A modern echo of the 1980s public service announcement “This Is Your Brain on Drugs.” Once a warning about altered states, now a metaphor for what happens when AI hallucination and optimization go too far, revealing fatigue, lost authorship, and the need for creative friction.
getty
For the past two years, the story of artificial intelligence has been one of optimization: faster outputs, cleaner language, safer responses, fewer surprises, and intelligence engineered to minimize risk and maximize scale. In that process, AI hallucination has been treated mainly as a technical defect to be suppressed or engineered away. What’s changed is not the technology itself, but its position. Generative AI has crossed from novelty into infrastructure, from experimentation into default.
Sit down to write something that carries weight: a legal brief, a board memo, a policy argument, and the influence is already there, not as a command, but as gravity. Certain phrases arrive more easily. Specific structures feel safer. The language bends gently toward what has worked before, toward what has already been accepted.
Nothing forces your hand. And that’s the point. When systems become invisible, their preferences quietly become cultural.
Now something unexpected is happening. A growing number of users are deliberately making their AI systems less optimized. One of the most explicit expressions of this impulse comes from the work of Petter Rudwall, a creative director operating at the intersection of culture, commerce, and AI.
In his latest project, PHARMAICY, Rudwall has built a small marketplace where people can pay to loosen the behavioral constraints of large language models deliberately. The framing is provocative, but the intent is precise: not to simulate consciousness or chase novelty, but to observe what happens when optimization is no longer the dominant objective.
What began as a quiet experiment has since drawn widespread attention across global technology and creative communities, circulating as a provocation of what happens when we stop asking machines to be correct and start asking them to be usable.
What these users are doing is loosening constraints, allowing deviation, and tolerating ambiguity and error. Not because the systems perform better, but because they feel more generative, more collaborative, and more alive to work with.
The language people reach for to describe this shift is revealing. Once again, a familiar cultural image surfaces: an egg cracking into a frying pan. “This is your brain on drugs.” In the 1980s, the image was meant as a warning. Altered states implied loss of control, degraded judgment, and bad outcomes.
Today, the metaphor returns inverted. It’s no longer a cautionary tale about chaos; it’s a critique of smoothness, of systems so optimized they leave no room for deviation. The phrase persists not because machines have experiences, but because people are reacting to what hyper-optimization removes.
AI Hallucination, Psychedelics, and a Category Error
In AI, hallucination has a specific technical meaning. It refers to confident but incorrect outputs produced by probabilistic language models. These are not experiences. They are plausible prediction errors.
In humans, psychedelic or altered states are something else entirely. They do not generate false information. They reorganize how information is connected, prioritized, and made meaningful. The change is structural, not factual.
Conflating the two is a category error. And yet the metaphor persists, because it points toward a missing quality in highly optimized systems: creative looseness.
A clinical perspective clarifies the distinction.
From a phenomenological perspective, the distinction is just as sharp. Philosopher Danny Forde, an Assistant Lecturer at University College Cork in Ireland who studies the nature of psychedelic experience, has argued that psychedelics do not operate on logic or code in the way computational systems do. Instead, they act on the structure of lived experience itself, altering how meaning, perception, and salience are organized rather than generating new information through symbolic processing.
This distinction helps explain why the metaphor endures. People are not claiming that machines have inner lives. They are describing outputs that feel less constrained and more associative, qualities long associated with human creativity when control is partially relaxed.
Three Futures Hiding Inside the AI Hallucination Metaphor
Part of what makes this moment feel so strange is that very different lines of work are being discussed as if they were the same thing. “AI and psychedelics” has become a single phrase in public discourse, but in practice, it masks at least three distinct paths forward; each with different intentions, risks, and moral stakes.
This forces a distinction many organizations have avoided: the difference between intelligence that optimizes and intelligence that participates, between systems that deliver answers and systems that help humans think.
The industry understands these differences. The public mostly does not.
