Steve,
Glad to see you're using what I'm extremely confident should now be the
definitive term here - and a fundamental part of any definition of AGI - REAL
WORLD INTELLIGENCE (and real world reasoning) - as opposed to the **artificial
world reasoning** of narrow AI.
The next step is to realise that logic, maths and algorithms are all strictly
artificial world reasoning..They are extremely useful *structures* to *apply*
to - and structure - real world reasoning, just as the artificial Cartesian
grid is useful in structuring a great deal of modern real world depiction, from
photography to movies to painting.
However, they *have* first to be applied by real world intelligence, which is
of a different kind. And they are not at all essential for real world
reasoning. A robot that wants to get from one place to another, across a field,
say, will NOT have to use them - but will be able to work with a crude
diagrammatic/ image-schematic form of reasoning - just as real world humans and
animals do. The fact that they were only invented at the proverbial second to
midnight of evolutionary history should tell us that. They are
super-sophisticated refinements, not primal foundations, of intelligence.
The ultimate step is to realise that only REAL WORLD AGENTS - robots - are
capable of real world intelligence.
You can't understand anything about the real world, including language, if you
can't say (to some extent) - "BEEN THERE, DONE THAT." You can't understand
journeys (& all human/animal activities can be, and are, seen as journeys) if
you haven't actually taken journeys.
AFAICT - correct me, Steve - you are making the same mistake that all AI-ers
have made - saying "*this* form of logic/maths/algorithm doesn't work, so
let's have *another* form..." No form will work. They are all forms of
formulaic thinking, and there isn't a formula for living in, navigating or
reasoning about the real world - it's unpredictable by any formula.. That's the
crucial, and completely inescapable, conclusion if you want AI to enter the
real world.
From: Steve Richfield
Sent: Tuesday, July 03, 2012 2:53 AM
To: AGI
Subject: [agi] re Computing functions versus solving equations (calculating
versus physical execution)
Going back on-forum after off-forum email and phone discussions with Dorian.
Irrelevant arguments about experimental methodology aside, the main point here,
in which Dorian and I are in FULL agreement is:
There is LOTS of experimental and mathematical evidence that the things that
are going on in our brains are WAY WAY beyond the commonly held beliefs of
brain function, based on long past partial observations. Further, the failures
of decades of AI and NN efforts to achieve anything like real-world performance
is further evidence that our the commonly held beliefs are very wrong.
The essential point here is that "modern" computers operating to numerically
evaluate unidirectional functions is at best incredibly inefficient, and
generally useless for real-world intelligence. This is like attacking algebra
problems with a calculator.
Now, when you start focusing in on just WHAT is happening, what could be
happening, and what should be happening in an intelligent system - you start
stirring up arguments, because with zero funding for this. all that is
available is unfunded work like Dorian's and mine..
I have mentioned in the past that if you compute the equivalent gain*bandwidth
product for neurons, you start getting numbers that are about the same as early
electronic operational amplifiers, e.g. as used in early analog computers like
those from Heathkit. In short, wetware is fast, even compared with electronic
circuitry. Further, wetware appears to be adapted for direct simulation,
equation solving, and other operations that are more like calculus than the
arithmetic that "modern" digital computers do.
Dorian and I, and hopefully others are interested in figuring out how things
REALLY do and should work, rather than trying to replace calculus with
arithmetic, as the present AGI efforts are attempting to do. We understand that
there is absolutely NO WAY to convert the present crop of arithmetic-enamored
AGIers to this sort of thinking, so please don't interpret these remarks as any
attempt to redirect Ben, Richard, or any of the other AGIers here.
We would like to pull together other like minded people, probably onto a
different forum, because these beliefs don't fit well in this
arithmetic-absorbed culture. Is there anyone else here who is interested in
looking at the many sub-disciplines related to intelligence via bidirectional
computing (a special interest of mine), the value of fields to bias decisions
in localized areas (a special interest of Dorian's), self-organizing simulation
systems, closing the loops that involve wet lab experiments, mathematics, and
simulation, etc., etc.?
If you know others who might be interested in these and related approaches,
please pass this posting on to them.
Perhaps if I/we can bring some reality to this discussion, there might someday
actually be an AGI.
Any interest?
Steve Richfield
AGI | Archives | Modify Your Subscription
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968
Powered by Listbox: http://www.listbox.com
