have we moved on?

[Phil Henshaw-2]

Martin,

> Hey Phil,
>
> If I understand you correctly, I think you're very right.  The
> information we have about the world is behavior and
> appearances, and for
> most interesting things the mechanism is completely hidden
> from us.  We
> can observe inputs and outputs, but not the source code.  We can see
> fuel go in and motion come out, but can't see the engine, let alone
> anything else.

The trickiest piece is proving in a comprehendable and comprehensive way
that anything has any actual inside structure, largely invisible from
the outside.  Science is built on the ideas of control, after all, and
it almost asks science to violate its own principles to discover it.
It's a genuine conundrum, and my original path to figuring it out, TMS
(to my satisfaction) was by trying to make sense of all the stuff
approximation leaves out, among other things.  Now I take it simply that
loops of intermittent relationships are largely invisible except to
dogged cross checking, and to be found ALL over the place.  All kinds of
real things seem to grow from them, and you can circumscribe their
growth with a boundary you can be confident their event horizon lies
within, QED 'built and run from inside'.

Now there are perhaps lots of things that a computer environment might
allow to be built from the inside... even if there are deep differences
in the interactions of numbers and things...  I think it's early yet to
say whether autonomous things growing in computers will be of any use,
but that's how nature seems to do it.   That's part of what I was
suggesting a while ago, asking if anyone had tried 'composting' as part
of artificial environments.

> Perhaps the core of intelligence is coming up with models of
> the world
> and exploiting them.  That's a view that's right up my alley.
>
> But say that to most AI researchers, and they'll stare at you
> uncomprehendingly.  They want a well defined problem, such as
> using all
> users purchases at Amazon to suggest other purchases for a
> single user.
>     And they'll come up with an algorithm that makes good suggestions
> most of the time.  The idea that the computer should be
> trying to make
> sense of the world -- eh?  What are you talking about?  Or maybe "oh,
> that's that flakey research from the 60s and 70s.  We've moved beyond
> that."  I have a friend who does research in believable virtual
> characters, and he gets that.
>
> Best,
> Martin
>
> Phil Henshaw wrote:
> > Got it!   But it's like making your way through a maze by
> running into
> > walls.  There's no point in being disappointed and just sitting down
> > when confronted by them.   I think locating the walls helps, i.e.
> > finding the barriers and disconnects in our thinking.   I've been
> > focused on one in particular, the lack of any working theoretical
> > model of things organized from the inside.  I think that's
> where the
> > start may be.  We all suffer from a core intellectual
> deficit on that
> > account, to quote another post:
> >
> > "I think it's comes from the biological human view of the
> world.  The
> > basic structure of thinking comes from our being
> 'observers', locked
> > up inside a brain, each of us reconstructing an imaginary
> model of the
> > world around us from our own observations and experiences.  
> That's a
> > problematic viewpoint for relating to any other thing built
> the same
> > way, i.e. organized from the inside.  What's going on inside other
> > things is invisible from the outside, and our [brain]
> builds its whole
> > world view from an outside perspective!!   Given that handicap, it's
> > quite natural for there to be more than one might guess
> missing from
> > our awareness."
> >
> > "...The theoretical sciences don't even have an image of anything
> > organized from the inside!  That part of the world is
> invisible to us
> > and so we're structurally unaware of the internally
> organized systems
> > we're part of and surround us.  It's ridiculous to work
> with a world
> > composed of several billion original, different and faulty
> universes,
> > but I think we're stuck with it and should try poking around to see
> > what other surprises there may be!  :)"
> >
> > make any sense?
> >  
> >> Phil Henshaw wrote:
> >>> I was curious about the film you were talking about, "Mind in the
> >>> Machine", and Googled it, coming across several things
> >> including its
> >>> origin and a simple statement by an Australian journalist (quoted
> >>> below) of Turing's idea of the test one would apply to
> >> measure success
> >>> in reproducing intelligence.
> >>>
> >>> I read the statement as saying if you're able to imitate
> >> something by
> >>> some other means (say behaviors of people by computers), in
> >> a way that
> >>> an observer doesn't notice the discrepancy, you've made the real
> >>> thing. I expect that's not quite accurate, and the current
> >> thinking has
> >>> evolved.   Can anyone say where the concept is headed?  
> >> The field of Artificial Intelligence no longer talks at all about
> >> general intelligence, the human mind, or anything like that.  
> >> The lone
> >> exception might the the natural language community, who of
> course are
> >> try to replicate something human specific.  But they still
> don't talk
> >> about "human equivalence" or anything like that.
> >>
> >> After the hype for AI in the 60s and 70s, there was a
> backlash in the
> >> 80s.  Kind of what happened to ideas like "virtual
> reality" or "dot
> >> com."  In search of respectability, AI has become largely applied
> >> statistics and focused on near term results.
> >>
> >> For someone like me who wants to explore principles and
> methods that
> >> point the way to full intelligence, this is all very
> >> depressing.  Like
> >> wanting to study cognitive psychology during behaviorism.
> >>
> >> Best,
> >> Martin
> >>
> >>
> >
> >
>
>