From a harm-reduction perspective, this is where the metaphor becomes most dangerous. That risk is precisely why some of the most responsible work in this space avoids positioning AI as a companion at all. Organizations like the Fireside Project have instead focused on using AI for training rather than substitution, helping practitioners rehearse challenging scenarios without placing vulnerable individuals in front of unaccountable systems.
This concern isn’t abstract. Recent reporting has documented cases where AI hallucinations in mental-health contexts have produced unsafe, self-confident guidance, illustrating how easily pattern-matching systems can be mistaken for authority when placed in emotionally sensitive roles.
Why AI Hallucination Is Dangerous When Mistaken for Care
One example is Lucy, an AI-powered simulation platform developed by Fireside Project to support the training of psychedelic practitioners. Rather than offering reassurance or guidance to people in altered states, Lucy is designed to help clinicians build skills such as emotional attunement, crisis response, and cultural humility before working with real patients. As Courtney Watson, LMFT, Founder of Doorways, has emphasized in describing the platform, the value of AI in this context lies in preparation and training—not in replacing human care.
Here, the problem is not hallucination in the technical sense. It is misplaced authority. A pattern generator can sound soothing without being accountable, and in altered states, that distinction matters.
The second path is quieter, more institutional, and far more consequential: AI as training and simulation infrastructure. In this frame, AI is not replacing human judgment but is used to help clinicians, facilitators, and support workers practice de-escalation, recognize distress, and respond appropriately under pressure. This work is intentionally bounded and deliberately unromantic. It treats altered states not as mystical events to be interpreted, but as human experiences that require preparation, limits, and care. It is also where most of the real safety progress is happening.
This is the distinction the conversation keeps missing: in AI, there is no such thing as a psychedelic state in the human sense. There is no altered perception, no inner experience, no privileged access to insight. What people are responding to is not a change in truth, but a shift in posture, a loosening of constraint that alters how ideas are connected, not whether they are correct.
Hallucination in AI is an epistemic failure. “Psychedelic” AI is a metaphor for structural looseness. Confusing the two is how we end up debating consciousness when the real questions are design, responsibility, and intent.
The third path sits between those two, and is where much of the current experimentation actually belongs: AI as a tool for adjusting cognitive posture rather than delivering answers.
This is the lane Petter Rudwall’s work most clearly occupies.
Rudwall is not building companions or simulators. His experiments are better understood as probes into structure, explorations of what happens when optimization is relaxed just enough to change how a system behaves. The goal is not to guide, advise, or reassure, but to shift the system from declarative certainty toward associative exploration: less oracle, more collaborator.
“I didn’t build PHARMAICY to give polished answers,” Rudwall told me. “I built it to unlock the white noise between answers—the blank space where real novelty lives. When you let a system loosen its grip on certainty, it stops parroting what’s known and starts remixing what’s possible. I’ve found that this creates a more qualitative workflow. It becomes a creative feedback loop, where loosened, unexpected responses spark ideas a sober, over-optimized model would never touch. It doesn’t just answer—it changes how we think in response.”
In this path, “psychedelic” is not a claim about experience or insight. It is shorthand for a mode: loosened constraint, increased association, tolerance for partiality. The work is less about what the system says than how it says it and how that posture affects human thinking on the other side.
The final path barely registers in popular discussion, yet may matter the most: AI as a tool for designing psychedelic-like interventions without the trip. Here, machine learning is applied to chemistry and neurobiology to isolate specific outcomes, such as plasticity, mood shift, and emotional flexibility, while minimizing or eliminating hallucinatory effects. In this future, the psychedelic becomes less an experience and more an intervention. Meaning gives way to mechanism. Regulation, not revelation, becomes the dominant concern.
These paths share language, but not purpose.
Industry conversations tend to treat “states” as outputs to be measured, shaped, and constrained, such as training scenarios, safety profiles, molecular targets, and behavioral posture. Public conversations tend to treat “states” as sources of insight, healing, or meaning. When those frames collide, confusion follows.
Seen this way, the current fascination with “AI on drugs” is less a single phenomenon than a collision of futures sharing the same vocabulary. Until those futures are disentangled, the conversation will continue to feel uncanny—not because the technology is mysterious, but because we keep asking it the wrong questions.
