First, I repost what I just sent to a list deeply involved in
policies which will affect the future of internet/AGI/IOT:
Could
it be that lack of understanding of COI mathematics for intelligent systems
might be the most serious threat to human survival itself at present, because
of how it connects to design of internet and to the challenges to building true
human-centered internet?
This
question hits me quite forcibly as I prepare for a plenary talk on the future
of internet/AGI/IOT for the next WCCI conference. Since about 1990, WCCI was
the lead conference in the world developing true AGI technologies like the
"deep learning" revolution which has only just begun a radical
remake of life on earth. (See the one page abstract attached for the overview.
)
But
will it raise the net value of having an internet or -- following recent trends
- will it set in motion a chain of events which simply kill us all in the end?
There
is lots of buzz out there in policy circles, but they still seem almost totally
oblivious to the biggest, most implacable looking modes of instability
and collapse, and to the fundamental mathematical principles which
could save us, with more work (which no one is doing now.)
Of
these... I would say that the concrete requirements to prevent a COI based
collapse are the most important. Unmanaged COI effects through both the human
and the app components can take many forms. (Stability problems tend to be like
what.) When lawyers or coders focus too exclusively on special cases, they can
easily generate systems easily flooded out.
Sadly,
this email, written on a small tablet as I travel to wcci2022, cannot do full
justice into the pieces I already know about risks and solutions, let alone the
essential new crossdisciplinary research not being done. Just a few key data
points.
As in
my sii proposal.. new platforms for integrated markets, designed to
harmonize many apps and human players, are essential. The experience of the
teams designing new integrative markets for electric power transmission systems
is one crucial part of what we all need to understand and build on better.
One
small but decisive part of that learning experience was the creation of
something called Sarbanes Oxley. A clear new set of rules constraining
information flows was necessary to prevent collapse, due to new modes of
instability resulting from New market flexibility (which is ever more necessary
in managing a system which is ever more complex demanding paths to allow
new technologies).
This
example is closely related to proving convergence to correct results in DHP, an
important RLADP technology discussed for example in chapter 13 of the Handook
of Intelligent Control. (That chapter is posted at werbos.com/mind.htm .)
Unrestricted inputs to all major components of an otherwise intelligent system
really can cause collapse. COI is a matter of mapping out what the major
components are, and what flows of information are allowed.
This
can be done in principle with computerized markets (whether for goods or for
information) [Google asks for a market in gods, not goods. Not in this email!!]
in managing apps from a unifying RLADP market system rather than in managing
humans. When training feedback is channelled from human to human directly.. the
effects of corruption and bias are so overwhelming across all of human history
(as I can see even when walking down the street in so many places) that we
would all be dead long ago if previous centuries had had technologies like what
we are building now.
I say
that on a day when I walk through Ravenna... after many many other pivotal
places.
I
really hope someone in a more professional position will be willing and
able to do justice to what I am too old and limited to lead myself, much as I
will try to help if you do.
*****************************
I
mentioned just two examples. I hope they ask for some of the crucial details..
like three more important examples.
In my
talk to the IEEE sustainability conference (which Yeshua helped me with),
posted at build-a-world.org,
I noted how COI has turned out to be the number one obstacle to the
cost-effective new technology deployment which we would need, in order to stop
the present drift towards total extinction of humans, sooner than most people
know is coming. COI in today's governance systems in all major nations....
could be prevented by new governance/market integration platforms with the
necessary level of COI protection.
Second
-- climate issues illustrate information flow issues which were solved, in
nature, by the evolution of the basic mammal brain. (See attached abstracts
with links.). A key property is that neocortex is evolved based on error
measures like accuracy in prediction, quite different from what action choices
"are paid." Here in Italy, I am reminded of several great novels by
Stendahl (not to mention Dante) about intelligence organizations paid to pacify
the boss by felling him what he wants to hear. Simple separation of allowed app
types into types like networks of truth, like the neocortex university of the
brain, versus action final layoffs, already handles about half the problems.
Also
a technical detail. Even in classical econometrics, as in the text by johnson,
there is a technique called instrumental regression, aimed at preventing a type
of bias which can happen even in simple linear statistics, when inappropriate
inputs are included but not sanitized. But proper use of time, as in
werbos.com/Erdos.pdf, handles part of the problem more directly and
efficiently.
======
The
mathematical equivalences across different measures of value, and of feedback
or prices used to guide actirs within a larger system, play a crucial role. In
meditation today.. i think...what harmony or peace really require is correct credit
assignment in the larger system.
