Page 40 - Kaleidoscope Academic Conference Proceedings 2020

simulation, emulation, modeling, statistics, and analytics,
have been the mainstay. They are still useful and many of the
processes built around them are hard to outperform. They
share one underlying philosophy, and that is to improve the
understanding of phenomena, effects, and outcomes by
explicitly trying to understand cause and effect. In that regard,
advances in mathematical methods, better instrumentation to
produce temporally and spatially higher resolution data,
fusion of data and information from many sources, and
greater computational power all contribute to improvements.
However, the greater the complexity the harder it has been to
make progress. Much of that progress has required the
dedication and attention of specialists and experts and lots
and lots of resources. It’s fair to say that such a focus has
allowed us to solve a few very hard problems by literally
pouring money at them. More often the efforts have been
confined to effectively make the greatest headway on the
easier, routine, and repetitive aspects of running an industrial
enterprise and its processes.

Artificial intelligence has gone through periods of great
enthusiasm, then neglect, followed by a rebirth a few times
over. We are in a period where there is an incredible
investment in AI, and it’s worthwhile to examine a few of
the key characteristics of what the term AI connotes and ask
why? I have included a few references [10], [11] that are an
attempt to capture the impacts that AI will have
economically, and for balance views that are worries about
how well AI will actually serve us [12].

As a label AI includes many different methods and
techniques (classification and categorization methods,
reasoning, knowledge, expert, and rule-based systems
among others). It is also identified with many areas of
specialization (planning, routing, machine vision, natural
language understanding, speech processing, robotics, and
many more). It also includes machine learning (ML) which
comes in a number of varieties (supervised learning,
unsupervised learning, deep learning, neural networks,
adaptive computing, and neuromorphic systems). Lastly,
there is what is referred to as “General AI”, which is about
artificial intelligence that can rival human thought and
display the properties of consciousness and has the ability to
deal with new situations creatively while displaying common
Figure 2 - The expectations for AI/ML and data sense. As previously discussed, General AI does not exist!
technologies and important subcategories in 2019 and 2020 We also have no inkling if it will and if it does when. In the
context of digitization, I would like to now narrow down to
4.1 Technologies in Focus - Artificial Intelligence just ML. There is something different here. What ML does
is fundamentally pattern recognition; unlike the methods we
Much of the value that comes from digitization in an developed in the past it does seek to find cause and effect.
industrial context is that it helps us make better decisions to The mainline ML algorithms have been around for quite
execute the right actions at the right time, and in the right some time but there are three advances that have made a
sequence, or that it allows us to create better experiences for tremendous difference in how well ML algorithms perform:
end users or customers. It also comes from how industries
optimize the management of time, assets, resources, and • The first is the ability to collect, store, and process
deployment of capital. The number of decisions and actions vast amounts of data. Not just any data but as
at a granular level, made in an industrial enterprise can complete and clean a set as possible to use in the
consist of tens of thousands or even hundreds of thousands training corpus for “teaching” and “tuning” a ML
of individual decisions; it is a difficult and extremely model. Data has to cover the full range of conditions
complex problem. The tools developed in the past, for that the input to the ML model will span.

– xxxvi –