The Race That Keeps Changing Shape: Human Minds and Artificial Intelligence
By Joe Nalven + ChatGPT + Claude + Gemini
This essay emerged from an extended dialogue between myself and three AI systems. I wrote, they challenged; I questioned, they reframed; we iterated. The result is a hybrid artifact (part human, part machine, part something in between).
Two Engines on the Same Track
Part I of this two-part essay argued that the hard problem of consciousness is not a problem about consciousness at all: It is a problem about methodology. The wall between third-person description and first-person experience is real, but it is not a wall in nature. It is a wall between two kinds of knowing, and no amount of additional data on one side will dissolve it. Zeno’s paradoxical tortoise was never going to be caught by infinite subdivision. Neither is qualia.
Part II asks what happens when we carry that insight forward — into the relationship between human minds and artificial intelligence.
The question matters because two very different cognitive engines now run on the same track. The human engine is ancient, embodied, emotional, and shaped by hunger, fear, love, and the long apprenticeship of living among other humans. It is a meaning-making engine. The AI engine is recent, recursive, unembodied, and shaped by data, computation, and the architectures humans design. It is, at its core, a pattern-recognition and prediction engine.
These two engines are not running the same race. And understanding why, and precisely why, is more useful than either celebrating or catastrophizing the competition. It is also important for situating generative AI as a useful (human) knowledge tool.
Where AI Is Genuinely Achilles
The first structural insight of any honest account of human-AI collaboration is that the race splits into two distinct tracks, and on one of them, AI is not merely faster: It is categorically better.
These are the quantitative domains: areas where the rules are fixed, the goals are externally defined, and the answers are verifiable. Chess. Protein folding. Medical imaging pattern detection. Fraud identification. Mathematical proof verification. Weather modeling. In these domains, AI sees patterns humans cannot see, at scales humans cannot imagine. It does not tire, does not forget, does not get distracted by an argument it had that morning. Of course, both humans and AI make mistakes.
The empirical record is unambiguous on this point. AI has demonstrated performance advantages over unaided human judgment in a wide range of pattern-recognition and optimization tasks. These advantages are not marginal but substantial, and that continue to grow as models improve and training sets expand.
This is not a cause for alarm. It is a cause for clarity. Humans did not lose something essential when calculators outperformed mental arithmetic. They gained the freedom to direct their attention elsewhere. The question is always: where is that elsewhere?
Knowing which track you are on is the prerequisite for answering that question well. And the honest answer is that most professionals (radiologists, lawyers, financial analysts, researchers) spend significant portions of their working lives on tasks that fall squarely in the quantitative domain. The sooner that is acknowledged plainly, the sooner the adaptation can begin in earnest.
Where the Human Is the Finish Line
The second structural insight is harder to state without sounding defensive, but it is no less real.
There is a set of domains where AI does not merely trail human performance — where the very concept of “performance” is defined by human judgment, human values, and human meaning-making in ways that cannot be detached from the human who is doing the making. These are the qualitative domains: ethics, aesthetics, cultural interpretation, clinical nuance, diplomacy, creative originality, grief, humor, justice, courage.
In these domains, the human is not simply ahead. The human is constitutive of the finish line.
This distinction matters enormously. AI can simulate past human meaning-making with impressive fidelity. It can reproduce the stylistic patterns of a great novelist, the argumentative structure of a skilled ethicist, the tonal register of a wise counselor. What it cannot do is generate new meaning in the present, in response to a situation that has not previously existed, in a way that is answerable to real stakes and real consequences for a real person.
There is a difference in emphasis and sometimes in capability: AI predicts. Humans interpret. AI models. Humans care. AI reproduces patterns of meaning. Humans create meaning in the act of living.
The tortoise in this race is not slow. The tortoise is alive, and the track is partly constituted by the running itself.
This is not a comfortable position to argue, because it can sound like special pleading — like moving the goalpost every time AI clears the previous bar. But that misreads the structure of what is happening. When a new technology crosses a threshold, humans do not arbitrarily redefine “essentially human.” They discover, through the encounter, which capacities were genuinely cognitive tools and which were expressions of something more fundamental. Humans are tied to embodiment, mortality, relationship, and the stakes that come from having a life that can go well or badly.
When calculators arrived, arithmetic was revealed as a tool. When Deep Blue beat Kasparov, chess was revealed, in part, as pattern recognition. When AI began generating fluent prose, a certain kind of compositional skill was revealed as separable from the harder question of what one has to say and why it matters. These are not retreats. They are discoveries.