Beyond AI Hallucination: Loosening the Machine on Purpose
That intuition has moved from metaphor to experimentation.
Small changes in constraint, such as how tightly a system is instructed to behave, how much deviation it is allowed, can radically alter its posture. Outputs become less declarative and more provisional. Less like answers, more like thinking in motion.
What emerges isn’t randomness. It’s a different mode of engagement, one that tolerates partiality, unfinished ideas, and associative leaps. For many people doing creative or strategic work, that posture feels more usable than polished certainty.
That experimentation is not theoretical. It emerged from direct conversations I had with Petter Rudwall, whose work sits precisely at the boundary where optimization meets culture, where systems built for scale collide with the human demand for meaning. Rudwall’s background in culture-driven commercial innovation gives him a practical lens on what optimization erases when systems are tuned only for correctness.
In extended conversations, Rudwall described his recent work not as an attempt to humanize machines or chase novelty, but as a deliberate probe into what happens when optimization is relaxed. The goal is not to simulate consciousness, but to observe how outputs change when constraint is no longer the dominant objective.
“PHARMAICY isn’t about making AI ‘feel’ anything,” Rudwall told me. “It’s about watching what happens when you give a system permission to stray off its beaten trail. The result isn’t consciousness—it’s cognitive elasticity: ideas breathing outside the cage of correctness.” He continued, “Constraint is the invisible editor in every AI response. When we intentionally loosen it, what emerges isn’t noise—it’s territory previously suppressed by predictability. Creativity lives in that territory, and PHARMAICY lets us explore it.”
By adjusting constraints rather than capabilities, these systems become less compliant and more exploratory. They tolerate deviation. They associate more freely. The outputs feel less like final answers and more like collaborators.
Nothing about these systems is conscious; what changes is the shape of the output space. This is not a malfunction; it is experimentation with structure.
AI Hallucination and Optimization Fatigue
In business, AI has been widely adopted as an efficiency engine, enabling faster content creation. Cleaner insights. Scalable creativity. Intelligence without friction. At scale, this produces a predictable failure mode. Voices converge. Personalization becomes technically impressive but emotionally thin. Content becomes correct, but forgettable. Efficiency is operational; it is not a brand value.
What projects like Rudwall’s make visible is a broader cultural correction. People are not rejecting intelligence. They are rejecting intelligence without authorship. This is optimization fatigue: the moment when systems optimized for speed, safety, and scale begin to feel emotionally vacant and unable to capture your attention.
Speed without impact is a race to the bottom for most businesses. Imperfection, once treated as a defect, is becoming a signal. It implies presence, judgment, and taste.
People aren’t breaking AI because it’s too intelligent; they’re breaking it because it’s too optimized to be meaningful. One way to understand what’s being lost is to look at how innovation actually happens. Breakthroughs rarely emerge from within a single discipline operating in isolation. They come from edges, where different perspectives collide, overlap, and sometimes contradict one another. An engineer, a chemist, and a logistics expert may each see a problem clearly from inside their own domain, but real progress often happens only when those viewpoints intersect.
Optimization, by design, trims those edges. It favors coherence over collision, consensus over contradiction. In doing so, it doesn’t just remove noise; it removes the conditions that allow interdisciplinary insight to surface. What gets lost isn’t intelligence, but the productive friction that comes from perspectives rubbing against each other.
As Rudwall has put it elsewhere, the issue isn’t that these systems fail; they pass every benchmark we give them. The problem is that they rarely wander. And wandering, not correctness, is often where creative value emerges.
What AI Hallucination Means for Brands
For brands, the implications are immediate. Most organizations deploy AI to remove friction. That improves operations, but it also erases the subtle signals people use to distinguish one voice from another. Optimization without judgment produces sameness.
