Page 78 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

66 CHAPTER 3 Third Gen AI as Human Experience Based Expert Systems

generate statistically sensor awareness FMF. Their Boolean Logic union and
intersection helps the final decision-making system. The averaged behavior
mimics the wide-sense irreversible “Older and Wiser” “EBES.”
2. Massively Parallel Distributed (MPD) computing architecture (e.g., iPhones,
graphic processors units 8 8 8 which have been furthermore miniaturized
in a backplane by Nvidia, Inc.) must match the MPD coding algorithm, for
example, Python tensor flow, like the well-fit “gloves with hands.” We consider
unlabeled data unsupervised deep learning (UDL) which is based on BNN of
both neurons and glial cells, the experience-based expert system can increase
the trustworthiness, sophistication and DARPA explainable AI (XAI).
3. Wide-Sense Ergodicity Principle (WSEP): The WSEP is based upon
Boltzmann’s formulation of irreversible thermodynamics, that is, the Maxwelle
Boltzmann assignment of probabilities to physical problems (canonical prob-
ability P(x o )). We note that by introducing the following set of notions from
statistical physics: (1) analyticity of energy, (2) causality local minima, and (3)
replacing temporal ergodicity of an ensemble with spatial ergodicity of an
ensemble. This perspective highlights the mean Ergodic theorem established by
John von Neumann, and the pointewise Ergodic theorem established by George
Birkhoff, proofs of which were published nearly simultaneously in PNAS in
1931 and 1932. We can apply these two principles to elucidate a new principle
that can be applied to computer science to explain the success of deep learning.
One can introduce the concept of spatial average AI computational approach to
replace by the temporal averaging technique that Monte Carlo with noise is used
in modeling physics and engineering problem. The deep learning community is
based on the MPD fast computers; tightly match the MPD smart. Although a
single computer time averaged over large number of stochastic runs is equiv-
alent to ensemble average of thousands of computers in both the mean and
variance moments. The Ergodicity can read both sides, unfortunately, the life
time duty cycle is not long enough for quick decision such as DAV. We count on
the spatial ensemble side, by increasing 1000 to 10,000, or million, and the
results are stored in the Cloud databases. Moreover, a large Cloud database
provides thousands to million machines for enough possible initial and
boundary conditions. The Cloud provides a means for implementing the spatial
form of the AI ergodicity principle by replacing the temporal average with
sharing spatial averages over multiple machines to gain more experience more
rapidly. (Note: we discuss this further in the document, but reserve a subsequent
paper to provide a more analytical formation in mathematician’s language.)
4. Biological Neural Networks (BNN): BNN requires growing, recruiting, prun-
ing, and trimming of 10 billion neurons and 100 billion glial cells for the
self-architectures, house cleaning (by astrocyte glial cells) that can prevent
Dementia Alzheimer Disease (DAD) ([5] Szu, Moon). DAD is the fifth major
disorder among diabetics type II, heart attack, strokes, cancers for aging WWII
baby boomers. It takes six dimensionalities to avoid DAD; three in physical:
exercise, eat right, sleep tight; and three in mental: stimulating games, social