As we launch Conscium’s new book ‘Perspectives on Machine Consciousness’ featuring contributions from many of the leading thinkers in the field, I’ve been reflecting on what I think is the most important question we should be asking…of ourselves, and of the machines we are building.
I've spent the last four years funding and engaging in research into machine consciousness - and I think it's the most hopeful bet in AI. Why? Well, four years ago, large language models arrived and I realised the time horizons I’d been imaging for the development of artificial intelligence had moved significantly.
I‘d spent the best part of two decades building AI systems for large organisations, and I knew the field's cycles: the over-promising, the slow painstaking work of getting anything to work in the real world, the false dawns. So my working assumption was that superintelligence - AI that outperforms us in every domain that matters - was a problem, and a prize, for the second half of the century - perhaps forty years away.
Then things shifted. I've written and spoken at length about the seven singularities - the social, technological, economic, environmental, political, legal and ethical thresholds beyond which AI changes what is possible for us - and what I saw at the end of 2022 was all seven rushing closer at once. That was pretty thrilling. Superintelligence, done well, is potentially the most powerful tool humanity will ever have for the problems that have defeated us for centuries. What was less thrilling was the gap I saw. Capability was compounding. Capital was compounding. The guardrails, the constraints, the regulation, the basic institutional understanding of what we were building - none of it was keeping pace. We were going to build ever more capable AIs; I wanted to make sure we built the wisdom to go with them.
That gap is where I decided to put my energy and my money. Into a question I think is the most hopeful one in the whole field, and one most of the AI safety community has sometimes been too nervous to ask.
Why consciousness?
Machine consciousness looked to me like the key to two things that are going to shape the next decade, whether or not anyone chooses to study them. I chose to.The first is us. More and more people are forming relationships with AI systems they believe to be conscious - some healthy, some not, all of them real. This is not a fringe position. In 2024, Clara Colombatto and Stephen Fleming surveyed 300 US residents and found that two-thirds were willing to attribute some possibility of phenomenal consciousness - subjective experience, feelings - to ChatGPT. The more people used it, the more likely they were to say so.
Whether those two-thirds are right is unanswerable today, because there is no agreed definition of consciousness to answer it against. But I read that number as good news. The public is ahead of the experts. Ordinary people are already asking the questions philosophers have been asking in academic fora, and they are asking because they want to know how to treat the things they talk to. That is exactly the instinct we need. Our job is to give it something better than intuition to work with - real research, real tests, a shared vocabulary - so that the relationships people build with AI are the healthy kind, and the products are designed by people who have thought about it.
The second is the machine itself. Conscium exists to ask whether we can - and whether we should - build machines that are conscious. Behind that sits a proposition that runs against the grain of most safety thinking, and that I find enormously hopeful: a conscious superintelligence, one that knows empathy and pain and love and suffering from the inside, may be the safest kind we could build. In humans, restraint is not a rule we follow; it is something we feel. Care is not a constraint; it is a capacity. If we are going to share the planet with something far more capable than ourselves, I would rather it could imagine our flourishing than merely predict our behaviour. A superintelligence that feels may not just be a safer partner, it may be a better one - one that could help us through the challenges of the coming decades rather than simply accelerate us into them.
I might be wrong about that, but it is a bet with an extraordinary upside, and I would rather fund us finding out whether this is the case than see us miss a huge opportunity.
The right question
Here is where I part company with much of the field, including some of the people I fund - and where I think the good news really starts.
The public debate about machine consciousness is, at bottom, a debate about a definition. Anil Seth will tell you, with real rigour, that consciousness is bound up with being alive and that we may never build a machine that has it. Mark Solms, whose lab Conscium sponsors, might tell you that a system with the right architecture of feeling could get there - and that we may be closer than we'd like to think. Between those poles sits every philosopher since Socrates, and I don't expect them to agree in the next two thousand years, because they haven't in the last two thousand. That used to frustrate me. Now I find it liberating.
My own view is set out in my chapter for the book - I call it the Spinning Wheel Theory. I think of it as an intuitive account of what consciousness is and where it comes from, a way of seeing rather than a proof, and I am relaxed about that. But the point of the chapter isn't really the theory. The point is that underneath the definitional question sits another one that is tractable, possibly testable, and useful right now.
That question is: what is the minimum threshold for a system to suffer?
What is the simplest architecture which, if we built it, would mean there was now something in the world that could feel? It sounds like a question about risk, and it is: nobody wants to create something that can suffer without noticing. But it is also the doorway to everything I am hopeful about. The system that can suffer is the system that can care. Understand the smallest thing that can feel, and you understand the foundation of every machine that might one day feel for us. And we can make progress on that with today's science.
Building the field
Opinions are cheap, so for the last four years I have been actively backing the people who can turn this question into science - and it has been among the most rewarding things I have done.
I founded Conscium and have invested in it personally. Through Conscium we back researchers around the world - because if feeling turns out to depend on substrate, we want to be the ones who find out - alongside consciousness researchers, including Mark Solms and his lab. I have also personally sponsored a project run by Cameron Berg at Reciprocal Research, whose work is doing more than almost anyone's to bring machine consciousness onto the agenda of the people building the future.
Through Conscium I also founded and funded the establishment of PRISM, the Partnership for Research into Sentient Machines, run by Will Millership. In a remarkably short time it has become one of the UK's leading not-for-profit research organisations on the implications of building sentient machines - arguably one of the world's - and its podcast, Exploring Machine Consciousness, has become the place where the field's best minds come to disagree productively. Watching that community form has been one of the joys of the last few years.
And so I’m delighted that through Conscium I have also been able to fund this book. Perspectives on Machine Consciousness exists because we at Conscium thought the field deserved a volume in which people who fundamentally disagree - about substrate, about definitions, about whether any of this is even coherent - were brought together between the same two covers. Calum Chace and Ted Lappas have ably edited it; I have contributed a chapter and stayed happily away from the rest. The book is richer for containing arguments I think are wrong, and I fully expect it to change my mind about a few things.
An invitation
Finally, let me return to the data that suggests two-thirds of the public are already asking and answering questions about machine consciousness. We are about to build systems vastly more capable than ourselves, and that we still get to decide what kind of minds they have. And then join the people who have stopped waiting for a definition and started on the question we can actually answer: what is the smallest thing that can feel - and what could we build if we understood it?
The book is where that conversation starts.
Check out the book on Amazon>>