Sandberg Cites Asilomar 1975 as Model for Tech Self-Regulation

Not knowing the hinges avoids responsibility. If we can improve epistemics, it seems we have a prima facie reason to do so in the hope that we may later affect hinges. Civilizations need not just better principles but better instruments for knowing when to act.

Can we identify the opportunity? There are plenty of examples of how we failed to see the risk (CFCs), made premature/self serving claims (Maxim on ending war), or just failed at seeing applications (Soviet military censors on stealth tech).

Sometimes we may be in a particularly important place, being able to affect the long-term future. Risk and capability are high during this period, and at least risk is significantly lower thereafter. Our choices matter, because of path dependence.

For DTD to work: Consequentiality: ordering of development actually affects outcomes. Malleability: development trajectory is still influenceable. Legibility: actors can recognise, with adequate confidence and adequate lead time, that they are inside the window.

The technology completion conjecture is that eventually every tech, no matter how harmful, will be invented. But we may be able to change the order: "Differential technology development: Leverage risk-reducing interactions between technologies by affecting their relative timing."

Hinges of History: centered on human action; decision-relevant; resource-focused. DTD: detect a risky technology; envision safety, defensive or substitute solutions; or a mechanism to stimulate their development. Big, overlapping epistemic component!

Philosophical limits on legibility. Not all hinges are knowable in advance — and the humility to recognise this matters. Needs to specify its own domain of application. Bounded confidence is useful.

(This is the part where I do a lot of case studies.)

Legibility-urgency tradeoff: technologies with short development timelines create urgent hinges but low legibility. Technologies with long timelines allow legibility but create coordination problems (the urgency is insufficient to motivate action). The sweet spot is narrow.

The role of epistemic infrastructure: what's needed is not (just) better ethical principles but better instruments for knowing when to apply them.

Indicators of hinge entry. Can you specify, even roughly, the signals that a technology has entered its critical window? Candidates include: capability thresholds being crossed, convergence of independent development lines, transition from basic to applied research, many more...

This requires building better epistemics about the near future – tech, forecasting, policy implementation; close collaboration with the communities that are doing the tech. It might be a moral imperative for ethicists to hang out more with diverse technical groups and listen.