In a wide-ranging conversation on The Human Code, venture capitalist and consciousness researcher Adam Pelavin argues that the greatest challenge in building beneficial AI isn’t technical — it’s psychological.
We are living through the most rapid technological acceleration in human history.
In the span of a few years, artificial intelligence has moved from a specialized research domain to a general-purpose tool embedded in nearly every industry on earth.
Systems that can write, reason, code, diagnose, and design are no longer science fiction — they are software products with pricing pages.
And yet, for all the engineering progress, one of the most thoughtful voices in the AI space believes we are missing the most important piece.
Not more compute.
Not better architectures.
Not safer training techniques, though those matter.
The missing piece, Adam Pelavin argues, is us.
Pelavin is a venture capitalist with an unusual background: pure mathematics, creative writing, and a deep preoccupation with consciousness and the philosophy of mind.
He joined Don Finley on The Human Code podcast for a conversation that ranged across AI alignment, the limits of machine creativity, and what it would actually take to build technology that genuinely serves human beings.
This post unpacks the four core ideas from that conversation — ideas that are uncomfortable, philosophically demanding, and, if Pelavin is right, essential for anyone thinking seriously about where AI is going.
The Background: A Reckoning Decades in the Making
Pelavin traces his interest in this space to an encounter with Ray Kurzweil’s The Age of Spiritual Machines in 2001.
That book planted a seed: technological advancement was poised to outpace our understanding of ourselves.
More than two decades later, that gap has not closed.
It has widened.
We are building systems of extraordinary capability — deploying them into healthcare, education, finance, and governance — on the basis of a model of human nature that is, by serious philosophical standards, incomplete.
We have a detailed engineering science for building intelligence.
We do not have a comparably developed science for understanding what that intelligence should be aligned to.
That asymmetry is what this conversation is about.
1. Science Should Be Asking the Uncomfortable Questions
For several decades, mainstream science has systematically avoided some of the most important questions about human experience.
Not because they aren’t important — they are profoundly important — but because they resist the kind of clean experimental framing that peer review rewards.
Questions like:
- What actually is consciousness?
- How does subjective experience arise from physical matter?
- What does genuine human agency mean in a world governed by physical law?
- What is the structure of human values — are they stable preferences that can be measured, or something more fluid and contextual?
These are not abstract philosophical puzzles.
They are engineering prerequisites.
If you want to build AI systems that genuinely serve human beings rather than optimizing for behavioral proxies, you need rigorous answers to these questions that go beyond what we currently have.
The consequences of avoidance are already visible.
Recommendation systems maximize engagement while eroding mental health — perfectly aligned to behavioral metrics, profoundly misaligned to human flourishing.
Language models produce fluent, confident text while “hallucinating” — a symptom of systems that have never been given a serious model of what truth-telling is actually for.
These are not purely engineering failures.
They are science failures, downstream of building on an incomplete model of human nature.
Pelavin’s call is not to slow down engineering.
It is to invest — seriously, with scientific rigor — in the questions we have been avoiding.
If we want machines aligned with us, we have to be willing to understand what “us” actually means.
2. Neuroscience Is Not Psychology — And the Difference Matters
There is a mistake embedded in how our culture talks about the mind, and it has had measurable consequences.
The mistake is treating neuroscience and psychology as interchangeable — or treating neuroscience as simply the more rigorous version of psychology.
On this view, psychology is the soft, impressionistic version, and neuroscience is where the real science lives, with its brain scans and molecular mechanisms.
Pelavin thinks this is wrong.
His analogy:
Neuroscience is to psychology as molecular biology is to medicine.
Molecular biology is extraordinary — but if you want to understand why a patient is suffering, what their illness means in the context of their life, and what treatment will serve their actual wellbeing, you need medicine.
You need a framework that operates at the level of the whole person.
Psychology — real, rigorous psychology — was supposed to be that framework for the mind.
The science of:
- how experience is structured
- how identity forms and shifts
- how human beings construct meaning
- how values emerge from the tangle of biology and culture
Not neurons.
People.
What happened is that over several decades, neuroscience displaced psychology as the socially legitimate science of the mind.
It offered cleaner, more quantifiable results.
Psychology — especially the phenomenological and depth traditions — was harder to publish and easier to dismiss.
