Hide table of contents

I have the impression it is usually supposed technological progress will come to an end soon after superintelligent artificial intelligence is developed, in which case the effective thinking time / computation needed to cause one unit of welfare would also stop decreasing. For example, William MacAskill says in MacAskill 2022 (emphasis mine):

The world growth rate is around 3.5% per year.⁴⁹,⁵⁰ It is not plausible that we can sustain such a high growth rate indefinitely into the future. To see this, suppose that in the future the world economy will grow by (merely) 2% per year indefinitely. If so, then after 10,000 years there would be 10^19 times present-day GDP for every atom in the galaxy. This is not a plausible outcome.

I believe that this appeal to rapid economic and technological progress is the strongest argument in favour of thinking that we live at an unusually influential time. The present time is certainly highly distinctive in terms of its growth rate. And [i)] even if you only think it 10% likely that the most influential time is at a period of unusually high economic growth, then you should give at least a 10% credence to the idea that we are among the most influential 10,000 years. And [ii)] there are positive arguments for thinking that we should expect the most influential times to be those of unusually fast technological progress: in particular, if the fate of the future is determined by how we manage the invention and deployment of particular technologies (such as artificial intelligence or particularly dangerous weapons), then at periods of unusually fast technological progress, we are moving faster through the space of all technological inventions, and are therefore more likely to discover one of the critical technologies.

I agree with ii), but not i). Intuitively, technological progress is caused by effective thinking time / computation[1]. So, even if the thinking time / computation per year is not growing, technological progress can still continue by means of finding more effective ways of using the constant supply of thinking time / computation.

I understand technological progress will tend to 0 as physical limits are approached. Nonetheless, it seems a priori very unlikely that the vast majority of progress will be made soon after the development of superintelligent artificial intelligence, because most effective thinking time / computation will happen after that. So will the vast majority of technological progress happen in the longterm future?

As far as I can tell, Will thought i) and ii) to be crucial considerations for assessing the hinge of history hypothesis. Indeed, this can be rejected if the vast majority of welfare will be caused in the longterm future, which follows from:

  • The vast majority of effective thinking time / computation being in the longterm future.
  • The effective thinking time / computation needed to cause a unit of welfare obeying something like the Wright’s Law (e.g. doubling the cumulative thinking time / computation results in a 20 % decrease in the cumulative thinking time / computation needed to cause a unit of welfare).

I believe the 1st of the above is usually accepted, but the 2nd is often rejected. Will’s claim that technological progress is associated with economic growth (see i) above) is in agreement with rejecting the 2nd point, but no arguments were given for/against it.

The invention of artificial general intelligence could make extreme types of lock-in technologically feasible, but that is not necessarily likely:

Note that we’re mostly making claims about feasibility as opposed to likelihood.

  1. ^

     Effective thinking time / computation factors in not only the amount of thinking time / computation, but also its quality / algorithmic efficiency.

New Answer
New Comment
No comments on this post yet.
Be the first to respond.
Curated and popular this week
Relevant opportunities