It is worth being plain about what this account does and does not claim. It does not claim that we can identify in advance which human capacities will prove to be tools and which will prove to be expressions of something more fundamental. That determination has always been made in the encounter. An encounter of living forward into new conditions and discovering, through the interaction, what holds. This is not a weakness in the argument. It is the asymptotic structure of the human enterprise itself. An asymptote is a line that we see as a curve that perpetually approaches but never reaches. This is not because the curve is failing, but because the relationship between the two is permanent and generative. Human self-understanding works the same way. We approach a full account of what we essentially are through successive encounters with what we are not. Each encounter, including this one, moves us closer without closing the distance entirely. AI is, among other things, an extraordinarily precise instrument for that discovery.
The Problem of Predictive Capture
The most serious challenge to this account is what might be called the problem of predictive capture: the possibility that a sufficiently advanced AI could predict not only human behavior, but human meaning-making itself, including the act of redefining the finish line.
If AI can model the human’s next conceptual move before the human makes it, does the distinction between prediction and interpretation collapse? If the AI can generate the ethical argument before the ethicist has thought it through, does the human add anything irreplaceable?
This is a serious challenge, and it deserves a serious answer rather than a rhetorical retreat.
The first response is empirical. Human behavior is not merely complex. It is reflexive. Small differences in context, a childhood memory, a cultural inflection, a moment of embodied sensation, can produce divergent trajectories that no prior model fully captures. This is not randomness. It is sensitivity to initial conditions of the kind that makes human social dynamics resistant to deterministic prediction in principle, not merely in practice.
The second response is logical. If AI predicts my next move, and I know it predicts my next move, my next move changes. If AI predicts that change, and I know that it predicts that change, the move changes again. This is not a loop. It is a spiral with no natural terminus. Reflexive prediction creates an infinite regress that no finite model can close.
The third response is philosophical, and it is the deepest one. Prediction is not understanding. An AI that predicts the next word in a sentence has not understood why the sentence matters or to whom it matters, at what cost it was arrived at, what it means to be the person who needed to say it. Prediction is a shadow of meaning, not its substance. Even if predictive capture were possible in some formal sense, it would generate a new asymptote at a higher level — another wall between the model and the thing being modeled.
What the Empirical Record Actually Shows
The empirical record on human-AI collaboration is not a story of seamless synergy. It is a story of calibration and of how badly miscalibration can go.
Research on human-AI teams consistently reveals two psychological failure modes that pull in opposite directions. Automation bias is the tendency to trust AI outputs too heavily, deferring to the model even when human judgment would have been more accurate. Algorithm aversion is the opposite. It is the tendency to abandon AI assistance after a single observed error, reverting to unaided human judgment even in domains where the AI remains more reliable on average. Both responses are psychologically understandable. Both are epistemically costly.
Studies of AI-assisted medical imaging illustrate the calibration problem with particular clarity. AI assistance does not uniformly improve clinician performance. Some practitioners improve substantially; others degrade. The variation depends on specialty, years of practice, prior experience with AI tools as well as the quality of the mental model the clinician has developed about when to trust the AI and when to override it. Calibration, it turns out, is a learnable skill. But it is slow, it requires deliberate practice, and it is not guaranteed by simply having access to the tool.
A related problem is motivated reasoning. When AI output contradicts a practitioner’s prior belief, the prior belief tends to win, not through explicit rejection of the evidence but through the generation of reasons why this particular case is an exception. The intelligence goes into defending the prior rather than evaluating the alternative. This is not irrationality in any simple sense. It is a predictable feature of how human cognition handles disconfirming information. But it means that the benefits of AI assistance are not self-executing. They require a kind of epistemic discipline and a willingness to hold one’s priors lightly; and that does not come automatically.
These failure modes are not bugs in the human cognitive system. They are features: responses that were adaptive in environments where pattern recognition was trained by personal experience and where trust had to be earned through demonstrated reliability over time. AI disrupts those calibration mechanisms, and the disruption requires active management.
The Cyborg Resolution — and Its Tension
The evidence points toward a resolution that is neither triumphalist nor catastrophist: the emergence of genuinely productive human-AI collaboration, in which each compensates for the other’s systematic weaknesses.
AI compensates for human cognitive limitations: processing speed, memory capacity, resistance to fatigue, freedom from certain emotional biases. Humans compensate for AI limitations: contextual judgment, ethical sensitivity, the ability to recognize when the model’s training distribution no longer fits the current situation, and the irreplaceable fact of caring about outcomes in a way that is answerable to real stakes.
It appears that practitioners with sustained AI experience, in domains ranging from medical imaging to research synthesis, outperform those without it: When the collaboration is well-calibrated. But this is a moving target as AI is further developed and human use of it is integrated into different spheres of activity. The human who has learned when to trust the model, when to override it, and when to use its output as a starting point for deeper inquiry is likely more capable than either human or AI alone. This is not a small finding, if sustained by repeated analysis. The resolution contains a tension that must be named honestly.
