1. 8

Readers have probably seen Elon Musk freaking out over the coming AI apocalypse. And most recently the FB CEO has fired back that he thinks Musk is wrong. Most AI researches I have listened to seem to think Musk is overreacting and some don’t.

This debate is a good example of a type of problem that I don’t feel I have the mental model to really consider, and that is if something’s progress is accurately described as exponential how do you think about it? In AI is exponential than even though things seem dumb right now, in 5 years they could easily be far beyond what we imagine.

But my question is not about if AI is exponential in its progress or if Musk is right.

Does anyone here have common mental tools they use to analyse and draw conclusions from things whose progress is non-linear over time?

  1.  

  2. 2

    http://lesswrong.com/lw/2bu/your_intuitions_are_not_magic/

    Basically, we always need to question our analysis of a situation. We need to put continuous effort into discovering our biases, and learning the shortcomings of our tools. Are you more curious about fitting models to data or predicting the future?

    1. 1

      The closest set of tools we have that I’m aware of are for things like credit/debit and the like, growth factors and so on. The interesting thing about those is that no exponential growth happens in the real world unabated, most getting an S shape before it’s played out. I suspect the same will apply to AI, even if it goes a few levels above us. Barring that, one potential mental model for such an AI would be to consider it supernatural, as that’s a common approach to things we don’t understand (not everyone takes this route, of course, but it is one that would need to be considered at least for some people).

      Any sufficiently advanced technology is indistinguishable from magic

      • Arthur C. Clarke
      1. 1

        Do you have an example of historical changes you’d consider to have shown nonlinear progress? It’d make it easier for me to think about to have a non-hypothetical example.

        1. 1

          Computational power via Moores Law is the most recent example. We’ve gone from human phone switches to the internet in, depending on what side you’re on, an unpredictable pace.