Security and Artificial Intelligence: FAQ

Jeff Crume
Doug Lhotka
Carma Austin
Security and Artificial Intelligence: FAQ

Introduction
In the world of cybersecurity, we can all agree on one thing: Change is constant. We must
continuously review what we did yesterday and identify ways to improve. To keep up with our
persistent adversaries, we must constantly try new technologies in an attempt to find better
ways to defend or proactively prevent an attack. We must assess our policies and enhance our
methodologies daily. In short, if we aren’t improving, we will not be able to maintain the level
of security needed to safeguard what is most important.

This foundational principle of change is well understood by every security vendor. It seems
that every year the industry adopts a new buzzword geared towards selling a new technology –
we’re all guilty of looking for the silver bullet to our security challenges. Terms like
“actionable” and “automated” have flooded our inboxes over the last few years, and now we
are increasingly seeing trending buzzwords including artificial intelligence (AI), cognitive
computing and machine learning.

Many vendors today claim that AI is their secret sauce, and this may, in fact, be true. This
critical technological advancement seems to have blossomed overnight, but many industry
professionals are skeptical of the promises being made. Sadly, the AI buzzword bandwagon
has contributed to the confusion, thereby inhibiting the market from truly adopting the
technology. There is no question as to whether or not AI exists today. What it is, and what it
can do for us, are the real questions. Herein lies the purpose of this paper: To set aside claims
and aspirations and attempt to define what AI and its components are capable of doing for us
in the security space today.

Frequently Asked Questions

• What is artificial intelligence?
Artificial intelligence (or AI for short) is defined by Webster’s Dictionary as “a branch of
computer science dealing with the simulation of intelligent behavior in computers.”1
The Oxford Dictionary elaborates on this basic definition, calling it “the theory and
development of computer systems able to perform tasks normally requiring human
intelligence, such as visual perception, speech recognition, decision-making and
translation between languages.”2

• OK, but what does that mean in plain English?

AI essentially involves making computers more able to match or exceed human

1
Merriam-Webster, https://www.merriam-webster.com/dictionary/artificial%20intelligence
2
Oxford Dictionaries, https://en.oxforddictionaries.com/definition/artificial_intelligence
Security and Artificial Intelligence: FAQ
intelligence in its various forms, by mimicking the human ability to discover, infer and
reason.

• Isn’t AI just another marketing buzzword? Is it real?

AI has been used to hype a lot of things that stretch the definition of the term beyond
recognition. That said, some promises of AI have already been realized, while others
remain in the realm of research. Instead of thinking of AI as a single, monolithic
feature, it is better to envision it as a collection of related technologies that support a
common goal. For example, the following is a list of some of those capabilities that,
when taken together, make computers more intelligent:
o Reasoning and problem solving
o Knowledge representation
o Planning
o Learning
o Natural language processing
o Perception
o Motion and manipulation
o Social intelligence
o Creativity
o General intelligence3

• What is machine learning?

Machine learning is a subfield of AI and computer science that has its roots in
statistics and mathematical optimization. Machine learning covers techniques in
supervised and unsupervised learning for applications in prediction, analytics and data
mining.4 Machine learning can be (and often is) used independently of other AI or
cognitive technologies. In fact, this is the most prevalent type of AI we see today. Many
machine learning algorithms and techniques are already in use within various solutions
that look for patterns or anomalies in data.

• What is deep learning?

Deep learning is a relatively new set of methods that is changing machine learning in
fundamental ways. Deep learning isn't an algorithm, per se, but rather a family of
algorithms that implement deep networks with unsupervised learning. These
networks are so deep that new methods of computation, such as GPUs, are required to
build them.4

3
Wikipedia, https://en.wikipedia.org/wiki/Artificial_intelligence
4
M. Tim Jones, “A Beginner's Guide to Artificial Intelligence, Machine Learning and Cognitive
Computing,” June 2017, https://www.ibm.com/developerworks/library/cc-beginner-guide-
machine-learning-ai-cognitive/cc-beginner-guide-machine-learning-ai-cognitive-pdf.pdf
Security and Artificial Intelligence: FAQ
• Deep learning sounds great. Are there any limits to what it can do today?
Yes. Despite the results of applying deep learning algorithms, problems exist that we
have yet to solve. A recent application of deep learning to skin cancer detection found
that the algorithm was more accurate than a board-certified dermatologist. But, where
dermatologists could enumerate the factors that led to their diagnosis, there's no way
to identify which factors a deep learning program used in its classification. This is
called deep learning's black box problem, and presents challenges for model
validation, particularly in gaining regulatory approval. The evidence shows that it’s
accurate and effective, yet if we cannot describe how the decisions are made, how can
it be fully validated?

