Page 67 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

1. Introduction 55

computers were abandoned due to the SWSP limitation of digital computers in the
1980s. By the early 21st century, Google et al. have taken massively parallel distrib-
uted (MPD) computer nodes and matching deep learning algorithm to test on real-
time AlphaGo to beat Oriental Go game genius Lee Se-Dol by 4:1 in games on web,
in March 2016. AI has been paraphrased by younger generation leaders in this coun-
try: Stanford Prof. Andrew Ng [2] “as if new electricity empowered every indus-
tries,” MIT Prof. Lex Fridman “Artificial General Intelligence to capture human
intelligence,” and NYU Courant Institute professor Yann LaCun commented about
ANN Deep Learning [1]. These remarkable accomplishments have been reported in
YouTube (e.g., https://www.youtube.com/watch?v¼-GV_A9Js2nM&t¼1514s) and
elsewhere that we shall not take the space to cover again. In this paper we wish to
point to the future, the next gen AI or third Gen AI, or i-AI.
ANN Deep Learning: Silicon-based digital computing machines are necessary to
coexist with human Carbone analog computing and thinking, which we will argue is
based on analog possibility rather than digital probability. Possibilities occur over an
open set of occurrences based on frequency datasets. Automation machines are
needed that “understand” the human analog fuzzy thinking process; otherwise, ac-
cidents are bound to happen. Recently, in Phoenix, Arizona, the binary logic DAV
killed a pedestrian when the human was crossing at a red light. What is Fuzzy Logic?
It is the logic of possibility; the logic is never fuzzy; it is the occurrence frequency of
the parameters relative to a Fuzzy Membership Function (FMF) plotted on the open
set occurrence. The rule is made to sensibly be broken in a possible human behavior
that machines got to learn. We shall introduce the cardinal rule, as swear by “first, do
no harm” to pedestrian (or “primum non nocere,” the Latin translation from the orig-
inal Greek Hippocratic physician oath). Once again, the word pedestrian is defined
as the FMF that includes a ball-playing boy, a handicap seeing dog, a senior on a
wheel chair crossing the street with or without zebra lines.
We consider how to shorten the 13-years time span required to develop DAV at
level-4 automation, before the final 5 [3]. The 100-billion dollar DAV community
consists of NASA (Martian Cruiser); DoD/DARPA (Grand Challenge 2004e07,
Humanoid Robotics 2012), Google (Waymo), Uber (Driverless Car Volvo), Tesla
(Autopilot), Japan, Italy, etc. and US National Highway Traffic Safety Administra-
tion (NHTSA), where we can learn the state of the art. They have implicitly adopted
two attributes, namely (1) Poincare-von Neumann recursion-Ergodic ensemble
average, justifying replacing the long duty cycle time span needed for trials and
errors with the statistical ensemble average of ANN deep learning results; and
(2) Unsupervised learning (with unlabeled training data) gradient descents based
on the MaxwelleBoltzmann minimum free energy (MFE) cost function of Helm-
holtz. In this chapter we prescribe 6 W FMFs for the third Gen AI, or i-AI. The
ACS community has not considered, namely “making things fuzzier in the begin-
ning without presumptuous relevance judgment can become sharper creative at
the end.” This fuzzy logic approach may be called human creativity with the help
of “funneling orifice convergence effect.” In many engineering applications, it is
necessary to deal with human “fuzzy” or analog thinking in the linguistic sense,