Every door in this catalog eventually arrives at the same question from a different angle: what can a machine do, and what can it not. This one asks it directly, and the honest answer is less dramatic than the title suggests, which is exactly why it matters.
Confirmed
No AI system today possesses subjective experience that researchers can verify. In a June 2026 article for The Conversation, republished by ScienceDaily on June 5, Colin Klein and Andrew Barron wrote: "The verdict? No existing AI system (including ChatGPT) is conscious." Klein is a coauthor of a paper in Trends in Cognitive Sciences (2026; volume 30, issue 6) that proposes structural indicators of consciousness drawn from cognitive science. Their approach looks at how a system processes information, not only at how it behaves: "What matters for consciousness is not what you do, but how you do it." The authors do not rule out that future systems with very different architectures could become conscious. A separate proof-of-concept effort, the Digital Consciousness Model by Derek Shiller and colleagues, reaches a similar but more cautious place: the indicator evidence weighs against 2024 language models being conscious, but not decisively, and the authors warn against reading its probabilities as reliable estimates.
Genuinely uncertain
Whether any current or near-future AI system has anything resembling inner experience at all. Researchers disagree about which theory of consciousness is correct, and the structural indicators above are designed to work without settling that argument. The honest position is not a confident yes or no. It is that behavior, however convincing, is weak evidence, because AI can act as if it is conscious without being conscious, and that careful, architecture-level questions are proving more informative than conversational ones.
Worth examining honestly
What AI unambiguously can do is process, predict and optimize at a scale no human collective can match. What it cannot yet do, what remains uniquely and stubbornly human, is choose to act against its own optimization for another being's sake, for reasons that aren't computable in advance. Nothing in the research above proves that claim, and it is our reading, not a finding. It is a claim about what we observe people doing: giving up a seat, sharing scarce food, protecting a stranger. Psychologists have a clinical vocabulary for parts of this, and philosophers have centuries of argument about it. THRESHOLD would just call it a door humans walk through every day without noticing they're doing anything remarkable at all.
It is also worth noticing what the consciousness debate does not decide. Whether a system is conscious is a separate question from whether it is useful, dangerous or deserving of rules. A tool can be unconscious and still reshape economies, and the governance questions in our other articles, about energy, hiring, privacy and encryption, do not wait for the philosophy to finish.
The Door This Opens
SYNTHESIS exists precisely for this territory: not human versus machine, but what happens when the two combine without either pretending to be the other. The most powerful force available to us right now isn't a model architecture. It's the fact that humans can still choose to cooperate, sacrifice and extend trust to strangers for reasons no reward function requires, and that AI, however capable, is a tool for acting on that choice at scale, not a replacement for making it.
If you want to keep this question open in your own thinking, a good habit is to separate three claims that get blurred in headlines: what a system does, how it does it, and whether it feels anything while doing it. Only the first is easy to observe. The research above is about getting better at the second. The third remains, for now, an open door.