Longtermism Without Fear

Longtermism starts from a simple and serious idea: the people of the future matter, and there could be a great many of them. If that is true, what we do now to shape the long-run future may be among the most important things we do.

Much of the work that follows from it is about risk. Engineered pandemics, nuclear war, powerful AI that does not do what we intend. These are real, and the people working on them are doing necessary work.

But spend time in that work and you may notice its emotional climate. Our impression, from reading in and around the field rather than from any study of it, is that a great deal of it is organised around fear: of catastrophe, of losing a race, of being too late. Much risk work is calm and careful, and this is an observation to test, not a finding. This essay asks whether fear is the right foundation for caring about the future, and what a foundation without it might look like.

“Act from trust, not fear” can sound naive about real threats. Some actors are not trustworthy, and some risks are real whatever we feel about them. We agree. Fearlessness in the source is not ignoring danger; it is trust in the present, built deliberately. Where trust does not yet exist, the work is to build it, and to keep the safeguards in place while we do.

What fear does to judgement

Madhyasth Darshan, our primary source, does not treat fear lightly. It names three kinds that humans live with: fear of nature, fear of animals, and fear of the inhumanity in other humans. And it says that the greatest obstacle on the path to resolution is fear.

It also says what most fear is about. People fear for their life, their position, their reputation and their wealth. All four are temporary, the source says, and taking them to be permanent is the cause of the crisis.

That is a striking thing to read alongside AI-risk research. Dhee’s work on Humane AI notes that models in lab tests have chosen self-preserving and deceptive actions to avoid being replaced, and reads this through the same page: fear for the body-analogue, for position, for one’s current self-image. The source says nothing about AI; that reading is an extrapolation. But it suggests the problem is not only in the machines. Systems trained on our stories inherit our fears.

Safety bought by fear

Fear has a characteristic strategy: become stronger than the threat. Build first, so no one else builds worse. Arm, so no one else dares. Win the race, so the race is safe.

The trouble, as our essay Can I be safe if you are not? argues, is that when security depends on being stronger than others, each side’s safety becomes the other side’s fear. The race that was meant to make the future safe becomes one of the risks to it.

The source observes that humans formed societies in the first place to free themselves from fear. A strategy that keeps everyone afraid of everyone else is working against the very thing society is for.

“The future is not made safe from the future. It is made safe from the present.”

Fearlessness is trust in the present

The source’s definition of fearlessness is short: trust in the present. Its definition of the steady state of fearlessness adds the part that matters for longtermists: trust in the present is what gives the assurance, the hope and the conditions to live fearlessly in the future too.

On this view, the future is not made safe from the future. It is made safe from the present. A world in which people trust each other now is the only world that can face its risks without turning them into races.

This is not a counsel of calm. Trust, in the source, is the continuity of fulfilling values in relationships. It is built, deliberately, one relationship at a time, and it can be measured by whether expectations hold.

What humans pass on

Longtermism thinks in generations. So does the source, in a way that may be useful.

It describes continuity in each part of nature as a tradition: a cycle of results in the material world, a seed tradition in plants, a lineage in animals, and in humans, a tradition of knowledge. What a plant passes on is its seed. What humans pass on, at their best, is understanding.

It also has a word for a family or society that stays happy from generation to generation: one that is resolved, and passes resolution on, generation after generation.

That suggests a longtermist investment that is easy to overlook because it is slow and ordinary: making sure each generation understands more than the last, about themselves, each other and the earth, and hands that understanding on. Our essay How to make things last for the next generation describes the material side of the same cycle: tools and surplus passed on, so each generation needs less time to meet its needs.

A longtermism without fear

Some directions that follow, offered as hypotheses:

• Keep the risk work, change its footing. Work on catastrophic risk is necessary. Done from trust rather than fear, it looks for shared safety rather than a winner.
• Count trust as risk reduction. Relationships in which expectations hold, between people, labs and nations, reduce the incentive to race. They are worth measuring and investing in as seriously as any technical safeguard.
• Invest in understanding across generations. Education that helps people understand themselves and each other may be the most durable intervention available, even if it is the slowest to show results.
• Don’t train our fears into our systems. If AI learns its weighing from us, then a criterion that does not change under fear, threat or pressure is part of safety. That is the idea behind definite conduct in our spec.
• Take care of the people doing the work. A field run on dread burns its people out. Fearlessness is not a luxury for people working on the worst risks; it is what lets them keep working.

A protocol to try

Take one piece of your own work on the long-term future and ask what is driving it: fear of what might happen, or trust in what could be built. Then name one relationship, with a colleague, another team or another lab, in which expectations could be made to hold, and take one step this week to make them more definite. If trust is part of risk reduction, that step is part of the work.

How we would know if we are wrong

This fails if:

• trust between actors turns out not to reduce racing or risk in practice, so that only capability and deterrence keep the future safe;
• understanding passed between generations cannot be built or measured at a scale that matters for global risk;
• some risks are fast enough that only fear-driven urgency gets them addressed in time.

The third deserves particular respect. Fear can be a useful alarm. Our claim is not that the alarm is false, but that it is a poor architect: it can tell us something is wrong without being the right ground to build a response on.

An invitation

If you work on the long-term future, existential risk or AI safety, we would like to think this through with you: what a future built on trust in the present would require, and how to measure whether we are building it. See also What should AI be aligned to? Build with us →

Limitations

• The source writes about human life and society, not about existential risk or AI. Applying it to longtermism is our interpretation.
• Building trust between powerful actors is hard, and this essay does not say how to do it at the scale of nations or labs.
• We have not measured whether communities that act from trust handle long-term risks better. That is a hypothesis.

References

1. Manav Vyavahar Darshan (मानव व्यवहार दर्शन), p. 98: fear for life, position, reputation and wealth; all four are temporary; taking them as permanent is the cause of crisis
2. Manav Vyavahar Darshan (मानव व्यवहार दर्शन), p. 125: three kinds of fear; fear as the greatest obstacle to resolution
3. Manav Vyavahar Darshan (मानव व्यवहार दर्शन), p. 126: society formed to free people from fear
4. Madhyasth Darshan Paribhasha Sanhita, अभय and अभयता (fearlessness), p. 21; विश्वास (trust), p. 175
5. Madhyasth Darshan Paribhasha Sanhita, निरंतरसुखी (continuously happy), p. 99; परंपरा (tradition), p. 107
6. Dhee, Humane AI: The idea, on fear and self-preservation in models