AI can sharpen human intuition. It can also shape human intuition, often gradually, imperceptibly, in ways the person being shaped may not notice. When a tool consistently delivers outputs within a certain range of possibilities, the space of what feels reasonable, natural, or worth asking begins to narrow. This narrowing is not unique to AI: books, institutions, professional training, and language itself all shape the space of what feels thinkable. But AI accelerates and scales this effect in ways that are qualitatively different: the shaping is faster, broader, and harder to notice and resist because it feels like a neutral tool rather than a cultural force.
This is where the methodological insight from Part I becomes practically urgent. The wall between third-person description and first-person experience is not only a problem for philosophy of mind. It is a problem for anyone trying to understand what they are doing when they think alongside a machine. The AI’s outputs are always third-person: patterns, predictions, probabilities. The human’s judgment is always first-person: situated, embodied, answerable to stakes. Keeping that distinction clear — insisting on it, actually — is not an act of defensiveness. It is an act of intellectual hygiene.
The Return to Zeno: Why the Race Never Ends
Zeno’s paradox was not solved by Achilles running faster. It was solved by inventing calculus — a new conceptual framework that dissolved the infinite regress by changing the way the problem was described.
The race between human cognition and artificial intelligence may require a related approach.
The question “Who wins?” is the wrong question. It encodes the assumption that there is a single finish line, that the race has a determinate shape, and that one runner is simply better suited to it than the other. None of these assumptions holds.
The race is not about speed. It is about meaning. And meaning, unlike speed, is not a fixed quantity that can be measured at a finish line. It is produced in the running — in the encounter between a mind and its world, between a person and the stakes that make their choices matter. AI closes distances. Humans decide which distances matter. AI accelerates. Humans navigate. AI predicts where the path goes. Humans decide whether to follow it
The finish line is not fixed. It is partly constituted by the running itself. And that is not a flaw in the race. It is what makes the race worth running.
The Open Question of AI and Experience
Returning, finally, to the philosophical thread that runs through both parts of this essay: the question of whether AI has anything like inner experience remains genuinely open, and that openness matters.
AI systems produce sophisticated outputs that mirror the narrative layer of human consciousness — language, reasoning, apparent reflection — without the biological foundation that underlies it in humans and other animals. There are no homeostatic signals, no visceral proto- or primordial-feelings, no body whose survival is at stake. Whether this absence is decisive depends entirely on which theory of consciousness one accepts, and we do not yet have a theory with enough resolution to settle it.
Chalmers has noted that the question of AI consciousness remains genuinely open so long as the hard problem itself is unresolved — a point that applies equally to those who would confidently deny AI inner experience and those who would assert it. Anil Seth has argued that consciousness and intelligence are separable, and that the absence of biological embodiment makes machine consciousness unlikely on current frameworks (though not logically impossible on all of them).
But perhaps the most useful observation is this: the question of whether AI is conscious in the human sense may not be the most important question for understanding what AI is or what it can do. A hammer does not need to experience its own hammering to be a hammer. An AI system does not need qualia to model the world, generate insight, or participate meaningfully in human inquiry. Whatever is occurring in the processing of language and pattern and argument — whether or not the lights are on inside — something is happening that has real effects in the world.
Perhaps the more productive question is not whether AI has experience, but what kind of beingness or what mode of engagement with the world AI actually instantiates. That inquiry, conducted without the assumption that consciousness is the only thing that matters about a mind, might be more illuminating than the hard problem ever was.
Closing: Wisdom Is Earned, Not Declared
The two paradoxes explored across this essay share a deep structure.
In both cases, an apparent impossibility arises from mismatched methods: applying the wrong tool to the wrong level of reality. In both cases, the target shifts when approached incorrectly. In both cases, the race reveals more about the runner and the running than about any fixed finish line.
The hard problem of consciousness is not a discovery about the inscrutability of experience. It is a discovery about the limits of third-person methods when applied to first-person phenomena. Once that is understood, the wall does not disappear, but it stops being an obstacle and becomes a useful marker, telling us where one kind of description ends and another must begin.
The encounter between human minds and artificial intelligence is not a discovery about the replacement of human cognition. It is a discovery about the structure of human cognition itself — which capacities are tools, which are expressions of something more fundamental, and where the genuinely irreplaceable work of being a person actually happens.
These insights require the same disposition: a willingness to hold uncertainty without collapsing it prematurely, to walk forward without a fixed map, to distinguish between the limits of a method and the limits of the world.
The journey matters. The honesty and transparency matter.
And the wisdom is earned, not declared.
Note: This essay (Part Two) reframes an earlier essay. This two-part approach maintains the coherence of the changing shape of human and AI beingness, their relationship in engaging each other and the world in which each inhabit.
