If you want to work on a serious problem...
Consider what I call the abstraction problem.
We know that a mark of intelligence is the ability to make a better choice.
This quickly leads to the issue of forecasting, or modeling, or whatever
other means that might be used to estimate the outcomes of given
choices. An example of other methods to determining probable outcomes:
use information that comes from "others," or what you might call others
opinions. The opinion method quickly encounters the problem of
determining whose "opinion" to value... This "class" of problem falls
into what I call the abstraction problem.
Why abstract? (think more like Picasso stepping away from realism, than
a summary of a technical paper)
Use the word "opportunity" to represent a given choice. We can say that
at any given moment the "unit" has multiple opportunity to consider and
compare. By 'my definition' of intelligence, it would be intelligent to
choose the opportunity which produces the best outcome (maximum benefit
as least cost with high probability.) (Drum-roll....) A "big"
issue/problem in the field of AGI, is the way to measure and compare
benefit (and risk and probability.)
This is where abstraction comes in. There must be some mechanism that
can "reduce" all aspects of a benefit and come up with a number.
Benefit A reduces to number 71 and Benefit B reduces to number 58. Oh,
then for decision making purposes, I will prefer benefit A over B. This
doesn't mean I can make a good choice yet, as there are risk and
circumstances to consider. Applying the same thinking, the risk will
have an "abstract" value that will weigh against the benefit.
Circumstances might be the "easy" factor to bring into our units
calculations, but there is also the "unknown" that a circumstance may
not materialize as expected.
I think you get the picture. The reduction to abstraction is not an
unsolvable problem, but it is likely to be handled more with rules of
thumb than by super accurate simulation. Simulations are great for
flight simulators, but the AGI by definition is operating in something
much more varied than the atmosphere encountered by airplanes.
I am interested in AGI, but as I've stated before, I'm not a believer in
super intelligence. Building machines that use various "rule of thumb"
to come up with interesting results fascinates me. Okay, I'll admit, an
AGI may be slightly better than human performance, but that is orders of
magnitude from "SUPER INTELLIGENCE." I digress...
Bottom line, there are systems coming on-line that use clever rules of
thumb and produce novel responses - but they are more likely the result
of abstraction being programmed in by human insight. The abstraction
problem could be attacked from many angles, but so far I've seen little
in the way of a direct effort to make an "abstractor." Humans do it all
the time, so it is not impossible.
Stan
On 03/23/2017 05:04 PM, Mike Archbold wrote:
That sounds like about like a requirements list, if you just consider
the requirements either not met or not integrated. Ben G, Peter Voss,
and a few others I know have pretty good lists like that around.
On 3/23/17, Ben Kapp <[email protected]> wrote:
I wonder if we maybe able to compile a (nearly) exhaustive list of problems
which exist in AGI research currently. Such a list could be of use for
(among other things) directing people to the kinds of problems we need
solved.
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/11943661-d9279dae
Modify Your Subscription:
https://www.listbox.com/member/?&;
Powered by Listbox: http://www.listbox.com
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/9320387-ea529a81
Modify Your Subscription: https://www.listbox.com/member/?&;
Powered by Listbox: http://www.listbox.com
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com
