On Sat, Mar 1, 2014 at 7:08 AM, Tim Tyler <[email protected]> wrote:
> On 28/02/2014 13:10, Matt Mahoney wrote:
>
> The bottleneck is not inference. For now, it is processing power. Vision, and 
> to a lesser extent hearing, language, robotics, and modeling human behavior 
> require enormous computation, even though they are not NP-hard. Moore's Law 
> should solve this problem in about 30 years if it stays on track.
>
> 30 years?!? Moravec predicted essentially this for around 2020 long ago - 
> and, these days,
> most seem to think that this was an over-estimate.
>
> http://singularitybookreviews.com/wp-content/uploads/2012/01/Moravec-Graph.png

If we had plotted a similar graph of automotive technology from 1890
to 1920, we might have predicted that cars would cost one cent and
travel faster than light. As late as the 1950's, we were predicting
nuclear powered flying cars by now.

I hate to be the one to throw a wet rag on Moore's Law, but the fact
is that machines are not now at monkey level intelligence as the graph
predicted in 1997. We are still a long way from building robots that
can weave spider webs.
However, we might be close to achieving nematode level.
http://www.artificialbrains.com/openworm

You might have noticed that several years ago, CPU clock speeds have
stopped getting faster.
http://mike.teczno.com-img.s3.amazonaws.com/nogis-talk/NoGIS-slides.016.jpg

Computers are still getting faster through parallelism, but soon we
will run into the fundamental problem that you can't make transistors
smaller than atoms. Right now the smallest feature sizes are about 100
atoms wide and a few atoms thick. Further shrinking will not make
transistors 10^5 times more energy efficient, which is what we would
need to do to match the power consumption of the human brain.

-- 
-- 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