On Thu, Aug 23, 2012 at 3:58 PM, [email protected]
<[email protected]> wrote:
>
> No, I want to build something that can tap into the knowledge & computing
> power of the internet just like we do.

That requires solving the remaining hard problems like language,
vision, robotics, and modeling human behavior. If we could do that,
then we could automate all human labor and stop paying people $70
trillion per year to work. I'm sure you aren't the first person to
work on this problem.

> As for the weaknesses of humans you list, I fail to see why they would
> have to apply to an artificial mind. Those are constraints placed on us by
> our particular implementation of intelligence.

Of course we would not for the purpose of working. But one application
of artificial stupidity would be uploading. If you want to resurrect
your deceased loved ones as robots, then you want them to act just
like they did when they were alive.

> And if you don't see the
> utility of having an autonomous thinking & decision making machine, you are
> in denial.

I see an existential threat to homo sapiens. Do you really want to
make machines that are smarter than you and not have control over
them?

> I would also like to be able to talk to a machine and know that it really
> does understand my wishes and requests, which search engines don't do yet.

Google is getting better at understanding natural language questions.
The only reason you don't think it "understands" you is because you
know it's a computer.

> And when search engines finally succeed at that, it will be because they
> will have implemented something analogous to human cognition.

It will be because they have successfully modeled your mind. A model
of your mind is a function that takes sensory input and returns a
prediction of your actions.

Models of human minds are very useful things. If I had a model of your
mind, I could predict what would make you happy, or what would make
you buy something and how much you would pay. I could program a robot
to carry out those predictions in real time and have an upload of you,
or just use it to steal your identity.

> Now what I'm working on isn't plugged into the internet for anyone to use
> like an (intelligent, conversational) search engine, but in order to lay the
> groundwork for such a service, people like me have to spend our time working
> on these sorts of projects. How else will we learn how to make the web truly
> intelligent? And how else will we make robots truly effective and safe,
> besides making them understand what's happening in the world around them?

The hard problems I mentioned (language, vision, etc) require a lot of
computing power. I estimate 10 petaflops and 1 petabyte, which is
about the power needed to run a human brain sized neural network in
real time. A lot of people think there are more efficient solutions,
but nobody has found any yet. The people who are actually making the
most progress in AI have big budgets and big buildings with cooling
towers filled with racks of servers. But good luck.


-- 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-c97d2393
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968
Powered by Listbox: http://www.listbox.com

Reply via email to