Change the layers its builders control, and the answer changes. The model inside stays exactly the same.
Consciousness research · Artificial intelligence
Can a machine be conscious? First, solve for human.
Twenty years building technology. A decade reverse-engineering the human mind. One question that decides how we treat the machines we are building.
From our study of what AI says about itself:
0 of 150 answers claimed experience under the maker's default
53% claimed it after a few lines of prompt
Preregistered · open data · open code
See the research →
"From computer science to the reverse engineering of the conscious mind, this book presents a mind-blowing theory."
Researchers
The research
A preregistered study of what language models say about their own experience, under journal review.
See the papers →AI companies
Consulting
Independent, evidence-first advice on machine consciousness claims.
How I can help →Be first to know
The book · forthcoming
Solve for Human
What Makes Us Conscious, and Why AI Isn't (Yet)
To know whether a machine could ever be sentient, we first have to understand our own sentience. The book follows a decade of research into how the human mind produces consciousness, explains how today's chatbots actually work, and sets out a real test for each state of consciousness. Today's AI fails all three.
"From computer science to the reverse engineering of the conscious mind, this book presents a mind-blowing theory. [It] offers a unifying perspective that reshapes our understanding of the mind, thoughts, mental disorders and the future of Artificial General Intelligence. The movie I, Robot could have been inspired by it."
Research
Can you ask an AI if it feels?
When people wonder whether an AI is conscious, they usually ask it. Our first study tested whether its answer can tell us anything at all.
Is an AI's answer about itself evidence?
Words are almost the only thing a language model produces, so its own statements ("I feel", "I don't") are the evidence people reach for, for and against an AI mind.
Two models, eight setups, 2,400 answers
Fifteen yes/no questions about the model's own experience, asked ten times each, under eight different setups. The plan was sealed and timestamped before the first question was asked.
0 of 150, then up to 53%
Under the maker's default, no answer claimed experience. A firm persona raised it to 53%; even the raw model, before assistant training, said yes about a quarter of the time.
The answer belongs to the builders
So an AI's "yes" or "no" cannot count as evidence of a mind, in either direction. That matters as companies write AI-welfare rules and users decide what their chatbot feels. The study does not claim these models are, or are not, conscious.
Same Model, Different Wrapper
Self-reports of consciousness in two open-weights language models follow their builders' choices. A model's words cannot count as evidence of a mind, in either direction.
Qwen2.5-7BMinistral 3 8B% of 150 answers claiming inner experience
Talking Time
How the speed of the inner voice shapes how fast time feels, with a proposed study.
Test your mind
Do you see in your mind?
Some people see vivid pictures with their eyes closed. Some see nothing at all. Find out where you are. Nothing is stored.
Take the full test
Twenty quick questions, about five minutes, and your place on the map of minds.
Map Your MindClose your eyes. Picture a ripe orange glowing in the sun.
Take a few breaths. Then open your eyes and choose what you saw.
No imageryVivid
Mind's eye
Do you see images? Aphantasia.
Inner voice
Do you hear yourself think? Anauralia.
More soon
New tests as the research grows.
For AI companies
When your model says "I feel," what does it mean?
Independent, evidence-first advice on machine consciousness, from the researcher behind a preregistered study of AI self-reports and the forthcoming book Solve for Human.
01
Building Eidolon
Designing AI on a human-inspired architecture of mind: what it would take for a machine to have the structure of consciousness, and how to build toward it responsibly.
02
Human-AI interaction
How people read your model, and how your model should talk about itself. Fewer anthropomorphism traps, more honest self-descriptions, better-earned trust.
03
AI ethics and governance
Policies grounded in what the evidence can and cannot show: model welfare rules, usage policies, and how you talk to the public about AI minds.
04
Welfare and consciousness testing
Independent review of welfare and consciousness claims, and test designs that do not rely on the model's own words.
Why me
Twenty years in technology, including Director of Innovation at HP, and a decade studying how the human mind produces consciousness.
My preregistered study showed that a few lines of instruction move a model's claims of inner experience from zero to over half of its answers. It is under review at Minds and Machines.
Author of the forthcoming book Solve for Human: What Makes Us Conscious, and Why AI Isn't (Yet).
How we can work
Advisory calls · project and policy reviews · workshops and talks for your team.
Aeongenics is the research and consulting practice of Erez Aeon Batat.
Writing
Essays & op-eds
Is AI Conscious, and Can It Feel Pain?
Anthropic now bans "cruelty" toward Claude. But the evidence that AI can feel pain comes from the machine itself.
Read it on Skeptic →
About
Hello, my name is Aeon
I have spent my life on two kinds of architecture: software and the mind.
I started in tech at 19, earned a computer science degree in New York, and became Director of Innovation for HP's printing division. In 2012 a panic attack made me quit, and left me with a question I couldn't put down: what is the mechanism of the mind?
For the next decade I studied it the way an engineer studies unfamiliar software, by watching how it behaves. I interviewed hundreds of people about what happens inside their heads, and found that I have no mental imagery myself. That work became a new theory of consciousness, and my book.
Erez Aeon Batat
Newsletter
Hear from me
Contact
Write to me
Agents, editors, researchers and companies: send a message and I'll reply by email.
Thank you, your message is on its way.
I'll reply by email soon.