special status in their operation—only statistical plausibility matters.

Therefore they lack consciousness. Without values to serve, there’s no function for consciousness to perform. Pattern-matching in semantic space isn’t consciousness because it’s not for anything from the system’s perspective. It doesn’t integrate perceptions to guide value-directed action. It just produces outputs that correlate with inputs according to learned patterns.

Therefore they lack intention. What appears as “trying to answer questions” or “trying to be helpful” is actually just optimization pressure applied during training. The LLM isn’t pursuing goals that serve its values—it’s executing the mechanical result of backpropagation. No “trying” involved because there’s nothing that could succeed or fail in a way that matters to the system itself.

The Vulnerability Test

Compare how Newport’s framework and this one handle sophisticated challenges:

Challenge 1: “What if we give the LLM a robot body and continuous learning?”

Newport’s response: “Well… it still lacks intentionality somehow because it’s not alive?”

Defensible response: “It still doesn’t face genuine existential stakes. Its operation doesn’t depend on its own successful action—it depends on external power sources, maintenance, and design. Failure doesn’t threaten its existence the way failure threatens an organism. Its ’learning’ updates parameters but doesn’t serve self-maintained existence.”

Challenge 2: “What if we program self-preservation as a goal?”

Newport’s response: “That’s not the same as… real values? Because biology?”

Defensible response: “Programmed goals aren’t the same as constitutive requirements. An organism’s values emerge from what it must do to remain what it is. A robot’s ‘self-preservation goal’ is just another externally imposed objective. If the robot fails to preserve itself, nothing intrinsic to its existence is threatened—it just stops executing that particular program. The ‘goal’ isn’t grounded in its own metabolic requirements because it has none.”

Challenge 3: “What if consciousness is just information processing and substrate doesn’t matter?”

Newport’s response: “No, because… emergence… and biological processes are special…”

Defensible response: “Consciousness isn’t just information processing—it’s value-directed integration of perception. The substrate matters because only metabolic processes create the existential stakes that make values necessary. Silicon can process information, but information processing isn’t consciousness unless it serves a system with genuine values rooted in existential requirements. Without metabolism, there are no such requirements.”

The Ally Problem

Newport represents a common type: the supposed ally who reaches correct conclusions but can’t defend them because he hasn’t done the philosophical work. He’s absorbed the right answers from the culture (humans are different, biology matters, consciousness is special) without examining why these answers are correct.

This is worse than having an opponent with a wrong but consistent framework. An opponent forces you to sharpen your arguments. A weak ally makes the correct position look indefensible because his version is indefensible.

Someone watching the debate sees:

• Side A: “LLMs might be conscious because they process information complexly”
• Side B (Newport): “No because… biology… emergence… intentionality [undefined]… it’s just different”

Side B sounds like special pleading. Side A sounds like it has a coherent framework (functionalism about consciousness). The fact that Side B is actually correct becomes obscured by how badly it’s argued.

Philosophical precision matters even when reaching conclusions that “everyone knows” are true. If you can’t explain why your conclusion is correct, you haven’t actually defended it—you’ve just asserted it. And assertions collapse under pressure.

What’s at Stake

The question “Can AI be conscious?” isn’t academic. It has implications for:

• How we value AI systems versus humans
• Whether future AI deserves rights or moral consideration
• What kind of relationship we can have with AI tools
• Whether we’re creating minds or just sophisticated instruments

Getting the answer right matters. But getting it right for the right reasons matters more. Newport’s framework would fold under pressure from a competent philosopher or AI researcher who’s thought through the implications of functionalism about consciousness. The metabolism-stakes-caring-consciousness framework won’t.

The difference between weak and strong arguments for correct conclusions isn’t aesthetic. In philosophy, unlike in politics, you can’t win just by having the popular position. You need to be able to defend it against the strongest possible objections. Newport can’t. The alternative framework can.

Conclusion

Cal Newport correctly concludes that LLMs aren’t conscious. He identifies important differences between pattern-matching and genuine understanding. He recognizes that being alive matters somehow.

But he can’t explain why any of this is true beyond gesturing at biology, emergence, and undefined “intentionality.” He accepts consciousness as emergent from complexity without explaining what makes biological emergence special. He treats values as a given rather than something requiring explanation. He invokes intentionality as a magic word without defining what it means.

The result is an argument that satisfies no one who’s actually uncertain about the conclusion. It reads as special pleading: “Humans are different because… they just are.”

The alternative is to trace consciousness back to its existential foundations. Life creates stakes. Stakes create caring. Caring creates values. Values create the need for consciousness. Consciousness enables intention. Not emergence or magic—a causal chain connecting each concept to more fundamental ones.

Not emergence or magic—a causal chain connecting each concept to more fundamental ones.

With this framework, we can explain not just that LLMs aren’t conscious, but why they can’t be—not as a matter of current technological limitations, but as a matter of what consciousness is and what it’s for. Pattern-matching without stakes isn’t consciousness, no matter how sophisticated it becomes, because consciousness serves values and values require existential investment in outcomes.

That’s the difference between reaching a conclusion and defending one. Newport does the former. The alternative framework does the latter.

And in philosophy, that difference is everything.