From: Matt Mahoney (firstname.lastname@example.org)
Date: Wed Feb 25 2009 - 09:21:38 MST
--- On Tue, 2/24/09, Eliezer Yudkowsky <email@example.com> wrote:
> Allow me to summarize the premises and conclusions of this
> 1) Markets are absolutely, perfectly, exactly efficient over all
> domains and between all times.
Giving all your money to a program that simulates you and then killing yourself (a.k.a. uploading) is not economically rational. But I argued based on automating the economy because I think that problem will be solved first. Passing the Turing test is easier if the people involved have not previously met.
My argument is that RSI is not economically viable until we can make AI cheaper than human labor, not just smarter. Not that we won't build uneconomical AI first. Maybe someone like Minsky might be able to build an AI that passes the Turing test for $10 million. But it won't be the kind of AI where he can sit back and watch it self improve. As it grows, he would have to spend more money for hardware and maintenance. It is kind of like Google has partially solved AI, but they still have to hire more people as it grows because the machines can't do everything.
> 2) AI would make a quadrillion-dollar profit.
Distributed among its 10^10 owners, yes. It is not that much. What value do you place on the benefits of technology to your life as opposed to, say, sustenance farming? How much would it be worth to get paid for doing exactly what you want?
> 3) AI can be obtained at any time by, and only by, spending a
> quadrillion dollars.
We don't have a choice. We want it, so we will spend it and get AI. Not out of some grand design, but out of an internet that gets incrementally smarter until it is no longer clear when you are talking to a human and when you are talking to a program.
> It doesn't lack for audacity, but I'm afraid it
> somewhat lacks for sanity.
Sorry, history is full of wildly optimistic projections about AI. If it were possible to solve a $1 quadrillion problem for only $1 trillion, we (who are collectively much smarter than you) would have figured it out long ago.
-- Matt Mahoney, firstname.lastname@example.org
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:01:04 MDT