Page 6 - EE Times Europe Magazine | June2020

6 EE|Times EUROPE

OPINION | ARTIFICIAL INTELLIGENCE

The Practicality of

Predictability

By Dennis Goldenson

Over the last several weeks, I’ve had a lot of time (as have many
others) to think about Covid-19. I’ve wondered whether any
artificial-intelligence technology can predict outbreak patterns and
warn us of a pandemic’s intended path. While many brilliant epide-
miologists are searching for a coronavirus cure, other researchers are
considering how AI can be effectively utilized to simulate and predict
how diseases will spread and how diseases can be contained. This is
the art of practical AI and the merging of science and technology to
predict the needs of a public health crisis worldwide.
Part of the science in predicting is the ability to predict in real time, based on unplanned
scenarios or across various internal and external environmental conditions. Our machines must
be able to adapt and respond like humans in order to provide more spontaneous and accurate
responses, especially in times of dire need.
There is no shortage of research on how artificial intelligence is making all our connected
machines smarter and more intuitive. However, it’s important not to overlook one of the most The other area of exploration is advancing
important aspects of AI: the ability to predict specific outcomes and anticipate trends to help AI contextual and adaptive reasoning, thereby
prepare for various conditional factors (e.g., pandemics). providing machines with the capacity to
There is also a need for AI and machine-learning systems to go beyond pattern recognition react to change by reusing existing data and
by providing correlations and beginning to address underlying causality. Can machines learn information for new and unfamiliar problems.
cause and effect? A strong key to AI success is its ability to react
dynamically to ever-changing context, select-
EVERYTHING HAS A PATTERN ing the best course of action. This, too, can be
Predictive analytics is the method of utilizing statistics, probabilities, data mining, and modeling applied to cause and effect in understanding
to project or make predictions. In basic terms, software will extract information to analyze what triggers logical thinking, reflection,
historical data trends and patterns to anticipate future trends. It’s sort of like linear or multiple explanation, and justification.
regressions on speed. We want our machines to use variables to learn a model and predict the
value of the response variable. While we can now predict future intent based on previous cus- AI FOR THE COMMON GOOD
tomer behavior and purchase patterns, the ability to project scientific outcomes in applications All of this comes down to prioritization and
such as health care is actually the most fascinating use case. the practical application of AI and predictive
analytics for the common good. This may be
ARTIFICIAL NEURONS the future of AI — training to understand not
The amount and quality of data are the crown jewels in the accuracy of these AI systems. The what we intend to do but why and how we
AI data should be verified against real-world outcomes to ensure its accuracy. With appropriate intend to do it for future application.
design and training, various data sources can be leveraged to expand the predictable nature Artificial intelligence and predictive
of unforeseen events. Much of the data we work with is unstructured; computers cannot glean analytics are powerful tools that hold great
much meaning from it. Unstructured data can be in the form of text, email messages, Word files, promise for the enterprise space, including
audio files, photos, video, and multimedia content. On the other hand, structured data — such health care, financial services, manufac-
as numbers, groups of words, and dates with defined length and format — is more tabulated and turing, and retail. The effectiveness in
usually requires considerable processing for computers to understand and interpret it. predicting outcomes and “outbreaks” will
Now, with the advent of neural networks and deep-learning architectures, we can derive more be contingent on the quality of data and
meaning from unstructured data. The algorithms cycle through these interconnected neural net- the ability to provide real-time analytics
works and layered processors like the 86 billion neurons in your brain. The data is passed from on performance, preferences, and pathways
one layer to the next in a cycle or in a feedback loop. This is the deep-learning approach through based on the relationship between cause
which the algorithms learn how to identify various objects. and effect. ■
It all comes down to training the neural-network algorithms to understand unstructured data
and to recognize the pieces of various items. For example, we can teach an algorithm to recog- Dennis Goldenson is director of artificial
nize new strains of a virus mutation by training it with microscopic images to identify various intelligence and machine learning at SAR
characteristics of an infectious agent. It can then begin to recognize unknown virus patterns Insight & Consulting, where he focuses on
with various components and characteristics. digital assistant platforms, natural-language
There are, of course, other training use case applications for artificial neural networks, such as processing, user interface technologies, IMAGE: SHUTTERSTOCK
emotion detection. AI algorithms are increasingly accessing neural networks to identify specific and machine learning to provide compre-
emotional responses and create insights that can affect behavior and how we think. This capabil- hensive coverage on AI market trends and
ity can be applied to cause and effect in understanding what triggers emotion. developments.

JUNE 2020 | www.eetimes.eu