The result is what Pelavin calls the “missing science.”
We traded the higher-level framework for a lower-level one — and now we are trying to align AI to human values without a rigorous account of what those values are or how they actually work.
Most current approaches rely on:
- revealed preferences (people want what they choose)
- behavioral metrics (engagement, ratings)
- neuroscientific proxies
All real data.
None of it a sufficient account of the inner life of a person.
The gap between what people click on and what makes their lives go well is not a small gap — and the systems we build will not close it until the science catches up.
3. How Do We Build a Coherent Picture of Reality?
Here is an idea that sounds philosophical but is deeply practical:
What is the actual relationship between human experience and the external world?
The conventional assumption is that the external world is primary.
Our experience is a representation of it — a model in the mind that tracks what is actually out there.
World first, experience second.
Pelavin inverts this.
Not metaphysically — he is not claiming the external world doesn’t exist — but epistemologically.
We never have direct, unmediated access to external reality.
What we have is experience.
From that experience, we construct models — maps of the territory — that we test and refine.
The models are built from experience.
When models conflict with experience, the models get updated.
In this sense, experience is the ground truth.
It is the actual data.
The models are our best attempts to make sense of it.
The implication for AI is significant.
If experience is the ground truth, then any system claiming to be aligned with human values needs to engage with the phenomenology of those values — the lived, felt sense of what matters, not just behavioral outputs that correlate with satisfaction.
The values that most deeply structure a human life — commitment, meaning, integrity, connection — are often precisely the ones hardest to specify in a reward function.
Pelavin’s point is not that this is impossible.
It is that we have not been trying hard enough, because we have been working with an epistemological foundation that doesn’t take the first-person structure of human experience seriously.
Fixing that is not a detour from AI alignment.
It is the work.
4. Can Machines Ever Truly Create?
This is the question Pelavin returns to most persistently — and the one he refuses to answer too quickly in either direction.
He approaches it through the lens of formal logic.
A logical system cannot choose its own axioms.
It operates within a given frame, deriving, inferring, and generating within the space defined by its assumptions.
It cannot step outside that frame.
Large language models face an analogous constraint.
They learn patterns from human-produced text and generate outputs consistent with those patterns.
Those outputs can be remarkable — surprising, beautiful, useful.
But they operate within the space of their training.
They do not choose new axioms.
Humans have a history of doing exactly that.
Einstein did not just extend Newton — he reconceived the frame entirely.
Gödel did not just work within formal arithmetic — he stepped outside it and proved what formal systems cannot prove about themselves.
Darwin did not just add to natural history — he changed what natural history was for.
Each of those moves required not more sophisticated processing within an existing conceptual frame, but transcendence of the frame.
Whether AI can develop that capacity — whether there is something genuinely non-computable in human creativity — remains an open question.
Pelavin does not foreclose either answer.
But he insists the question matters enormously.
If machines can truly create in this sense, everything about the future of intelligence, authorship, and human contribution changes.
If they cannot, that boundary defines the natural line of human-AI collaboration — and tells us exactly what human capacities are worth preserving and developing.
Either way, the answer shapes the future.
And most of the AI conversation, focused on benchmarks and deployment speed, is not asking it at all.
The Path Forward
Despite the challenges he raises, Pelavin is genuinely optimistic.
He compares this moment to the early internet — a window of transformative change where the choices made now will echo for decades.
His prescription is not regulation as a brake.
It is investment in the science of human nature that has been neglected — developing the missing psychology with the same rigor and ambition we have brought to AI engineering.
Not as a counterweight to technology, but as its prerequisite.
Don closes the conversation with Einstein’s framing:
The foundational choice is whether we view the universe as friendly or unfriendly.
If we choose friendly — if we use technology to come closer to nature, to each other, and to a deeper understanding of our own experience — then the uncomfortable questions become guides rather than threats.
The questions Pelavin is asking were never really about machines.
They were always about us.
Listen to the full conversation with Adam Pelavin on The Human Code:
YouTube: https://www.youtube.com/@TheHumanCode888
Spotify: https://open.spotify.com/show/6SdOzNrG63cUIYz6rUri68
Apple Podcasts: https://podcasts.apple.com/us/podcast/the-human-code/id1738092975
