Real time systems need to execute processes, often within a hard time limit to avoid catastrophic consequences. ‘Tasks’ or ‘jobs’ arrive at the task scheduler queue with a deadline and a worst case computation requirement and must be performed in an order which minimizes the number of missed deadlines. Put it this way, real time scheduling adheres to the ‘better never than late’ policy.
I was quite intrigued to notice the striking resemblance between real-time scheduling and existential risks. Both require foresight and offer no second chances, irreversibly overruling any chances of survival or redemption. Real time systems, fortunately, have a provably optimal algorithm to schedule tasks – it’s called the Earliest Deadline First scheduling algorithm. As the name suggests, it prioritizes tasks based solely on their deadlines and seeks to meet the earliest deadline at all times.
Building an armoury to defend against an alien attack is utterly futile if we’re more likely to be be obliterated by a large-scale nuclear war or a global pandemic before that. Likewise, focusing the spotlight entirely on avoiding global pandemics isn’t a great idea if the melting ice-caps will drown us first. We must adopt an earliest deadline first strategy rigorously and spend more than we do on cigarette ads to ensure that humanity survives this century. Of course, X-risks arrive with probabilistic deadlines and the worst case cost of avoiding them is also highly uncertain. I’m not aware of any probabilistic EDF schedulers but I plan to read about them soon and see how they can be incorporated in policy making more rigorously.
Disclaimer: I no longer agree with most of the content written below.
- (1) If Earth Overshoot Day isn’t a publicity stunt after all, it might have dramatic consequences for effective altruists. Even if the date (which conspicuously is always an exact date instead of a range, given the uncertainty and seeming intractability of the problem) they claim every year is exaggerated, the possibility of wrecking our planet faster than it can heal warrants an intelligent inquiry, particularly because of the eventuality it entails.
- (2) The creation of life is generally assigned positive value. I believe the creation of life has high positive value and death has high negative value, not necessarily exactly canceling out.
Is there a point beyond which we should consider collectively slowing down population growth or even decrease the population despite the ‘inherent’ value in creating life? A reasonable tipping point seems to be when the marginal value of creating one more life is exactly canceled out by the inconveniences caused to the rest of the world. One could envisage a starving planet, where every new birth engenders more starvation and suffering – an unlikely scenario where (classical ?) utilitarianism would advise against creating more lives.
- (3) In a recent post, 80000 hours and Toby Ord discuss that ‘the welfare of future generations should be our number one moral concern’ as only a small fraction of all the people whose well-being we care about are alive today.
- (4) In Altruism and Profit, Paul Christiano says:
If you complete a high-impact technology project which other technologists would have tackled in a year, the counterfactual doesn’t just involve that project finishing a year sooner—it also involves those technologists going off to do whatever else they would have done.
I argue the net benefit should also include the effects of cascading subsequent discoveries by -1 years. If the discovery improves the quality of everyone’s life by a factor of 1.000001, it doesn’t just ‘save’ 0.000001 * 7 billion QALYs, the counterfactual must also involve the preponement and acceleration of subsequent discoveries that might slightly improve the quality for an extremely large number of generations (In other words, it should have a high multiplier).
Any analysis of the benefit of an outcome must also take into account the impact it has on the (hopefully countless) future generations. The marginal value of giving birth to one more child must also include the impact on all future generations.
The marginal benefit of producing one more child must at least offset the cost of depriving future generations of the resources it consumes. Let it sink in, (if) we are consuming more than we can replenish, (then) with the birth of every new child Earth Overshoot Day is getting closer to January 1. Pause and ponder over the possibility of starving (or maybe just depriving) countless future generations, and the expected impact of averting it. The world needn’t be starving already for the expected value of creating human life to be negative, increasing the likelihood of starving future generations counts as well.
If Earth Overshoot Day is for real, natalism calls for antinatalism.