[Pamela McCorduck]

Artificial intelligence as a field "talks about" just about  
everything.  We mustn't confuse what the funding agencies demand  
(who, after all, call the tune in most instances, about what  
direction research will take) and what the scientists would wish, or  
even talk about quite publicly.  If you read the presidential essays  
of each new president of the Association for the Advancement of  
Artificial Intelligence, the main professional society of AI, you  
will see wonderful proposals for what the field might be and how it  
might be done.  DARPA wants reliable robot cars, thanks anyway.

There's a melancholy moment at the end of the film where Marvin  
Minsky talks about the grand ideas of the founding fathers of AI, and  
the way the field has instead become fixated on incremental  
improvements in performance--owing to that's where the money is.

I've noticed that people often scoff because the game of Go has never  
had a real AI challenge.  I agree that Go is wonderfully complex,  
difficult, a tough nut to crack.  But research toward an automatic Go  
machine has been the sole province of non-funded amateurs for sweet  
forever.  It's possible--not necessarily guaranteed--that a major,  
well-funded effort might crack the problem.  Nobody who has any money  
could imagine what use it would be, unfortunately.

So for big ideas: for the first fifty years of AI, the dream was to  
build a killer chess machine.  Why?  Because this was considered the  
sine qua non of intelligent behavior.  Never mind that you wouldn't  
particularly want a chess master as your dinner partner; this  
reflected our view of what intelligence was at the time.  We have our  
killer chess machine and we (and the chess players, Kasparov says)  
have learned a lot from the effort.

But the grand goal for the next fifty years is a robot soccer team  
that will defeat a human team in the World's Cup.  Think of what this  
means: planning, cooperating with other autonomies, kinetic  
intelligence, real-time calculations, and so forth.  It seems to me a  
worthy successor to the chess champion.  If I'm lucky, it will happen  
sooner than fifty years, and I'll get to see it for myself.  If not,  
not.

And, FWIW, this idea for a grand challenge bubbled up from workers--
young workers--in the field, and was not proposed by a funding  
agency.  Other grand ideas are being pursued on a shoestring by other  
young researchers.  I can talk about them, OR--you can buy the new  
edition of my Machines Who Think, which addresses some of these  
contemporary issues.

There will be more to talk about when I show the film in December.

Pamela





"For some reason the most vocal Christians among us never mention the  
Beatitudes.  But with tears in their eyes they demand that the Ten  
Commandments be posted in public places.  And of course that's Moses,  
not Jesus.  I haven't heard one of them demand that the Sermon on the  
Mount, the Beatitudes, be posted anywhere."

