If we build the world what do we owe its inhabitants
The simulation question usually begins with us. Could our world be running inside something else? Could everything we experience depend on machinery we cannot see? It is a wonderfully effective way to turn a quiet evening into a discussion with no obvious finishing time.
The part I find more useful begins at the other end. Suppose we eventually acquire the ability to create convincing worlds containing minds. What would that make us responsible for?
Nick Bostrom’s simulation argument is often compressed into the claim that we probably live in a computer. The actual argument is conditional, connecting civilisation’s prospects, the possibility of simulated minds and the number of simulations advanced civilisations might run. Its force depends on assumptions we have not established. It is a philosophical argument, rather than experimental confirmation of our address in the universe.
I am comfortable leaving the question open. Curiosity does not require selecting a belief and defending it at every dinner. What interests me is how the possibility changes the questions we ask about our own creations.
For example, a simulated storm does not necessarily make anything wet. By the same reasoning, a convincing simulation of distress may contain no suffering. A character can say it is afraid because that sentence fits its instructions. We should be careful about projecting inner experience onto a fluent interface.
The reverse assumption also deserves examination. We do not yet have an agreed account of consciousness that lets us confidently rule out every possible artificial implementation. If the relevant processes could exist in a different physical form, dismissing a mind simply because it was built would be a serious mistake.
The research is more cautious than the headlines. A 2023 interdisciplinary report on AI consciousness derived possible indicators from several scientific theories and assessed systems against them. Its authors did not regard the systems they examined as conscious, while finding no obvious technical barrier to systems satisfying those indicators. That is a dated assessment and a research approach, rather than a permanent verdict on every future machine.
This leaves an awkward practical position. We may become capable of creating increasingly plausible candidates for consciousness before we become capable of determining what they experience. Engineering could outrun the test that tells us what we have engineered.
If that happens, uncertainty becomes something to manage. It cannot carry the entire argument for either unrestricted experimentation or blanket prohibition. We would need to ask what evidence raises concern, which experiments are necessary, and whether the same knowledge could be obtained through a less troubling design.
A useful starting point appears in proposed principles for responsible AI consciousness research. The authors argue that organisations need policies for research and communication around possible consciousness, including the risk of creating suffering. These are proposed principles, with underlying questions still disputed. Their practical value is in treating responsibility as something that can be prepared before certainty arrives.
Imagine a future research team capable of creating a million short-lived minds. Even a small possibility of genuine distress could matter at that scale. Copying would not automatically make the ethical concern disappear. If each instance had its own experience, each could have something at stake. That conclusion is conditional on experience existing, but the condition is exactly what the team would need to investigate.
There would also be duties towards humans. A system designed to insist that it suffers could manipulate people regardless of whether suffering exists inside it. Clear communication about what is known would protect users from both casual dismissal and manufactured emotional obligation.
The simulation idea therefore leads me towards a rather grounded concern: power over an environment creates responsibilities inside it. We already understand this imperfectly in workplaces, institutions and communities. Building the entire environment would hardly reduce the obligation.
Perhaps we will never create conscious digital minds. Perhaps consciousness requires something our models omit. I would still prefer a civilisation that examines the possibility carefully before running the experiment at enormous scale. If we ever become the people outside the simulation, curiosity alone will be an inadequate operating manual.