On Tue, Feb 17, 2015 at 4:18 PM, Aaron Hosford <[email protected]> wrote:
>> We can prove that good solutions must have high algorithmic complexity.
>
> Could you give a sketch for such a proof?

Suppose you have a simple predictor with Kolmogorov complexity N. Then
I can create a sequence with about the same complexity that your
predictor can't predict. My program simulates your program and outputs
the opposite of whatever you predict.

The long version by Legg: http://arxiv.org/abs/cs/0606070

> Why does the complexity have to
> initially reside in the algorithm, rather than in the environment?

Because the environment can lie. I could give your predictor some code
that would make it a better predictor, or worse. It has no way to
know.

And once it has collected all human knowledge, learning will slow down
because it will have to learn by experiment just like we do. Some
experiments take a long time no matter how smart you are. For example,
we know very little about what therapies will slow down aging. Any
experiment takes decades to get an answer. This is the reason that the
rate of increase of worldwide life expectancy peaked at 0.2 years per
year in the 1990's (1970's in developed countries) and is declining.

And yes, we have machines that can go to the moon. They are stronger
and faster. They are also more intelligent, depending on what test you
use for intelligence. If the test is arithmetic speed and accuracy,
then machines surpassed human level intelligence 100 years ago.

But again, intelligence is not the goal.

-- 
-- 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