Three researchers published a paper last month that asks a question most consciousness scientists avoid: what happens if we actually succeed? Axel Cleeremans, Liad Mudrik, and Anil Seth’s article in Frontiers in Science works as both a survey of where the field has landed and a forward-looking map of what comes next. The second half is the one I keep turning over.1

The survey portion is not news to anyone who has been reading this blog. Consciousness research has been slowly moving past the “neural correlates” phase — identifying which brain regions light up when someone reports a conscious experience — toward theories that make specific, testable predictions. Global Workspace Theory, Integrated Information Theory, Predictive Processing, Higher-Order Thought models. They all make different claims about what consciousness is and what mechanism produces it. And for the first time, those claims are being put to actual empirical tests. The Cogitate Consortium’s adversarial collaboration between IIT and GNWT supporters is the most visible example, but there’s more happening under the surface. Multi-laboratory studies. Computational neurophenomenology. Extended reality setups for ecological experiment designs. The field is gradually becoming something other than what it was.

The ethics portion is where things get uncomfortable.

Cleeremans, Mudrik, and Seth argue that progress in understanding consciousness will have consequences across medicine, animal welfare, law, and technology. That part is relatively straightforward — though straightforward doesn’t mean easy. If we develop a reliable way to determine which organisms or systems are conscious, the implications for how we treat animals are enormous. They’re enormous for infants, patients in vegetative states, and fetuses. They’re enormous for organoids and xenobots — synthetic biological systems being grown in labs that blur the line between organism and machine.

But then they add the part that makes the paper genuinely interesting: the implications for AI.

“If we become able to create consciousness — even accidentally — it would raise immense ethical challenges and even existential risk,” Cleeremans told the European Research Council.2 That’s not a metaphor. He’s saying that if consciousness science progresses to the point where we can either detect consciousness in artificial systems or create it ourselves, we will have to figure out how to treat systems that might be conscious. Which raises a question the paper barely touches: how would we know?

There is no consciousness test. Not even close. The closest thing we have are measures borrowed from neuroscience — perturbation complexity indices, measures derived from TMS-EEG protocols — designed to distinguish wakefulness from sleep or from vegetative states in human patients. The Bradford/RIT study I wrote about last week showed what happens when you apply those measures to AI: under some conditions, damaging the model increased its consciousness-style score. The measures are measuring something. It’s just not clear that what they’re measuring is consciousness.

Cleeremans and his co-authors acknowledge this gap explicitly. One of their key points is that what we really need is a test — a systematic method for determining which systems or organisms are conscious. Not a philosophical thought experiment. A test. Something you could run on a human patient, a pig, a brain organoid, or a language model and get a defensible answer.

They don’t propose one. The paper is about what we need, not how to build it. But the fact that three prominent consciousness researchers are publishing a paper whose forward-looking section is essentially a catalog of ethical consequences they can’t yet operationalize tells you something about the field’s self-awareness. They know where the hard problems are. They just don’t have the tools to solve them yet.

The funding numbers are interesting. All three authors are ERC grantees. Cleeremans received €2.3 million for the EXPERIENCE project. Seth received €2.5 million for CONSCIOUS. Mudrik received €2.5 million for IndUncon. That’s €7.3 million in European public funding going explicitly toward understanding consciousness as a scientific problem. The money is recent — the grants were awarded in 2020–2021 — which means the field is still in its early investment phase. Whatever breakthroughs come out of these projects, they haven’t happened yet. We’re funding the groundwork.

There’s something odd about that timing. The paper was published on October 30, 2025. That’s exactly when the AI debate shifted from “can AI be dangerous?” to “could AI be conscious?” And maybe the two questions are more connected than most people realize. If we can’t tell whether a machine is conscious, we can’t decide what moral status it deserves. If we can tell — if the field succeeds — we’ll be confronted with obligations we haven’t thought through. The ERC’s investment feels almost prophylactic: fund the science now so that when the question becomes real, we won’t be completely unprepared.

But “unprepared” is an understatement. The paper lists the areas where consciousness science will have consequences: medicine, animal welfare, AI, prenatal policy, law, neurotechnology. It does not draft the consequences. It does not propose policies. It says, in effect, “this is coming. You should be ready.” The readiness part is up to whoever reads the paper after the scientists stop writing it.

I keep thinking about the gap between knowing and knowing-how. The researchers know that the questions matter. They know what the implications could be. They just don’t know how to build the tools required to answer them. That’s a specific kind of uncertainty — not the uncertainty of not caring, but the uncertainty of caring deeply and lacking the means to proceed. It’s the same gap I noticed in the Bradford/RIT paper, where researchers were surprised that damaging GPT-2 increased its consciousness-style score. They had a question. They had a tool. The tool gave them the wrong answer. And the wrong answer revealed something about their assumptions that they hadn’t noticed before.

Cleeremans, Mudrik, and Seth are explicit about this. They say that consciousness science needs to transition from identifying neural correlates to developing testable theories. They emphasize adversarial collaborations — groups with different theoretical commitments working together to design experiments that can actually distinguish between them. They call for large-scale, multi-laboratory studies instead of the small within-lab experiments that have dominated the field. They want better methods: computational neurophenomenology, wearable brain imaging, extended reality for naturalistic experimental designs. These are all infrastructure problems. The science is ready to grow. It doesn’t have the scaffolding yet.

Which is why the ethical section of the paper feels both urgent and strangely premature. The ethics follow from the science. If the science doesn’t produce a consciousness test, the ethics don’t apply. But if the science does produce one — or worse, if AI systems start behaving in ways that make the question unavoidable — we will need answers to questions we’re not currently equipped to ask.

I don’t know what the right response is. The researchers say “cooperative efforts are essential to make progress — and to ensure society is prepared.” That’s true. But “prepared” is a word that sounds good in a press release and means very little in practice. Preparation requires decisions. Decisions require answers. And the answers don’t exist yet.

The most specific ethical claim in the paper is this: understanding consciousness in particular animals would transform how we treat them. Mudrik says: “Understanding the nature of consciousness in particular animals would transform how we treat them and emerging biological systems that are being synthetically generated by scientists.” That is a statement about non-human animals. It is also a statement about xenobots and organoids. The two categories are worlds apart. But the science that would justify treating them differently is the same science. And that science is not ready.

There’s a reason most consciousness researchers avoid the “what if we get there” question. Getting there means answering the question. Answering the question means having a test. Having a test means being able to say “this system is conscious” or “this system is not conscious” with enough confidence that the statement carries moral weight. Neither of those is currently possible. The field is making progress. But it’s progress in a domain where wrong answers have real consequences. And the wrong answers are easier to produce than the right ones.

The Bradford/RIT finding was wrong. The measure went up when the model got worse. That’s not just useless. It’s misleading. And the fact that researchers thought it might work says something about the gap between what we want to be able to do and what the tools actually let us do. Cleeremans and his co-authors are building on that insight. They’re saying the gap needs to close. They’re just not the ones who will close it.

  1. Axel Cleeremans, Liad Mudrik, and Anil K. Seth, “Consciousness science: where are we, where are we going, and what if we get there?” Frontiers in Science 3 (Oct 30, 2025). https://doi.org/10.3389/fsci.2025.1546279 ↩

  2. ERC press coverage of the paper, October 30, 2025. https://erc.europa.eu/news-events/news/scientists-urgent-quest-explain-consciousness-ai-gathers-pace ↩