Another application, called Deep Patient, was able to successfully predict disease given
a patient's medical records. The application proved to be considerably better at
forecasting disease than physicians – even for schizophrenia, which is notoriously
difficult to predict. So, even though the models work well, no one can reach into the
massive neural networks to identify why.4

Deep learning has significant potential for a certain class of problem solving, but is not
suitable for all situations, is difficult, costly and time consuming to implement, and
works best when focused on a narrow target.

• What is cognitive computing?

Cognitive computing is a subfield of AI which builds on neural networks and deep
learning. It applies knowledge from cognitive science to build systems that simulate
human thought processes. However, rather than focus on a singular set of technologies,
cognitive computing covers several disciplines, including machine learning, natural
language processing, vision, and human-computer interaction.4 Of those, Cognitive
computing focuses most heavily on natural language processing.

• What is so special about cognitive computing?

Here’s an excerpt from a white paper by IBM’s Head of Research John Kelly,5 that
describes it well:

Cognitive computing refers to systems that learn at scale, reason with purpose, and
interact with humans naturally. Rather than being explicitly programmed, they learn and
reason from their interactions with us and from their experiences with their environment.
They are made possible by advances in a number of scientific fields over the past half-
century, and are different in important ways from the information systems that preceded
them. Those systems have been deterministic; cognitive systems are probabilistic. They

5
Dr. John E. Kelly III, “Computing, Cognition and the Future of Knowing: How Humans and
Machines are Forging a New Age of Understanding,” October 2015,
https://cra.org/crn/2016/09/computing-cognition-future-knowing-humans-machines-forging-
new-age-understanding/
Security and Artificial Intelligence: FAQ
generate not just answers to numerical problems, but hypotheses, reasoned arguments
and recommendations about more complex – and meaningful – bodies of data. What’s
more, cognitive systems can make sense of the 80 percent of the world’s data that
computer scientists call “unstructured.” This enables them to keep pace with the volume,
complexity and unpredictability of information and systems in the modern world.

• Is cognitive computing the same as AI?

Cognitive computing is an AI technology, but there are other aspects of AI that don’t
involve cognitive.

• How do all these AI technologies relate?

Deep learning is a special type of machine learning. Cognitive computing uses both. All
three are examples of AI technologies. Here’s a stylized Venn diagram to illustrate.

• What is IBM Watson®?

Watson is IBM’s AI, machine learning and cognitive computing platform. It provides a
wide range of AI technologies to process both structured and unstructured information
from a wide range of sources, understand what they mean, and add them to its body of
knowledge (a.k.a. corpus) for subsequent use. Some components of the platform can
then respond to complex questions using natural language based upon this corpus to
augment human intelligence and improve insight and efficiency in critical tasks. Other
components provide machine learning and deep learning capabilities for use on
structured data.

Watson Platform Solutions are not a single product, but rather a suite of integrated
components that have specialized analytical capabilities that can be leveraged as
individual components to build AI-enabled solutions. While all involve AI, some provide
cognitive capabilities, many use machine learning or deep learning, and some use more
traditional analytics.
Security and Artificial Intelligence: FAQ

• Sounds good, but can AI and cognitive computing be used to actually solve real
world problems?
Yes. For example, IBM Watson is being used at the University of North Carolina at
Chapel Hill Cancer Center in order to identify and recommend treatment options for
patients who have not responded to standard therapies. With roughly 8,000 new
medical research papers published every day, it is impossible for a doctor or even a
team of doctors to keep up with all the latest developments. Watson, however, can
consume all this new research and then quickly apply it to improve patient outcomes.
In a test of 1,000 patients, Watson’s recommendations matched those of the experts
99 percent of the time. More importantly, Watson found other treatment options in
about 30 percent of cases which had not been identified by the experts.6

This is just one of many examples. Another is how IBM is now applying AI and cognitive
technologies to the cybersecurity space, in order to allow organizations to identify
threats and respond more quickly. Watson for Cyber Security has ingested over 2 billion
documents in the corpus and is adding thousands more every day. It’s reduced the time
to analyze an incident from hours to minutes, greatly accelerating mitigation and
reducing the impact to the organization.

• Will AI make humans obsolete?

No. The goal of AI is to augment human intelligence – not replace it. There are still
significant limits to what cognitive technologies can do, especially in the area of
decision making, where humans are able to weigh factors that can’t easily be expressed
in algorithmic terms. AI is a tool. Like other tools, AI amplifies the work that people do.
AI processes vast amounts of unstructured information into a coherent whole, resulting
in greater efficiency and insights.

