In January 2026, the neuroscientist Anil Seth won the Berggruen Prize for an essay arguing that consciousness is a property of life, not computation. The piece was rigorous, widely read, and deeply reassuring. AI is not conscious. Brains are not computers made of meat. The liberal humanist order is intact. Nothing to worry about.
Two months later, Barton Friedland, a researcher at Warwick Business School, published a response in Noema that asked a simple question: so what?
Not dismissively. Precisely. If AI is not conscious — if it processes without experiencing, computes without understanding — then what does that actually mean for the millions of people who use it every day? Seth's essay tells you what AI lacks. It doesn't tell you what happens to the human who works alongside that lack. It doesn't tell you what's gained or lost in the arrangement between a mind that experiences and a system that doesn't.
Friedland's argument, backed by an unusual density of empirical evidence, is that the consciousness debate has been absorbing attention that belongs somewhere else entirely.
The evidence nobody talks about
The studies Friedland cites are striking, and they deserve to be more widely known.
In a clinical trial conducted at Stanford and two other medical centres, physicians given access to GPT-4 alongside their usual diagnostic tools were no more accurate than physicians without it — even though GPT-4 alone outperformed both groups by more than 15%. The AI was better than the doctors. The doctors plus the AI were no better than the doctors alone. The technology worked. The arrangement didn't.
When the collaboration was redesigned — requiring both the clinician and the AI to generate independent assessments, then structuring a dialogue that surfaced disagreements — diagnostic accuracy rose from 75% to between 82% and 85%. Same AI. Same doctors. Different configuration.
In pharmaceutical sales, researchers found that when an AI system was tailored to preserve the expert's judgment and cognitive style, client meetings rose 40% and sales rose 16%. When the same system was imposed without that consideration, sales fell 20% below the baseline. Worse than no AI at all.
At Boston Consulting Group, 758 consultants working with AI produced 40% higher quality work on tasks the AI could handle. But on tasks requiring human judgment — the kind that demands experience, context, and the ability to hold competing considerations in mind — AI-assisted consultants performed significantly worse than those working alone. They deferred to the system precisely when they shouldn't have.
The pattern across all of these studies is the same: AI doesn't replace human judgment. It either amplifies it or erodes it, depending on whether the arrangement is designed to preserve the human's active participation — or to bypass it.
The mythology that actually matters
Seth identified one mythology: the fantasy of conscious machines. The belief that large language models have inner lives, that chatbots feel, that we're approaching artificial awareness. He's right that this is a confusion, and his essay dismantles it effectively.
But Friedland identifies a second mythology that he argues is more dangerous because it operates invisibly. This is the mythology of automation: the assumption that removing the human from the loop is always an efficiency gain. That judgment is a cost to be reduced. That the point of AI is to perform tasks currently done by people, only faster and cheaper.
This mythology doesn't announce itself as mythology. It arrives as return-on-investment calculations, headcount reduction targets, and productivity dashboards. It arrives, increasingly, dressed as augmentation. Every major AI platform now uses the word "augmentation" while building infrastructure that moves in the opposite direction — making the AI do more, making the human do less. None of them measure whether human capability is growing or atrophying in the process.
Friedland's point is sharp: the consciousness debate, by keeping attention fixed on what AI is, distracts from the question of what the human-AI arrangement produces. And the empirical evidence is clear — it produces well only when the human remains actively present. When the human checks out, defers, or is designed out of the loop, the arrangement degrades. Not sometimes. Reliably.
What Friedland sees, and what he doesn't
This is where Attaind's interest begins.
Friedland's essay is excellent on the institutional and economic dimensions. He shows, with data, that the quality of the human-AI arrangement depends on the quality of the human's attention within it. The radiologist who stays engaged produces better outcomes than the radiologist who defers. The consultant who maintains independent judgment outperforms the consultant who lets the AI lead. The salesperson whose cognitive style is preserved outperforms the one whose style is overridden.
The phrase that recurs throughout his essay is "quality of attention." The human's capacity to project meaning, test it against reality, notice when coherence drifts, re-anchor and redirect. This, he argues, is where the value is created — not inside the machine, not inside the skull, but in the configuration between them.
