Ah, but there is the rub. Do we have any *reason* to suppose that they
can be designed in such a way that (a) they produce the same powerful
operators which are responsible for intelligence, and (b) they
nevertheless do that without being complex?
Yes. Human beings exist.
Hey, I'm not saying "no!", I am just saying that we have never succeeded,
we have no theory that says that they can succeed, and it is also kind of
noticeable that whenever we try to make such systems non-complex, they get
out of our control and end up being complex anyway (plus they don't
work!).
I think that the reason they don't work is *because* they are complex. Once
we understand enough to make it so that they aren't complex, I think that
the problem will turn out to be surprisingly easy.
The only response I have ever heard to this question is: "I just don't
think it is going to be a problem". I need more than that.
Understood. Especially since "I just don't think it is going to be a
problem" is also said by the people who don't see or understand the problem
and then just blithely go and make complex systems which are then guaranteed
not to work. I understand what you say the problem is. I'm just saying
that I think that it is avoidable -- with the rational that the design of
human beings avoided it, so we can too.
Tricky. There may not be a clean version that works. And we do not find
it easy to figure out what the operators are. And, worst of all, nobody
is actually trying to figure out those human-cognitive operators (they all
say they do not need to copy human cognition at all).
I don't believe that you can make a complex version that works.
Yes, it is *not* easy to figure out what the operators are. It's going to
take a while. But we'll get there eventually. It took me about thirty
months to get through the ethics problem. The operators are going to be
worse but I think that it is going to be amenable to divide and conquer and
I believe that a lot of the work has already been done and ready to be
gathered up once you have a decent framework to hang it all on.
4. this operator system *MUST* be able to build operators
after-the-fact (and preferably, be able to manipulate core
operators with sufficient safety precautions)
Nope. It can build new ones, by more of the same sort of kludging. And
it does not have to be able to go back and manipulate old ones.
Nope. You resort to kludging and you *WILL* go complex and it won't work.
Stick to *YOUR* argument. Complex = No go.
5. with sufficient analysis and clean-up tools, you *can* get a
non-complex version of an intelligent system working (assuming
that you believe that an F-14 is non-complex i.e. an adaptive,
controllable system working in a complex world)
Why?
Consider: what if every act of cutting-edge creative thought (of the sort
that happens when someone mentions two unrelated phenomena, and all of a
sudden an analogy pops into your head that enables you to see a
commonality between the two things, where before you never thought that
they were related) is the result of new operators being spawned from old
by those immensely tangled, complex processes that are what builds
operators?
After all, this idea of "operators" finds its most obvious expression in
analogy-making.
What if the whole business of analogy-making is the result of building new
operators, and the only way we know of to do that is through the tangled,
complex mechanisms that humans probably do use to build new operators out
of old ones?
The whole business of analogy-making is the result of building new symbols
AND new operators. BUT the new operators are built according to
well-defined rules and their behavior in any given situation is eminently
predictable. I think that you are actually sometimes confusing complex and
complicated just like the rest of us.
(To be more specific, I suggest that operators are not built by an
"operator-builder' mechanism of the sort that you might be able to clean
up and reverse engineer, but instead operators are built by other
operators, so the more operators you have, the more ways you have to build
new operators .... which makes the stock of operator-building machinery
always increase).
Ever hear of an inductive proof? The "operator-builder" mechanism IS the
first set of operators. Of course the more operators you have, the more
ways you have to build new operators. That's why it's called Seed AI. But
there also, necessarily has to be a lot of pruning and storing off to
archive so that the operating-building doesn't become cancerous and grow out
of control.
If real, high-level human-equivalent intelligence is inextricably bound up
with the human ability to build analogies, and if analogy-making is as
complex as I just implied, how could you ever parse the complexity out of
the system?
Again, I think that you're confusing complicated with complex. The base
rules, I believe, are going to be very simple. What is going to make the
*system* entirely non-deterministic is it's interaction with the world --
but this happens with any sufficiently complicated system and is not an
argument that the system is complex.
6. you are far more like to get a clean, non-complex system working
than a horrible mess of kludges like the human brain (and it will
be *much* safer)
Hmmm.... not necessarily at all. But that is a big argument.
Why would you *possibly* think that a horrible mess of kludges is more
likely to work that a clean, well-designed system?
7. with sufficient analysis and clean-up tools available to the
system, it will *NOT* slowly diverge from stability as it
interacts with the world (the F-14 model)
I have to say I reject the F-14 model, since it depends on treating all
complexity as irrelevant noise.
OK, so here's my big return. What complexity is there that I, as a human
being, don't treat as irrelevant noise? I can handle complicated things by
divide-and-conquer. I handle truly complex things by a combination of
divide-and-conquer and approximation. Why can't my AGI do exactly the same.
We have plenty of evidence that complexity exists in the system: the boot
is on the other foot.
I don't believe so. Not at the level of rational thought. Yes, things are
certainly non-deterministic but human beings are composed of extremely
well-bounded systems (else we go off the rails and die). I say that you
need to prove your complexity exists argument. :-)
Can you?
-------------------------------------------
agi
Archives: http://www.listbox.com/member/archive/303/=now
RSS Feed: http://www.listbox.com/member/archive/rss/303/
Modify Your Subscription:
http://www.listbox.com/member/?member_id=8660244&id_secret=101455710-f059c4
Powered by Listbox: http://www.listbox.com
