Written by Shahrokh
Shahidzadeh, CEO at Acceptto
It's estimated that artificial intelligence (AI) will be
a $47 billion industry by next year, and therefore
it's no surprise that AI is a hot topic amongst CISOs and IT security
professionals. But for many, AI can often times be a term thrown around without
complete understanding. There is still a lot to be learned when it comes to
artificial intelligence. But starting off, the basic concepts still cause some
confusion.
A helpful way of grasping the concept of AI is to describe
AI as the capability of a machine to imitate intelligent human behavior. It's a
branch of computer science or a kind of computer system. What makes this system
special is its ability to perform tasks that usually require human
intelligence. These include decision-making, translation, and visual
perception.
Many people don't realize that the concept of AI has been
around for a long time. Forms of AI exist as far back as Greek mythology, in
stories of ‘mechanical men' that are described to mimic the behavior of humans.
Along with this, the very first engineers were known to understand their job,
in part, as attempting to create mechanical brains.
Our understanding of technology and neuroscience has come
leaps and bounds since these early days. As a result, the concept of what
actually constitutes AI has changed drastically. Today, AI offers a pretty
legitimate human-to-machine interaction. Machines are moving toward an ability
to connect data points, understand requests, and even draw conclusions.
Inference vs.
Prediction
One basic way of staying ahead of the curve when it comes to
understanding AI is the difference between inference and prediction. Often
people will confuse prediction with inference. While the differences are
actually very subtle, they have magnitudes of significance when it comes to
making decisions on who, when and how one should be accessing your information
infrastructure. CISOs implicitly understand that it only takes one bad actor to
gain access to their information assets and cause millions in damage. Therefore, how identity
authentication solutions make decisions on if the user is truly who they say
they are, versus merely identifying a bad actor impersonating a valid
credential, can mean the difference between safety and remediating cyber
damage.
What Is Inference?
Inference is simply a way of you asking yourself questions
to figure something out in order to reach a conclusion, but you are not always
able to confirm the results by the end of the situation.
According to a video from the Johns Hopkins University
course on "Managing Data Analysis,"
the goals of inferential questions include:
- Association
backed with outcome and key predictor while adjusting for confounders
- Single
or small number of key predictor(s)
- Sensitivity
analysis to check associations
It may be better to understand why we need to make
inferences:
- Inferences
help the source algorithm understand things the
target wants them to know but do not directly relate to situation. As it
relates to artificial intelligence and machine learning (AIML)
authentication, this is the identification of cyber applications and
associated hardware.
- Inferences
help the source algorithm to understand behaviors, what they may have done
in the past and what they may do next. As it relates to AIML
authentication, this is the cataloging of behavior patterns with those
cyber applications and associated hardware.
- Inferences
help the source algorithm draw logical conclusions about what is
happening. As it relates to AIML authentication, this is the determination
of whether or not the credential being used is actually the one intended for
use based on previous behaviors.
Inferences alone aren't an adequate method of determining
immutable identity for cybersecurity authentication. An effective solution will
also understand how to make predictions.
What Is Prediction?
Predictions are simply a way of you asking yourself what
will happen next and confirming your thoughts by the end of the situation.
Going back to the Johns Hopkins University video, the goals
of predictive questions include:
- Develop
a model that best predicts the outcome
- Use
all available information
- No
predictions favored over the others
- Little
focus on mechanism
Ultimately, we make and confirm prediction to better understand the
complexity and entirety of the situation. As it relates to identity authentication,
it is a way of building up a knowledge base of accurate inferences and learning
from inaccurate inferences to better predict immutable identity authentication
with less drag (need for further authentication) in the future.
Though our understanding of AI still contains several holes,
it is becoming (and really, already has) such an impressive and regularly used
part of today's technology world that it's best to jump on the learning wagon
now. AI learning is the future and it only makes sense to build skill in this
arena. A lot of possibilities are available with its implementations and its
benefits span from industry to industry.
##
About the Author
Shahrokh Shahidzadeh
leads a team of technologists, driving a paradigm shift in Cybersecurity through
Acceptto's Cognitive Continuous AuthenticationTM. Shahrokh is a
seasoned technologist and leader with 27 years of contribution to modern
computer architecture, device identity, platform trust elevation, large IoT
initiatives and ambient intelligence research with more than 20 issued and
pending patents. Prior to Acceptto, Shahrokh was a senior principal
technologist contributing to Intel Corporation for 25 years in a variety of
leadership positions where he architected and led multiple billion-dollar
product initiatives.