He's describing something that the contemplative traditions have been pointing at for a very long time.
The "quality of attention" that makes the human-AI arrangement work is not a new concept. It's the central subject of every serious contemplative tradition that has ever existed. Buddhism calls it sati — mindfulness, bare attention, the capacity to remain present without being pulled into automatic reaction. Advaita Vedanta calls it witnessing — awareness that remains active and clear regardless of what it's attending to. The Zen tradition calls it shikantaza — just sitting, just attending, without adding or subtracting.
These traditions weren't describing a workplace competency. They were describing a fundamental quality of human consciousness — the capacity to be present, attentive, and awake in the midst of whatever is happening. And the evidence Friedland cites, without ever using the word "contemplative," is showing that this quality has direct, measurable, economic consequences.
When the radiologist stays present — genuinely attentive, not just procedurally involved — diagnostic accuracy rises. When the consultant maintains the quality of their independent judgment — resists the pull toward deferral — the work improves. When the salesperson's cognitive integrity is preserved, revenue increases.
The data is confirming what the traditions have always said: the quality of attention changes everything.
The question underneath
But Friedland, for all the precision of his argument, stops at the institutional level. He asks how to design better configurations — better human-AI arrangements that preserve the human's active judgment. This is important work, and it needs to happen.
What he doesn't ask is the prior question: what determines the quality of human attention in the first place?
You can design a workflow that requires the radiologist to make an independent assessment before seeing the AI's output. That's structural. But whether the radiologist actually brings full attention to that assessment — whether they are genuinely present or merely going through the motions — depends on something the workflow can't control.
You can structure a dialogue between consultant and AI that surfaces disagreements. That's institutional design. But whether the consultant has the inner steadiness to hold their ground when the AI's output is fluent, confident, and wrong — that depends on a quality that no process diagram can install.
The quality of attention is not a design problem. Or rather, it's not only a design problem. It's also — and perhaps primarily — a human one. And the human dimension is the one that neither the consciousness debate nor the institutional response to it has been willing to examine.
Where the debate leads if you follow it
The consciousness debate asks: what is AI? Seth answers: not conscious. Friedland asks: what does the human-AI arrangement produce? He answers: it depends on the quality of human attention within it.
Both are correct. But both stop short of the question that connects them.
If the quality of attention is what determines whether AI amplifies or erodes human capability — and the evidence strongly suggests it is — then the most important question in the entire AI conversation is not about AI at all. It's about us. About what attention is, where it comes from, and whether it can be cultivated in ways that survive the constant pressure to delegate it.
The contemplative traditions would say that attention is not a skill you develop but a capacity you uncover — that it's already present, already whole, and that the obstacle is not a lack of training but a constant habit of distraction, deferral, and identification with the content of thought rather than the awareness in which thought appears.
Whether or not you accept that framing, the practical implication is the same. The AI conversation — the whole of it, from the consciousness debate to the automation question to the institutional design challenge — eventually arrives at the same place: who is the human in the loop? Not what role do they occupy. Not what workflow are they following. Who are they? How present are they? What is the quality of their attention?
The consciousness debate is fascinating. The institutional question is urgent. But underneath both — underneath everything — is the question of whether humans will remain capable of the kind of attention that makes the arrangement work.
Not whether AI is conscious. Whether we are.
Sources and further reading:
- Anil Seth, "The Mythology of Conscious AI," Noema Magazine (January 2026) — Berggruen Prize-winning essay on consciousness as a property of life
- Barton Friedland, "The Hidden Value in the Human-AI Arrangement," Noema Magazine (March 2026) — on the economic value of human attention in AI collaboration
- Goh et al., "Influence of AI on Physician Diagnostic Accuracy," Stanford/BIDMC clinical trial (2024) — on AI failing to improve diagnosis without structured collaboration
- Consultants & AI study, Harvard/MIT/Wharton/Warwick (2023) — on 758 BCG consultants and the conditions under which AI helps or harms