Demystifying
Artificial Intelligence not a new term …
But with many new applications!
Buzzwords - often used interchangeably
Not quite the same thing
Both terms connected with Big Data & Analytics
• AI is the broader concept of machines being able to carry out tasks
in a way that we would consider “smart”, and
• ML is an application of AI based around the idea that machines
with access to data can “learn” for themselves
Artificial Intelligence - 1950’s
In 1952, Alexander Douglas at University of Cambridge developed one of the first known
video games - playing tic-tac-toe with a computer (guess who won!)
Artificial Intelligence - 1990’s
In 1996, after four decades of on-going battle, Deep Blue became the first machine to win a
chess game against reigning world champion Garry Kasparov
Artificial Intelligence - 2018
“Artificial Intelligence is like summoning the demon” - Elon Musk
ML is a branch of AI, concerned with the design and development of algorithms
that allow computers to evolve behaviors based on empirical data, i.e. improve (or
optimize) performance with experience (or more data)
https://blogs.nvidia.com/blog/2016/07/29/whats-difference-artificial-intelligence-machine-learning-deep-learning-ai/
Training examples of a person
Test images
http://www.uk.research.att.com/facedatabase.html
Which of these things is NOT like the others?
Which of these things is like the others? And how?
Why “Learn”?
• There is no need to “learn” to calculate payroll
• Learning is useful when:
• Human expertise does not exist (navigating on Mars, choosing an action)
• Humans are unable to explain their expertise (speech/image recognition)
• Solution/action changes with time (routing on a network, conditional access)
• Solution needs to be adapted to particular cases (using biometrics)
• ML is the increasingly preferred approach for:
• Speech recognition & Natural Language Processing
• Image analysis
• Medical outcomes analysis
• Robotics controls
• Computational biology
• Security analytics
• Accelerating trend on:
• Improved machine learning algorithms
• Improved data capture, networking, faster computers
• New sensors / IO devices
• Demand for self-customization to user, environment
• Supervised Learning
• Unsupervised Learning
• Reinforcement Learning
• Concept Learning
• Clustering Algorithms
• Connectionist Algorithms
• Genetic Algorithms
• Explanation-based Learning
• Transformation-based Learning
• Case-based Learning
• Macro Learning
• Evaluation Functions
• Cognitive Learning Architectures
• Constructive Induction
• Discovery Systems
• Knowledge capture
• Data Mining
• Probability & Statistics
• Information theory
• Numerical optimization
• Computational complexity theory
• Control theory (adaptive)
• Psychology (developmental, cognitive)
• Neurobiology
• Linguistics
• Philosophy
AI & ML Applications
Autonomous Vehicles Home Automation
and many more…
• Recognize faces
• Transcribe speech
• Translate between hundreds of languages
• Spot subtle financial fraud
• Find relevant web pages for ambiguous queries
• Map the best driving route
• Automated cancer diagnoses
• Housecleaning robots
• Automated scientific discovery
Over the next five to 10 years, Facebook A.I. will
“get better than human level at all of the
primary human senses: vision, hearing,
language, general cognition.” - Mark Zuckerberg
Personal Assistants
Cyberspace landscape is rapidly
changing
Virtually anything can be attacked
Security skills are in short supply
99 days Median # of days between infiltration and detection
88% Of companies concerned
about cyberattacks in 2017
$4M Average cost of a data
breach in 2017
$8 trillion Cost (USD) of cybercrime
to global economy by 2022
750+% Growth in # of ransomware
families in 2016
Cybersecurity: in the News, in the Boardroom
Criminals are and
Cybercrime & Cybersecurity challenges
Too much to handle
False positives & Prioritization
Lack of in-house expertise
Opportunities from the Challenge
Towards Insight and Intelligence
6.5T threat signals
analyzed daily
Microsoft Trust Center
Bucketing Functions • Extract key dimensions
• Dependent on richness of the data
Promotion Signals • Signal of insight of “new issue”
• Selector for higher analysis
• Key to scalable system
Supervised Automatic Learning • Seek signals for self-correction
• Resiliency and Robustness
• HITL Human Over The Loop
Signals Labels
Higher dimensional data
Lower dimensional data
Labeling • Analyze and decide verdict or labels
• E.g. mapping, analysis, classifier, …
Identity at the center of security leveraging AI & ML
Corporate Network
Geo-location
MacOS
Android
iOS
Windows
Windows Defender ATP
Client apps
Browser apps
Google ID
MSA
Azure AD
ADFS
Employee & Partner Users and Roles
Trusted & Compliant Devices
Location
Client apps & Auth Method
Conditions
Microsoft Cloud App Security
Force password reset
Require MFA
Allow/block access
Terms of Use
******
Limited access
Controls
Machine learning
Policies
Real time Evaluation Engine
Session Risk
3
40TB
Effective policy
Security Operations leveraging AI & ML
Digital Cybercrimes Unit Leading the fight against cybercrime
Protecting people, organizations and our cloud through
global disruptions and enforcement actions against
cybercriminals leveraging AI & ML
Investigations, forensics and analytics
Machine learning, AI and data visualization
Public and private partnerships
Creative legal strategies
PhotoDNA 99% of Cyber Tip reports to the National Center for Missing and Exploited Children, are generated by PhotoDNA (8.2M+ annually) PhotoDNA has enabled increasing law enforcement engagement, and introduction of private sector participation (130+ private organizations) PhotoDNA is built into international law enforcement forensic tools Leveraging membership organizations like Internet Watch Foundation and Protect.org, to drive increased licensing/usage of PhotoDNA Innovation key to progress: PhotoDNA Cloud Service and PhotoDNA for Video
Online Child Exploitation
Every day, billions of unique images are uploaded and
shared via the Internet. Finding known child sex abuse
images is like trying to find a needle in a haystack.
AI & ML is critical to making sense of the information.
Perpetuation of child exploitation / revictimization
Law enforcement challenges from the scale of problem
At Microsoft, we do not take your trust for granted
• We live by standards and practices designed to earn your confidence and trust
• We collaborate with industry and regulators to build trust in the cloud ecosystem
• We leverage the scale and scope of the cloud, and cutting-edge AI & ML
• We are leading the deliberations on ethics in AI
“Businesses and users are going to embrace technology only if they can trust it.”
—Satya Nadella
• Increase in the number of
products and services
implementing AI effectively in
multitude of domains has led to
increased interest in
standardization of AI
• Countries and regulators have
begun taking interest in many
aspects of AI
• Standards are needed to enable
wider acceptance of approaches
and concepts, and to address the
interests and concerns of
governments and society
• Standards provide requirements, specifications, and guidelines that can be used
consistently to ensure that AI technologies meet objectives for functionality and
interoperability, and that they perform reliably, safely & ethically
• Techniques and algorithms in AI & ML are evolving rapidly
• More and more applications of them are being adopted
• They will continue to play an important role in all spheres of life
• Their usage in everything including Security will grow exponentially