On Thu, Feb 27, 2014 at 5:31 PM, Samantha Atkins <[email protected]> wrote:
>
> Exactly.  If P=NP then any level of godlike machine you can dream of can be 
> built.   All encryption is pointless as a minor side effect.  :)

P=NP won't help you build AI. Neural networks will still take just as
long to evaluate and train. Vision, language, robotics, and modeling
human behavior are hard because of the amount of data that has to be
processed and learned, but it is already O(n). It will not speed up
search problems like simulated evolution.

P=NP doesn't solve the halting problem or get around Rice's theorem or
the uncomputability of Kolmogorov complexity. In practice, this means
that software testing would still be hard and imperfect. Theorem
proving and AIXI would still not be computable. AIXI^tl would still
requires exponential time because it is not an NP-hard problem.

P=NP won't solve quantum mechanics. This means you still couldn't
write a program that inputs chemical formulas and outputs chemical
properties like the freezing point of water.

P=NP doesn't eliminate chaos. Weather forecasts would not be any more accurate.

P=NP doesn't eliminate Goedel incompleteness. There will still be math
problems we can't solve (like whether P=NP).

P=NP will break all encryption as you said. This will complicate
decentralized AI designs (like my proposed CMR) because agents would
not be able to securely identify themselves to establish reputations
as reliable data sources. The Internet would be far more vulnerable
and probably useless and unusable. Electronic banking would require
moving information over physically secure channels, making credit
cards mostly impractical.

-- 
-- Matt Mahoney, [email protected]


-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com

Reply via email to