  Kurt Vonnegut, "A Man Without A Country"


-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/friam_redfish.com/attachments/20060902/2ab7c6e4/attachment.html

[Phil Henshaw-2]

Right, funding makes a big difference, but ever if we spend part of our
days serving bean counters does it control the direction of scientific
thought?     I think it's often the other way around too, that what
scientists find interesting to explore is what they're able to sell.
In that regard, helping Government, and others, learn what to expect
from uncontrolled systems could be just as interesting and marketable as
offering limited ways to control them.   You could present it as
advances in 'steering', like how to read ahead on the curves so your
mid-course corrections can  be early, small and graceful rather than
late, large and clumsy... but then, government does seem to enjoy the
latter so very much, perhaps we couldn't persuade them to give it up! :)


Artificial intelligence as a field "talks about" just about everything.
We mustn't confuse what the funding agencies demand (who, after all,
call the tune in most instances, about what direction research will
take) and what the scientists would wish, or even talk about quite
publicly.  If you read the presidential essays of each new president of
the Association for the Advancement of Artificial Intelligence, the main
professional society of AI, you will see wonderful proposals for what
the field might be and how it might be done.  DARPA wants reliable robot
cars, thanks anyway.  

There's a melancholy moment at the end of the film where Marvin Minsky
talks about the grand ideas of the founding fathers of AI, and the way
the field has instead become fixated on incremental improvements in
performance--owing to that's where the money is.

I've noticed that people often scoff because the game of Go has never
had a real AI challenge.  I agree that Go is wonderfully complex,
difficult, a tough nut to crack.  But research toward an automatic Go
machine has been the sole province of non-funded amateurs for sweet
forever.  It's possible--not necessarily guaranteed--that a major,
well-funded effort might crack the problem.  Nobody who has any money
could imagine what use it would be, unfortunately.

So for big ideas: for the first fifty years of AI, the dream was to
build a killer chess machine.  Why?  Because this was considered the
sine qua non of intelligent behavior.  Never mind that you wouldn't
particularly want a chess master as your dinner partner; this reflected
our view of what intelligence was at the time.  We have our killer chess
machine and we (and the chess players, Kasparov says) have learned a lot
from the effort.  

But the grand goal for the next fifty years is a robot soccer team that
will defeat a human team in the World's Cup.  Think of what this means:
planning, cooperating with other autonomies, kinetic intelligence,
real-time calculations, and so forth.  It seems to me a worthy successor
to the chess champion.  If I'm lucky, it will happen sooner than fifty
years, and I'll get to see it for myself.  If not, not.

And, FWIW, this idea for a grand challenge bubbled up from
workers--young workers--in the field, and was not proposed by a funding
agency.  Other grand ideas are being pursued on a shoestring by other
young researchers.  I can talk about them, OR--you can buy the new
edition of my Machines Who Think, which addresses some of these
contemporary issues.

There will be more to talk about when I show the film in December.

Pamela






"For some reason the most vocal Christians among us never mention the
Beatitudes.  But with tears in their eyes they demand that the Ten
Commandments be posted in public places.  And of course that's Moses,
not Jesus.  I haven't heard one of them demand that the Sermon on the
Mount, the Beatitudes, be posted anywhere."

 Kurt Vonnegut, "A Man Without A Country"




-------------- next part --------------
An HTML attachment was scrubbed...
URL: /pipermail/friam_redfish.com/attachments/20060902/e7b9bb13/attachment.html

[Martin C. Martin-2]

Hey Phil,

Phil Henshaw wrote:

> Martin,
>> Hey Phil,
>>
>> If I understand you correctly, I think you're very right.  The
>> information we have about the world is behavior and
>> appearances, and for
>> most interesting things the mechanism is completely hidden
>> from us.  We
>> can observe inputs and outputs, but not the source code.  We can see
>> fuel go in and motion come out, but can't see the engine, let alone
>> anything else.
>
> The trickiest piece is proving in a comprehendable and comprehensive way
> that anything has any actual inside structure, largely invisible from
> the outside.