• Will AI become sentient and take over the planet?

You’ve been watching too many sci-fi movies.

• Is AI Big Brother?
Like any technology, AI can be misused. For example, Watson for Cyber Security needs
to be trained with data to provide insights for identifying threat and attack scenarios.
You have complete control to configure what type of data is sent to Watson for Cyber
Security for analysis. We know that the bad guys are interested in abusing AI for their
own ends, but we’re far ahead of them at the moment. AI is just a tool that can be
leveraged for good or malicious work.

6
“Artificial Intelligence Positioned to Be a Game-Changer,” 60 Minutes, CBS News, October 9,
2016, https://www.cbsnews.com/amp/news/60-minutes-artificial-intelligence-charlie-rose-
robot-sophia/
Security and Artificial Intelligence: FAQ

• What are predictive analytics?

As its name implies, predictive analytics is a branch of advanced analytics which are
used to make predictions about unknown future events. Predictive analytics extracts
information from data using many techniques from data mining, statistics, modeling,
machine learning and artificial intelligence to analyze current data to make predictions
about the future.7

• What can predictive analytics do? What are the limitations?

The answer here is “it depends.” The predictive power of analytics derives from the
quality and relevance of the data set being used, as well as the accuracy of the
algorithm processing that data, and may even extend to the capabilities of the
individual interpreting the insights gleaned from the data. That said, some events are
easier to predict than others, and some first-of-a-kind events may never be predicted
(this type of event is sometimes referred to as a “black swan”).

For instance, in a security context, a zero-day attack would be very difficult to predict,
whereas a widespread malware campaign would be much easier. In other words, even
though the name is “predictive,” these techniques actually provide probabilities of a
future occurrence, and by definition, black swan events cannot be predicted.
Unfortunately, security black swans are not that uncommon. Combine this with the fact
that predictive analytics for security is still quite new, so results vary, and expectations
should be conservative.

• What is threat hunting?

In the context of cybersecurity, threat hunting involves searching through large
volumes of data to identify bad actors and threats to an organization’s IT infrastructure.
The goal is to prevent attacks before they happen and eliminate or minimize their
effects. Threat hunting tools may ingest threat intelligence feeds, vulnerability analysis
reports, risk assessments, malware analysis, HR employee records, security event data,
system logs, social media feeds and more.

While threat hunting leverages tools, much of the work is manually driven by an
investigator who researches the answers to questions they develop proactively. For
instance, they may choose to look for people in a particular organization with access to
highly sensitive resources, who have recently expressed a negative sentiment toward
the organization, as they could be potential insider threats. There are, of course,
limitations to this sort of activity, since it involves a great deal of factors to try to detect
anomalous activity and then predict future events. Because such activity is human

7
“What Is Predictive Analytics?” Predictive Analytics Today,
https://www.predictiveanalyticstoday.com/what-is-predictive-analytics/
Security and Artificial Intelligence: FAQ
intensive, expectations should be set appropriately. There is no magic AI technology
that can find a rock in a large ocean (or swamp) of data on a consistent and reliable
basis.

IBM Security and AI

• What AI offerings does IBM have?
Watson for Cyber Security is an instance of IBM's cognitive computing capability
focused specifically on the cybersecurity space. It consumes structured security
information from threat intelligence feeds, security events and related data, as well as
unstructured sources such as research papers, security blogs, websites and advisories.
It then stores this as a massive corpus (i.e., knowledge base) consisting of more than
10 billion elements and refreshes its understanding at the rate of 4 million more
elements each hour. In a sense, Watson for Cyber Security is like a security expert who
reads the web 24x7, never forgets, formulates hypotheses about attacks based upon
this highly dynamic knowledge base and gets smarter over time. Today, Watson for
Cyber Security can be accessed via QRadar Advisor with Watson.

QRadar Advisor is an application which extends IBM's QRadar Security Information

and Event Management (SIEM) platform by sending indicators of compromise to
Watson for Cyber Security for processing, which then returns insights into the nature
and extent of cyber attacks. This app merges the security analytics in QRadar (which
itself uses AI technologies like machine learning) to discover anomalies, along with the
reasoning skills and knowledge base of Watson for Cyber Security (which is cognitive
focused), to provide rapid analysis of the full scope and context of an offense.

i2 Enterprise Insight Analysis can be used to perform threat hunting and

investigations. i2 QRadar Offense Investigator integrates the QRadar SIEM
capabilities with i2 in order to improve the efficiency of the investigations.

