Electricity leases signed through 2048. Published models that can never be withdrawn. Skills no longer taught because a tool covers them. Each of these decisions is being made today, by people who will not bear the main consequences. That is the very definition of an intergenerational problem, and we have no method for dealing with it.
We have covered delegating decisions and responsibility. Here is the long-term version, and it is harder.
Three things that cannot be undone
Infrastructure. A data centre commits land, electricity and water for decades. We have documented a twenty-year lease on 191 megawatts. Those who will live with these facilities in 2045 have no say in the matter, and for good reason.
Irreversible publications. This is the heart of the debate on open weights. A published model can no longer be recalled. If a capability turns out to be dangerous in ten years, the decision made today remains in effect.
Abandoned skills. This is the most discreet and perhaps the most profound. When a generation stops practising a know-how because a tool covers it, that know-how is no longer passed on. You can relearn from books; you do not easily relearn what was transmitted through practice.
The problem of representation
Here is the central difficulty, and it is an old one. Those most affected by these decisions cannot take part in making them, because they do not exist yet.
Our institutions are built to arbitrate between present interests. A vote, a negotiation, a market presuppose parties capable of defending their position. Future generations have none of those means.
This produces a systematic bias: immediate benefits are defended by those who enjoy them, deferred costs are defended by no one.
One might object that it is enough to account for future interests. The problem is that we do not know how to do that. We do not know what people will need in thirty years, nor what capabilities they will have. Reasoning that claims to maximise future well-being rests on unverifiable projections, and it can justify almost any present sacrifice. That is why the most robust approaches do not seek to optimise the future, but to preserve options.
The criterion that seems to hold
Since we cannot know what they will want, a more modest rule imposes itself: leave them the capacity to choose.
That gives an operational criterion: prefer what can be undone. Between two options of comparable efficiency, the one that leaves room for turning back is better, even if it is slightly less performant.
Applied concretely, this guides several debates we cover. Reversible infrastructure is better than a twenty-year commitment. A skill maintained even when a tool covers it is an insurance policy. Dependence on a single supplier is a choice imposed on those who come after.
This is exactly the reasoning we use for practical reasons — do not depend on one supplier, keep the ability to switch. It turns out it also has a moral justification.
What we cannot promise them
We must be honest about the limits of this reflection.
We cannot guarantee them a better world, nor even foresee what will seem valuable to them. Every generation has judged the next ungrateful, and every generation has inherited things it would not have chosen and made do with them.
There is also a form of arrogance in deciding for them in the name of what we believe we know. Those who delayed technologies in the name of future generations have sometimes been badly wrong.
What to take away
The question does not call for a maximalist answer, and being wary of reasoning that sacrifices much of the present to a hypothetical future remains wise.
What seems right can be said in few words: we do not owe them a perfect world, we owe them not closing doors on their behalf. A decision that can be undone leaves them the choice. An irreversible decision takes it away.
And there is one thing we can pass on to them regardless of everything else: the ability to understand what they inherit. A generation that knows how the systems it depends on work can change them. A generation that endures them without understanding them cannot, no matter the quality of what was left to it.