The BASIC interpreter on my Commodore 64 had self-modifying code. It
is an interesting concept, but not new. More importantly, it does not
address the high cost of AGI in hardware, software, and training data.
I do not believe (1). Computers today have nowhere near the
computing capacity of the human brain with respect to neural models of
vision, language, and robotics.
If you looked at my web site where I have a full spec of my project, you
wouldn't compare it with Commodore 64 Basic. I did program a few business
systems using the Commodore back in 1978, however. The kind of flexibility
I have created in code generation can't be had, even in Lisp.
The cost can change based on the kind of tools used. You say high cost but
I would say some solutions are now a reasonable cost.
I don't think comparing a brain to a computer is reasonable at all.
1. I don't think humans have a monopoly on intelligence and
therefore AGI doesn't have to mimic a human.
2. Human brains are good at some things but very poor (compared to
computers) at doing other things.
3. My observation of intelligence in humans is that for most of the
intelligent stuff, humans don't do a very good job.
4. I believe that most of our intelligence is simulated at a very
high and inefficient level in our brains and that most of the theoretical
processing power of our brains is wasted.
5. The raw processing of our senses are definitely an exception to
point 4 but that processing doesn't constitute intelligence IMO.
I could add many more points but am trying not to bore people.
I have designed and built robots and I still keep up on all the latest
developments. Check out Rodney Brooks' latest human compatible robot
creation. Vision recognition is much better today and at a much lower cost.
Language translation is so good and so low cost that professional
translators can hardly find a job.
Even with stating the previous paragraph, I don't think any of these items
are what I would call intelligence!
Every major effort I have seen that tried to create AGI got to a certain
point where the money ran out for the programming and then it stalled.
Maybe they had the right ideas or maybe not but the programming effort, in
the end, stalled each project. If this statement is in dispute, I would be
happy to get educated differently. I want AGI to succeed. I want Ben's
OpenCog to succeed. I would like to see many more people actually trying to
create AGI rather than just writing about it.
If human programmers have to write all the code that I think is needed to
create AGI then AGI won't get solved for a long while yet.
David
-----Original Message-----
From: Matt Mahoney [mailto:[email protected]]
Sent: January-10-13 4:08 PM
To: AGI
Subject: Re: [agi] Why Logic & Maths Have Sweet FA to do with Real world
reasoning
On Thu, Jan 10, 2013 at 12:54 PM, David Clark <[email protected]>
wrote:
> 1. I believe that AGI can be created using software we have today,
on computers that exist today. AGI has not been created yet so I have no
proof that this is true, even though I am acting as if it is.
>
> 2. I believe that AGI can't move forward until programs have the
unlimited ability to create other programs. This hasn't happened yet and
therefore I have no proof that this capability will be more successful than
current approaches BUT at least it isn't one of the techniques that we know
doesn't work.
>
> I am acting as if these 2 beliefs were true while I have no proof they
are. Guilty as charged.
>
> If people on this list don't also believe my first assumption, I don't
understand why they would be on this list!
Minsky and Kurzweil are not on this list. When this list was formed, there
were several people who were working on AGI projects. All but Ben have left.
> I have many arguments that point to my second assumption being correct but
no proof.
The BASIC interpreter on my Commodore 64 had self-modifying code. It is an
interesting concept, but not new. More importantly, it does not address the
high cost of AGI in hardware, software, and training data.
I do not believe (1). Computers today have nowhere near the computing
capacity of the human brain with respect to neural models of vision,
language, and robotics.
--
-- 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