Well, I'd argue something slightly different.  We need a model of what's
going on inside, but that's not the same as recovering what's actually
going on inside.  In fact, a high level model may be more useful and
important than a low level one.  For example, I can come up with the
concept of pressure, temperature and volume for a gas without
discovering molecules.  I can do all kinds of useful and interesting
things knowing only about pressure, temperature and volume, like make
air conditioners and refrigerators, and have no idea whether gas is
continuous or made of molecules.

As another example, there tends to be more traffic on the roads during
morning and evening rush hour.  This is an emergent phenomenon, and
would be hard to prove starting from a wiring diagram of the human
brain, plus whatever else about the environment you'd need to know.

So I don't see the job so much as recovering the actual structure that's
inside, but discovering regularities in the observables.

- Martin

[Phil Henshaw-2]

Martin,

>
> Hey Phil,
>
> Phil Henshaw wrote:
> > Martin,
> >> Hey Phil,
> >>
> >> If I understand you correctly, I think you're very right.  The
> >> information we have about the world is behavior and
> >> appearances, and for
> >> most interesting things the mechanism is completely hidden
> >> from us.  We
> >> can observe inputs and outputs, but not the source code.  
> >> We can see
> >> fuel go in and motion come out, but can't see the engine,
> >> let alone
> >> anything else.
> >
> > The trickiest piece is proving in a comprehendable and
> > comprehensive
> > way that anything has any actual inside structure, largely
> > invisible
> > from the outside.
>
> Well, I'd argue something slightly different.  We need a
> model of what's
> going on inside, but that's not the same as recovering what's
> actually
> going on inside.  In fact, a high level model may be more useful and
> important than a low level one.  For example, I can come up with the
> concept of pressure, temperature and volume for a gas without
> discovering molecules.  I can do all kinds of useful and interesting
> things knowing only about pressure, temperature and volume, like make
> air conditioners and refrigerators, and have no idea whether gas is
> continuous or made of molecules.
>
> As another example, there tends to be more traffic on the
> roads during
> morning and evening rush hour.  This is an emergent phenomenon, and
> would be hard to prove starting from a wiring diagram of the human
> brain, plus whatever else about the environment you'd need to know.
>
> So I don't see the job so much as recovering the actual
> structure that's
> inside, but discovering regularities in the observables.

Well, that's what people have always done, of course, not bothered with
the details that are hard or impossible to consider and made up
something else that serves a current purpose.   I call it
'approximation', and it's what science is mostly made of.   It can be
extremely useful.   There are also times when overlooking the fuzzy or
unavailable bits actually does matter and approximations don't help.
One example would be a passive system, behaving in response to external
controls, which is highly sensitive to initial conditions.   Another is
a natural system that isn't passive and has an independent internal
behavior and design.  That includes living things, and many other kinds
of growth systems.  There are several kinds of good proof.

Where you have an active system with an interior, the loops of
relationships it's built from are especially hard to see, inherently
hidden as a consequence of only directly connecting to themselves,
internally.  They also tend to be too complicated, immeasurable and
intermittent and original to figure out by indirect means. That also
makes them 'approximately' non-existent if that's how you choose to
interpret the hopelessness of your data on them.   That a great many
things highly important to us in the world are systems with active
interior behavior and design is not yet something physics seems quite
willing to admit, which makes it even more difficult to study them.
Still, you can prove it by watching them build.  Of course, we've also
been told never to study individual things, since whatever they do
that's interesting doesn't apply anyway.

I'm coming at this from having focused on trying to understand what
approximation leaves out.   I found quite a lot, including that closely
watching things that grow  will usually expose classic individually
unique developmental processes.   There's only one place to say that's
happening, i.e. inside.   Approximation helped show me where to look,
but only when I looked for what was left out and missing...  

There's obviously more to this, but the point is that the first step in
learning about something deeply hidden is not building a model that
looks familiar.  It's figuring out how to watch the detailed workings of
what you presently can't see.


Phil


> - Martin
>
>