• Where else in its security portfolio does IBM leverage AI technologies?

We are actively infusing AI capabilities across the portfolio. For example, IBM QRadar
has used machine learning to identify security anomalies in the environment for many
years. More recently, the QRadar User Behavior Analytics app leverages advanced
machine learning to detect changes in user behavior and assign risk scores.

IBM MaaS360 uses other AI technologies in the Watson family to provide insight into
mobile endpoint vulnerabilities from unstructured data.

IBM AppScan Source leverages machine learning to reduce false positives when
looking for potential security vulnerabilities in source code.
Security and Artificial Intelligence: FAQ
• Is this stuff real? Is anyone actually using cognitive capabilities to improve their
security posture?
Yes. IBM customers using QRadar Advisor with Watson typically see an improvement in
their security posture because they are able to complete investigations faster, more
thoroughly and more consistently. They are also able to get through their daily backlog
of events – even with a smaller, less experienced team of analysts.

Here are some real quotes from real customers.

• Can security response be automated?

Yes and no. For example, malware attack responses have been automated for years,
with generally good results. Typically, this involves moving infected files into a special
quarantine container to prevent further dissemination. Even then, there are regular
reports that a signature identifies an operating system or other application component
as malicious, which results in system outages.

Well-known security expert Bruce Schneier put it this way:

Incident response is fundamentally uncertain and that makes it hard to

automate. For a whole lot of reasons. All attacks are different. All networks are
different. Security environments are different. Organizations are different.
Regulatory environments are different. In a lot of cases the political and
economic considerations matter a lot more than the tech.8

Fundamentally automation requires certainty, and the one thing certain in security is
variability. For specific cases where the downside risk is small (e.g., isolating one user’s

8
Bruce Schneier, “Security and Privacy in a Hyper-Connected World,” InterConnect 2017.
Security and Artificial Intelligence: FAQ
machine), automation may be appropriate. For cases where the risk of disruption is
high, such as in electrical utilities or healthcare delivery, automation should be used
sparingly and with caution. Best practice at the moment is to have a human in the loop
for all critical situations. For these reasons, we refer to the broader concept as
orchestration, rather than automation – humans are in the decision loop, with machines
rapidly executing the approved tasks.

• Where can I get more info on all of this?

Here are some good references for further reading:

IBM Cognitive Security White paper

A Beginner's Guide to Artificial Intelligence, Machine Learning and Cognitive Computing

5 Things You Need to Know About AI Buzzwords: Cognitive, Neural and Deep, Oh My!

Conclusion
The concept of artificial intelligence has been around since the 1950s. Its intent has
always been to mimic how the human brain identifies and interprets complex patterns
with the hopes of simulating neural networks that could someday begin thinking for
themselves. Since the ‘50s we have progressed by leaps and bounds, and while we are
nowhere near the classic science fiction image of a machine that thinks like a human,
we are now entering the era of cognitive computing, where the interaction between
computers and humans does take place.

Cognitive diverges from core machine learning and machine intelligence by having the
ability to understand, reason, learn and interact with humans using natural language.
Cognitive systems represent the first time a computer can provide insight into an
unstructured data set that we, as humans, might not have been able to identify.

IBM Watson for Cyber Security is an AI system made up of many different subfields of
AI, including natural language recognition and processing, predictive analytics, data
mining, machine learning, deep learning and knowledge graphs by which to display the
relationships within the data sets. This architecture works together to create a massive
corpus of knowledge containing billions of data elements that were previously
inaccessible to both humans and machines at scale. From every experience, the corpus
gets larger and smarter, and delivers that intelligence to our finger tips.

QRadar Advisor with Watson has only begun to scratch the surface of the Watson
corpus’s capabilities, by creating a bridge between the world of unstructured data
(Watson), deep analytics and machine learning on structured data (QRadar), and the
human security analyst to accelerate the breadth, visibility and speed of an
Security and Artificial Intelligence: FAQ
investigation. We don’t know exactly what the future holds, yet it’s undeniable that the
AI era is here and it promises to be a game changer for cyber security.

If you are interested in learning more about the IBM perspective on artificial
intelligence, or the IBM offerings that utilize the technology, contact any one of the
authors.

Jeff Crume, Distinguished Engineer, IT Security Architect, IBM Master Inventor, IBM
Security Solutions Tech Sales
[email protected]

Doug Lhotka, Executive CyberSecurity Architect, CISSP-ISSAP

[email protected]

Carma Austin, Executive Security Advisor

[email protected]