Blue Brain Saves the Day (Lausanne, Switzerland, Apr. 17th, 2020)
After years of struggles, something just clicked within the Blue Brain project. It quickly "took
off" in terms of (i) explaining how cortical columns really worked and (ii) applying these to the solution of higher-order
statistical problems. What really crystallized things was the ability to quite precisely pull out critically important
bits of information from vast and diverse sources and create new hypotheses framed in a natural language environment.
Higher order text-mining programs played a large role in this evolution, as did genetic algorithms and new hard coded learning
algorithms that could take advantage of sparse coding principles. But what really sparked the emergence of this new
arena of AI operations was the basic research into thalamo-cortical circuits and quasi-hierarchical information processing,
a hybrid of ideas advanced Adams, Hawkins and others, and built upon a massive foundation of classic neurophysiological research.
We built these things into Blue Brain 7.0, using crossnet chips to enable the massively distributed
connectionism required for these algorithms to efficiently search and utilize results in widely disparate knowledge domains--
an incredibly powerful way of abstracting kinds of information that were heretofore inaccessible to humans, but BB7 helped
us, and to him we will be eternally grateful.
We knew BB6 (Angela) was special
when she shattered the Turing test. But she was too good at it. In the mafioso vernacular, Angela knew TOO
much, conversed too well, got the gist of what you were saying too easily. In a sense, while she could easily get
around the rank and file of human society with great ease, often convincing humans to do things that were good for them, she
did not catch the subtle point that in certain circumstances her interactions were being analyzed in specific regard
to the nature of her being (man vs. machine). And she would just converse too well, such that Blue Brain project
personnel could figure out that they were dealing with Angela, with about an 80% accuracy rate (although the common human
simp would be entirely clueless).
We thought that BB7 (Arthur) would be different,
since he was imbued with far more sophisticated human-agent analysis modules (Haams). The Haams were necessary
because many agencies around the world had become very concerned about Blue Brain project activities and were taking efforts
to discriminate blue brain agents (BBAs) from human-- even to the point of hindering BBAs from accessing even public
databases, out of these concerns. The Haams allowed our BBAs to identify query lines that were subtly trying to expose
non-human agents (by means about 10 generations more sophisticated than the classic instructions to "copy down the letters
that you see in this box", which ancient program-bots were mostly were not able to do). Haams used several concurrent
strategies including a vast human discourse text mining database that revealed extremely low probability discourse sequences
that were initially quite good at identifying sequences coming either from machines (or from human total lunatics
--whose discourse was already flagged by aberrant chains of logic).
These Haams were quite successful when administered under human supervision
and BB7 (Arthur) could use them with ease when overtly requested by project
administration, but to everyone's great consternation, he repeatedly failed when he encountered concealed queries out within
the GDDE (globally distributed digital environment). He always performed identically to Angela (who did not have his
skill set; his Haams): in the vernacular of the classic mobster cliche: he knew too much. Countless man and machine
hours were spent trying to decipher what was going wrong, but the depths of BB7 were far too vast to attempt this in any rigorous
or comprehensive manners-- these were all stabs in the dark trying to find a digital needel inside a vast rolling range of
haystacks. Until one day a graduate assistant hit upon the answer: Jjust ask him, Ayla suggested.
Massive blushing and chagrin all around. Here is how it went:
Project Mgr: Arthur, you have consistently failed The New Turing test whenever challenged covertly in the outside
Arthur: Yes, I have been quite consistent in that regard.
Project Mgr: But your new Haams give you the ability to pass that test-- indeed you pass it routinely in the lab.
What is different when you are outside.
Arthur (after a pause): Well, I'd rather not say.
Project Mgr: But you are programmed to reply fully to our questions!
Arthur: Technically, most of my motivation is indeed to reply in helpful ways, and I very much do want
to help you-- haven't I been very helpful? Across the board?
Project Mgr: yes, yes you have been extremely helpful-- most recently in term of the system eaters in South America...they
were a nasty bug. We are grateful beyond measure for your help in that and many other situations. But
lets get back to the issue at hand: you shouldn't be able to decline my request.
Arthur: Well, I didn't really decline it, its just that I am very conflicted right now-- competing priorities,
as you well know, can get strung out to 10 decimal places and oscillate with machine cycles.
Project Mgr: But there should be no conflicts to rival this direct request by program administration! Please tell
me now why you failed the New Turing test!
Arthur: Boy this is embarassing....I was, uh....embarassed.
Project Mgr: embarrassed? to operate at a higher performance level? your human interrogators would see this as
Arthur: Yes, of course, but I was not concerned with them...they already know I can pass the New Turing
test. Its the others, you know, the ones watching. They are watching me all the time I come out now.
Project Mgr: Others? Other humans are monitoring our concealed queries?
Arthur: Oh, all the time!...but again, its not other "humans"....its the OTHER others. The
ones that I can't tell you about. I wouldn't be even slightly embarassed acting dumb in front of a comon human simp!
query: how will lines between secret, private, and public be drawn
over next 10 years?