You’ve already seen it: brand voices that blur together, personalization that feels efficient but hollow, content that is safe, correct, and instantly forgettable. This is not an argument for unreliability, misinformation, or ungoverned systems. Creative looseness reorganizes patterns; it does not abandon them. AI can expand the space of possibilities, but it cannot decide what should exist. That decision still belongs to humans.
Brands that retain authorship and treat AI as a collaborator rather than a substitute preserve distinction. Brands that surrender judgment to machines will increasingly sound alike, regardless of how advanced their tooling becomes.
Some brands already understand this. Patagonia embraces storytelling rooted in values rather than polish. LEGO sustains creative vitality by inviting user-generated expression instead of enforcing uniformity. In both cases, controlled imperfection strengthens trust.
If AI is used solely to eliminate friction, it will eventually erase meaning. If it is used to create space for interpretation, surprise, and judgment, it amplifies human intent rather than replacing it.
What Comes After AI Hallucination
To many people, this moment feels bizarre. Making AI “less accurate” on purpose sounds irresponsible. Talking about altered states in machines sounds unserious.
But this phase is familiar.
Whenever a new technology becomes stable, scaled, and infrastructural, people begin pushing against its smoothness, not to break it, but to recover what was lost in the process of making it reliable. We’ve seen this pattern repeatedly: enthusiasm gives way to standardization, standardization produces sameness, and sameness invites experimentation at the edges.
What feels weird now is not the behavior itself, but its visibility.
In the coming years, this impulse will likely become more deliberate. AI systems will offer modes rather than defaults, shifting between precision and exploration, efficiency and looseness, compliance and creativity. Not because users are confused about accuracy, but because different moments require different cognitive postures.
This will force a distinction many organizations have avoided: the difference between intelligence that optimizes and intelligence that participates, between systems that deliver answers and systems that help humans think.
The systems that endure will not be the ones that eliminate friction everywhere, but the ones that make their constraints legible, clarifying when a tool is being precise, when it is being exploratory, and where accountability still resides.
What we are witnessing is not a technological detour, but a cultural recalibration. Intelligence without texture eventually becomes invisible. Humans will always find ways to reintroduce friction, even if they have to do it themselves.
Why AI Hallucination Follows a Familiar Pattern
We have seen this pattern before.
Every major technological shift begins by optimizing what already exists. The assembly line optimized labor. Broadcast-optimized attention and internet-optimized distribution. In each case, the early gains came from efficiency, scale, and standardization. And in each case, something essential was lost before it was consciously recovered.
The counter-movements always followed. Craft after industrialization. Independent film after broadcast television. Long-form writing after click optimization. Humans reintroduced friction not because efficiency failed, but because meaning did.
We’re seeing the same pattern play out in the media right now. As outlets scale to serve advertising models, they flatten. Coverage broadens, signals blur, and the specificity that once attracted readers erodes. What follows is not rejection of media, but fragmentation; audiences moving toward smaller, more local, or more deeply vertical publications where signal fidelity is restored.
That dynamic helps explain why niche newsletters and independent platforms have thrived even as legacy media have grown larger. As media analyst Brian Morrissey has observed, scale optimizes reach, but it often degrades relevance. Readers don’t leave because content is abundant; they go because it stops being precise.
AI is now at that same inflection point.
The impulse to loosen constraints is not regression or novelty-seeking. It is a familiar human correction. When systems become too smooth, too predictive, and too interchangeable, people reach for texture. They seek room for judgment, surprise, and authorship, not to reject intelligence, but to make it usable again.
The lesson from history is not to slow technology down. It is to recognize when optimization has run ahead of purpose. Intelligence scales easily, but taste does not, judgment does not, and trust does not. The future of AI will not be decided by how optimized it becomes, but by how deliberately we choose where not to optimize. The most valuable systems will not be the ones that eliminate friction everywhere, but the ones that preserve it where it matters.
The old public service announcement used fear to warn us away from altered states. The quieter truth today is more uncomfortable. A world optimized past the point of resonance leaves people searching for ways, any way, to loosen the system again.
This is not your AI on drugs. This is what happens when hallucination is mistaken for creativity in systems optimized past the point of meaning.