On Tue, Feb 12, 2019 at 10:52 AM Jim Bromer <[email protected]> wrote:
>
> Human beings make machines that are more powerful than they are, and
> can do things that they cannot do.

I pointed this out when someone asked on Quora when machines would
surpass human intelligence. Machines are already a billion times
faster than us at math, defeat world champions at chess, Jeopardy, and
Go, translate hundreds of languages, recognize billions of faces, fly
spacecraft, etc. Yet we still don't consider them more intelligent
than us. 
https://www.quora.com/How-long-will-it-take-until-machines-greatly-supersede-human-level-intelligence/answer/Matt-Mahoney-2

The short answer is that you can't compare human and machine
intelligence. The two most widely accepted measures, the Turing test
and universal intelligence, give vastly different answers.

The best a machine can ever do in a Turing test is equal human
intelligence by being indistinguishable from human. It can never
exceed it.

The second test is Legg and Hutter's universal intelligence. We can't
measure accumulated reward over an infinite set of environments, but a
good practical approximation would be dollars per hour. The tipping
point would be where machines are earning half of the world's income
(for their owners), seen as a doubling of world GDP from a baseline
agricultural society. This happened sometime in the 19th century
around the inventions of the railroad and telegraph. Today we have a
100 fold increase, meaning machines are doing 99% of the work.

Then where is our singularity? Well, we do have a super-exponential
growth rate of world knowledge and computing power, and have for
centuries. I don't know if or when it will peak, but I doubt it will
be soon enough that we will live to see it.

-- 
-- Matt Mahoney, [email protected]

------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/Ta6fce6a7b640886a-Md068625f32f07c28c6b39f9c
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to