Mark Waser wrote:
As I wrote to Robin Hanson earlier today, the fact you don’t agree
with what we view as the relatively high probability of success for
our approach does not reflect poorly on either your intelligence or
your knowledge of AI. If you haven’t spent a lot of time thinking
about a Novamente-like approach there is no reason, no matter how
bright you are that you should be able to understand its promise.
Maybe I shouldn't get into this, but . . . .
I've read the detailed Novamente design. It meshed very well with a lot
of my previous intuitions and in some areas went into a lot more detail
(and went into a lot less detail in others). I firmly believe that an
AGI could be built on top of Novamente's design.
That being said, I don't believe that Novamente is particularly close to
the fastest path to AGI for several reasons.
First, Novamente is a discovery system (and a *really* good one). The
other parts of it's design, however, are not fully fleshed out and there
are huge "a miracle happens here" holes. This is not to denigrate Ben
and his team in any way, shape, or form. They've done wonders with
their resources and can't do everything.
Second, over the past few years, I've become more and more convinced
that discovery systems, while they do "learn", are not the type of
learning that I think is necessary for AGI. Novamente can certainly
tease out patterns from large quantities of data but it isn't fully
designed (at this point) to do anything like reasoning by analogy, for
example. Ben does have some plans for this but, my opinion is that, he
is still in the realm of "a miracle happens here" on this subject.
Third, and I've said this before, there are some fundamental engineering
features (scale-invariance of knowledge, ways of determining and
exploiting encapsulation and modularity of knowledge without killing
useful "leaky" abstractions, etc.) that aren't implemented yet in
Novamente that really need to be implemented much earlier rather than
later. Also, I have a lot of questions about Novamente's "memory" design.
In particular, I think that Novamente's foray into learning in a virtual
world is either going to be incredibly useful or a rather large bust
because it is precisely the type of learning that Novamente hasn't
specialized in before this point.
A number of people on this list seem to regard Ben as almost a deity or
a prophet. Ben is intelligent, creative, has a solid background, and
gets to work hard in the field so he looks a lot better than most
everyone else. It also means that he has polished his ideas and
eliminated the most obvious problems. This does not, however, mean that
he has a provably correct path. Novamente may lead to AGI (with *a lot*
more hard work). Personally, as I've said, I believe that it is *a
path* but one which will be overtaken and passed by a shorter, easier
path (just as I believe that brain emulation is a path that will be
overtaken and passed by a shorter, easier path).
When one simply looks at the difference between the brain emulation path
and the Novamente path (much less other paths like Hawkins, etc), one
has to realize that there is a *wide* range of potentially viable paths
to AGI. What is particularly distressing is those individuals who insist
on being Novamente fanboys without pointing to any specific features
that are particularly important or unique. Ben and, for example,
Richard argue in specific details. They pretty much understand where
each other stands but disagree with some fundamental (but unprovable)
assumptions on the other's part. Personally, it seems to me that
Novamente could answer Richard's complaints with some tweaking and
minor/moderate change of focus (since Novamente is actually more a
framework than an absolutely rigid design in many ways) but that the two
of them are currently more interested in being different that working
together.
But this has gotten rather long so I should sum up . . . . Novamente has
great promise -- but part of the reason why it has such great promise is
because so much of it *hasn't* been fully determined yet. The design is
still open enough that it can be stretched to fit many things. The
problem is that stretching it in some directions may/probably will make
it less adept at other things (jack of all trades/master of none) and it
may well be (and this is my primary complaint) that it is *so* general
that, while it could serve as the basis of an AGI, it is far more
complicated than necessary to do so (just as a bird's biology is not
necessary for flight). Thus, those blindly insisting that Novamente is
the be-all-and-end-all and that all other approaches should be abandoned
are not doing any of us a service. I want to see Novamente go forward
but we shouldn't put all of our eggs in one basket.
I want to say that this sums up my feelings pretty well, too, with some
minor differences or differences of emphasis.
My biggest issue with Novemente (and other systems) is what I have
generally referred to as the complex systems problem, which has a number
of specific manifestations, which Mark points to.
So, when Mark says
> ... there are some fundamental engineering
> features (scale-invariance of knowledge, ways of determining and
> exploiting encapsulation and modularity of knowledge without killing
> useful "leaky" abstractions, etc.) that aren't implemented yet in
> Novamente ...
... I would say that things like the "scale invariance of knowledge"
problem are an aspect of the CSP: the mechanism does not necessarily
work as expected when things scale up, because long-range dependencies
begin to take their toll and make the overall behavior impossible to
predict from the local mechanisms. It is a subtle point, but suffice it
to say that when I use the blanket term "complex systems problem" I
really mean this as a shorthand for a bunch of Gotchas that come along
when the proposed AGI gets up to such a size that subtle interactions
between its components start to dominate the behavior.
Another example would be the reliability of inference control engines:
inference itself is designed to make sure that truth preservation
happens, but the control structures that govern the actual extent of the
inferences carried out are the things that actually govern the overall
intelligence of the system.
Anyhow, it is this fundamental set of issues that constitute my problem
with Novamente (and other systems): I think they are a manfestation of
something much, much deeper, so it would not be enough to just throw a
Ph.D. at each one separately.
Richard Loosemore
-----
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=8660244&id_secret=64173188-ec64e4
