--- "YKY (Yan King Yin)" <[EMAIL PROTECTED]> wrote:
> On 5/6/07, Matt Mahoney <[EMAIL PROTECTED]> wrote:
> > YKY, what do you mean by "scruffie"? Is that anyone who doesn't think
> FOPL
> > should be the core of an AGI?
>
> "Scruffies" tend to think AGI consists of a large number of
> heterogeneous modules. Let's try to avoid this debate by saying "we'll
> build as many modules as we see fit."
>
> Now FOPL is not the only KR language out there, but anyone can be reasonably
> familiar with it so it can serve as a foundational framework. If we are to
> agree on a KR scheme, it should be one that can be explained in 15 minutes.
I don't think there is an elegant solution to AGI. First, people have been
working on this for a long time, and if there was a simple solution we likely
would have found it. Second, the complexity of AGI, prior to any training, is
bounded by the complexity of DNA, which is quite high. Consider the
complexity of programming a robot spider to weave webs, not by training, but
by writing the algorithm yourself. Spiders are born with this knowledge.
Then consider the complexity of a human brain compared to that of a spider.
As for FOPL or probabilistic FOPL ("for most x, p(x) is usually true",
formalized with numeric probabilities), people have been down this path many
times and it is a dead end. What theoretical insight do you have that would
lead me to believe that your system would succeed where others have failed?
> I used to be pretty good at C and assembler hacking =) but we definitely
> should not worry about hardware at *this stage*. We should first focus
> on the algorithms.
We need to keep in mind that the current version requires 10^15 bits of memory
and 10^16 operations per second. Why would we evolve such large brains if
there was a shortcut?
> I think we should not go FOSS just because we arn't confident of ourselves,
> or to try to avoid competition. We love our work and should go the extra
> miles to make it profitable. Those who're not interested in business
> matters can leave that to somebody else in the group.
The problem with closed source is you have to pay your employees. Personally,
I am not interested in making a lot of money. I already make enough to buy
what I want. It is more important to have free time to pursue my interests.
AGI, especially language, is one of my interests. But I don't want to build
something aimlessly like Cyc. I would like to see an application, a goal in
which progress can be measured. I currently use text compression for this
purpose. Do you have a better idea?
-- Matt Mahoney, [EMAIL PROTECTED]
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?member_id=231415&user_secret=fabd7936
