I think this is an excellent question and hasn’t (yet) received the discussion it deserves. Below are a few half-baked thoughts.

The last couple of years have significantly increased my credence that we’ll see explosive growth as a result of AI within the next 20 years. If this happens, it’ll raise a huge number of different challenges; human extinction at the hands of AI is obviously one. But there are others, too, even if we successfully avoid extinction, such as by aligning AI or coordinating to ensure that all powerful AI systems are limited in their capacities in some way (for example, by lacking long-term planning or theory of mind, or being subject to constant monitoring by AI law enforcement).

One framing is:

• Think of all the technological challenges that we’d face over the coming 500 years, on a business-as-usual 1-5% per year growth rate.
• Now imagine that that occurs over the course of 5 years rather than 500.
• And now imagine we only have months to years to respond to each new challenge, rather than decades.

A few things leap out to me:

1. One’s biorisk estimates should go higher because:
   1. We’d develop even better biotech than we would otherwise (in the same time period).
   2. There is little institutional time to respond.
   3. The technology is even more likely to be highly democratised. (E.g. open-source models able to answer highly sophisticated questions about biotech available to everyone.) 
2. The point of time at which we expand beyond our solar system might be within our lifetimes. This could be one of the most influential moments in human history: the speed of light sets an upper bound on how fast you can go, so who leaves first at max-speed gets there first. And, plausibly, solar systems are defense-dominant so whoever gets there first controls the resources they reach indefinitely. 
   1. To my knowledge there’s been almost no work done on what governance regime should be in place to guide this, and an anarchic land grab doesn’t seem ideal.  
   2. This warrants at least some work on thinking through governance, building up a coalition of people (across countries, labs, and at the UN) who take this seriously. I’m hoping to do some work on this topic this year. (This relates to discussion of the long reflection, which has remained hand-wavy to date.)
3. Relatedly, laws around capital ownership. If almost all economic value is created by AI, then whoever owns the aligned AI (and hardware, data, etc) would have almost total economic power. Similarly, if  all military power is held by AI, then whoever owns the AI would have almost total military power. In principle this could be a single company or a small group of people. We could try to work on legislation in advance to more widely share the increased power from aligned AI. 
4. More generally, in-advance discussion of what sort of world we’re aiming for in a post-AGI future seems helpful.
   1. One counterargument is: “This discussion will all become irrelevant post-superintelligence.” And that might be right. But I’m not confident that’s right. And, even if so, progress on this question can make things go better pre-superintelligence. For example, it’s plausible to me that, with cooperation and trade, a wide range of value systems can get 90%+ of what they value post-superintelligence. If this is right, and became widely known, then it could reduce the desire for different actors to race. 
5. One’s estimate of existential risks from unknown tech within our lifetimes should go higher, too. This favours some amount of general-purpose responsive infrastructure, like crisis response teams within government, cash reserves that can be drawn upon, and pre-approved emergency policy regimes. It also favours, more generally, governments that can nimbly respond in times of crisis, and philanthropists holding on to savings.
6. Work on certain key ethical issues, like the moral status of digital beings and concern for digital sentience, seems even more important.  These might seem wacky now, but might be very real in the next couple of decades. 
7. Scenario planning to extend this list seems very important.

Then, in terms of impacts on other existing EA cause areas:

• Global health and development:
  • Near-term explosive economic growth would greatly increase the value of saving lives relative to the value of improving lives or generally making poor people richer. If AI goes well, then it could greatly extend currently-existing lives, and greatly increase their quality of life, too. On a person-affecting view of population ethics (which you should give some weight to, and might be why you focus on this cause area) that’s huge. The benefit to the child of saving a child’s life would be much greater than just ~40 QALYs, perhaps many orders of magnitude greater.
  • The medium-term benefits of global development would be smaller. Even if wealth is highly concentrated after superintelligence, there would probably be trickle-down benefits that, in absolute terms, would be transformative for the world’s poor; previous economic gains would be largely washed out. So the discount rate should be higher. This is relevant for, say, deworming versus bednets, though the previous consideration about saving versus benefitting lives is much larger.
• Farm animal welfare: 
  • Some animal welfare work is about changing values and attitudes to non-human beings. This looks more promising than it did, especially among key decision-makers (e.g. politicians, people at AI labs), because those value changes could significantly affect the long-term future, for example via attitudes to digital sentience.  (Though there’s a significant worry that any actions on this will get washed out, because the volume and quality of information and arguments that people are exposed to will potentially be much greater after the advent of advanced AI, and have a much bigger impact on their views and values.)
  • Activities with medium-term payoff, like advancing technology on cultivated meat, don’t look as promising. The technology will get created soon anyway, and will make a big dent into rates of animal farming; the costs to raise animals more humanely will also decrease post-superintelligence.  

I’m confident that the above list of considerations is extremely non-exhaustive! 

I think it mostly means that you should be looking to get quick wins. When calculating the effectiveness of an intervention, don't assume things like "over the course of an 85-year lifespan this person will be healthier due to better nutrition now." or "this person will have better education and thus more income 20 years from now." Instead just think: How much good does this intervention accomplish in the next 5 years? (Or if you want to get fancy, use e.g. a 10%/yr discount rate)

See Neartermists should consider AGI timelines in their spending decisions - EA Forum (effectivealtruism.org)