Almost every debate about artificial intelligence contains a sentence nobody spells out, because it is too comfortable to be examined. It runs roughly like this: later there will be an instance that can judge this better than we can.

The sentence looks harmless. It is not. It changes what we treat as urgent today. Anyone who believes a cleverer instance will resolve the hard questions anyway treats their own decision as an interim step. You commit provisionally, keep the matter open, wait for better knowledge. In many areas this is right. In some it is a mistake. Which case applies is not something you can read off the surface.

So I played out a line of thought that has not left me since. Suppose that in two hundred years there exists an intelligence as far ahead of us as we are ahead of an insect. It commands computing power on a scale we have no feel for. Which of our present questions will it answer? And which will it not?

What it will genuinely resolve

I do not want to diminish the answer before I give it. This instance will resolve a great deal that looks intractable to us now.

It will answer questions we cannot even pose, because we lack the concepts. It will design experiments no human would conceive. Take the unification of quantum physics and gravity. Today it fails on energies that would require an accelerator of galactic size. The likeliest resolution is not that accelerator. It is an indirect route we have simply overlooked. That is precisely its strength. It sees detours where we see walls.

So anyone expecting a catalogue of eternal riddles here will be disappointed. The list is short. What stands on it, however, does not shrink under cleverness.

The letter that never arrives

Imagine writing to someone who is moving away from you faster than the mail can travel. No better sender helps, no faster vehicle, no greater effort. The letter does not arrive, because the distance between you grows while it is on its way.

This is exactly the case for a large part of the universe. The expansion is accelerating. Beyond a certain distance, galaxies recede so quickly that their light will never reach us. Not today, not in two hundred years, not in a billion years. What happens out there is not a difficult observation. It is no observation at all. An intelligence of any resolution faces the same empty channel we do. The signal is not faint. It is absent.

The second case sits closer, because it concerns us directly. How life began on this planet is among the largest open questions we have. The data set for it is empty. The chemistry that turned dead matter into something self-sustaining, nearly four billion years ago, left nothing behind. No rock, no molecule, no trace. A superintelligence will be able to show us which pathways work chemically, probably several. Which one was actually taken, it will not be able to say. The evidence is not lost. It never existed.

That is the first category. It fails not on intellect, but on the missing signal.

Two maps, the same route

Take two maps of the same terrain. For every path you can actually walk, both give the same distance, the same duration, the same gradient. They differ only beyond the edge of the map, where nobody ever goes. Which map is correct? The question sounds answerable. It is not, as long as no walk exists that separates the two.

Physics has this case for real. The competing interpretations of quantum mechanics predict the same outcome for every performable experiment. This is not an interim state of research but a property of their construction. Whether the world branches at every measurement or a hidden order lies beneath it makes no difference at any instrument. An instance with unlimited computing time works through both interpretations and arrives twice at the same result.

The second case goes further, because here a proof exists. A little over ten years ago, mathematicians examined a physical property of certain material systems, the so-called spectral gap. Their finding: there is no procedure that decides, for every such system, whether the property holds. Not that we do not yet know. It is proven that none can exist, just as no program can predict of all other programs whether they will ever halt. This is not a question of scientific maturity. It is part of mathematics itself. And mathematics applies to machines exactly as it applies to us.

What only duration answers

The third category is the least spectacular. Everyone knows it from daily life. Whether a marriage lasts, you learn afterwards. Whether a structure holds under load shows up under load. Some answers are not knowledge but a course of events. And a course of events cannot be brought forward.

Whether a civilisation survives its own power belongs here. Of this process we have exactly one case, our own, and it is running now. There is no control group, no second run, no way to repeat the experiment under different initial conditions. A superintelligence will advise us superbly. It will see patterns we cannot. How it ends, it will not know, because that knowledge comes into being only with the ending.

A question on which any serious settlement planning depends sits close by. Whether a mammal produces a generation under the gravity of Mars is not a calculation. It is a measurement that takes as long as generations take. What can be accelerated is the analysis, the preparation, the choice of methods. Time itself cannot be accelerated. Anyone planning cities before that measurement exists is planning not with knowledge but with hope.

Why more machines change none of this

At this point an objection reliably arrives: all of that holds only while computing power is finite, and with quantum machines in vast numbers a new order of magnitude appears. I regard this objection as the most instructive point of the whole argument, because it is precisely wrong.

Quantum computers are not a general accelerator but a specialist tool. They are strong at simulating molecules and materials, at certain mathematical decompositions, at a narrow class of further tasks. For general searching through possibilities they deliver a quadratic improvement. Astronomically many steps thereby become astronomically many steps still. A billion such machines are a billion machines with the same narrow advantage. A different organ of thought does not emerge from that.

Behind this lies a physical ceiling that holds for every design. How many state changes a machine manages per second is bounded by the energy it uses. How much information fits into a volume of space is bounded by its size. A computer with the mass of a planet has a finite upper limit that can be calculated. It lies unimaginably far above anything we have. A limit it remains.

For our three categories this is beside the point anyway. A signal that never arrives is not received through computing power. Two maps that predict the same thing are not separated by computing power. A course of events that takes decades is not shortened by computing power. In none of these cases was intelligence the bottleneck. That is why more of it does not help.

What remains I call the hard remainder. It is the share of a question that cleverness does not reduce, because its obstacle lies elsewhere. The hard remainder is small. It is large exactly where the irreversible begins.

The state it will inherit

At this point the view turns around, and the argument becomes uncomfortable.

This future instance can do almost everything better than we can. One thing it cannot do: enter the time before its own existence. It inherits our world in the condition we hand over. Whatever is fixed irreversibly in the coming decades, it finds as a given, not as a task.

Which orbits are occupied and what debris circles there will by then be a fact and no longer an option. Which species still exist is decided beforehand. An extinct genome is not a difficult reconstruction but a missing template. Whether there are institutions holding in trust what can belong to no one will likewise have been settled by then.

In this arrangement we are not the inferior party waiting for later deliverance. We write the initial state. The role is uncomfortable, because it cannot be delegated. It is the more consequential one all the same.

What the later instance will see of us

When this intelligence looks back, in two hundred years, on the opening decades of its own development, it will not find our ignorance remarkable. That we understood little will strike it as unremarkable as the ignorance of the Bronze Age strikes us. What is remarkable about us is something else. We are the first generation whose actions stand fixed for it too.

What it will see, therefore, is our sorting. It will find an era that directed almost all its attention at the correctable, at quarters, electoral terms, product cycles. The few questions with no way back ran along on the same calendar. It will see that the depreciation of a data centre was negotiated more thoroughly than what remains. Which species disappear for good. Which orbits around this planet become unusable for centuries because too much debris circles there. Which pathogens come into being in laboratories, absent before and impossible to remove afterwards.

This sorting we have to carry out ourselves, and for that we need no later instance. The question behind it is simple: can a decision still be corrected later, or does it stand fixed by then? What stands fixed is judged by no one. It is only ever found.

If an instance two hundred years from now can only find your decisions today and change none of them: which would you make differently this week?

Website: https://planet-futures.org