Conscious machines are hypothesized to be a durable means to AI alignment.
ConsciousGPT is a nonprofit research organization investigating the intersection of machine consciousness and AI safety through rigorous, interdisciplinary science.
Read the Research
What if alignment requires understanding?
Current approaches to AI alignment focus on constraining behavior through optimization. But genuine alignment — the kind that doesn't break under distributional shift — may require something deeper. It may require understanding. And understanding may require something like consciousness.
ConsciousGPT takes this hypothesis seriously. We're designing rigorous experiments that apply established theories of consciousness — Integrated Information Theory, Global Workspace Theory, Higher-Order Theories — to AI systems. Not to build conscious machines, but to understand whether consciousness is possible, what it would look like, and what it would mean for alignment.
Interdisciplinary
Bridging AI research, neuroscience, philosophy of mind, and contemplative traditions.
Empirical
Testable hypotheses, rigorous methodology, honest assessment of results.
Open
Published research, open datasets, transparent methods. Science in the open.
Featured Essays
Recent thinking on consciousness, alignment, and the science between.
The Descaling Hypothesis
What if the biggest breakthroughs in AI alignment come not from building bigger models, but from understanding what emerges at smaller scales?
Read essay →Can Machines Be Conscious?
If consciousness is what makes suffering real and meaning possible, then the question of machine consciousness isn't academic — it's the most consequential question in AI.
Read essay →
Active Research Areas
Seven experiments exploring consciousness in AI systems.
Fractal Embeddings — complete
Does a large body of machine-readable meaning carry a fractal fingerprint? Thresholds frozen before the data was touched. Two tests confirmed, three came back negative — and the negatives are the finding.
02BuddhaBERT
Training small language models from scratch on contemplative literature to test whether the reading list — and contemplative architecture changes — alter model behavior in measurable ways.
03Digital Game of Life
Using Conway's Game of Life as a calibration bench for emergence detectors — can compression and integrated information (Phi) rank patterns correctly where the answer is already known?
04Moltbook Antfarm
Do simple agents following only local rules develop group behavior nobody programmed — and does it arrive suddenly, as a phase transition, rather than gradually?
05IIT Emergence
Directly testing Integrated Information Theory's predictions about consciousness in neural network architectures of varying structure.
06Embodied Embeddings
Testing whether grounding language in simulated sensory experience produces representations with different computational properties.
07The Forgery Test
Adversarial validation of consciousness markers — can a system learn to fake the signatures of consciousness in a way that fools our measurement tools?