Page 22 - MSDN Magazine, October 2017

ArtificiAlly intelligent FRANK LA VIGNE Exploring Azure Machine Learning Studio
Longtime readers of my blog (franksworld.com) have noticed over the past 18 months a marked shift in content toward data science, artificial intelligence (AI) and machine learning (ML). So, it’s only fitting that this column also shifts gears and focuses on the revolution happening all around us: The AI Revolution. Not too long ago, AI was the stuff of science fiction. Now we can add intelligence to virtually any app or Web site. In fact, many of your favorite apps and Web sites already employ some form of AI. Cortana and many other voice assistants are obvious examples of AI in the UI layer. Less obvious, but no less important, are intelligent algorithms optimizing resources, rec- ommendation systems telling you what movies you might like and determining what you see in your social media feeds.
Over the last 10 years, the focus of many developer and IT organizations was the capture and storage of Big Data. During that time, the notion of what a “large” database size was grew in orders of magnitude from terabytes to petabytes. Now, in 2017, the rush is on to find insights, trends and predictions of the future based on the information buried in these large data stores. Combined with recent advancements in AI research, cloud-based analytics
Figure 1 Generally Accepted Artificial Intelligence Phrases
tools and ML algorithms, these large data stores can not only be mined, but monetized.
With the cloud providing affordable computing power and stor- age, even small businesses can predict the future by anticipating customer behavior and identifying trends at the individual level and at scale. Organizations that can discover and deploy actionable predictive models before the competition does will dominate their market segment. Properly leveraged, AI can add serious value to any business. As Peter Drucker put it, “The best way to predict the future is to create it.” In that spirit, here’s a deeper look at AI and ML.
Getting the Terms Right
Before getting into an AI project, it’s important to define the scope of what exactly is “artificial intelligence.” This will be important as future columns will rely on a common set of meaning for terms as- sociated with this field. A quick Internet search of the term “artificial intelligence” yields a lot of various results, from chatbots and com- puter vision systems to debates on the nature of consciousness itself. While there’s no firm consensus as to what the term means, most ex- perts agree generally on basic phrases, which are listed in Figure 1.
The Power of the Cloud in the Palm of Your Hand
Around the middle of the last decade, I was an early adopter and proponent of the Tablet PC platform. As I would deliver presentations to various user groups and speak at conferences, one criticism would inevitably come up: lack of high-performance hardware. The reason for the lack of serious computing power on these devices had more to do with the constraints of making a tablet device viable: namely weight, battery life and cost. Many would object that while they admired the tablet PC form factor, they needed a device with a more powerful CPU. Fast forward 10 years and the limitations of cost, battery life and network connectivity have largely gone away. Any device within range of Wi-Fi and 4G networks can now connect to limitless computing services and storage resources in the cloud.
AI in the Cloud
As a developer, you have choices in terms of what types of intelligent services to consume. If an app or Web site requires image recognition or natural language processing, then Microsoft has made several ser- vices available as part of the Microsoft Cognitive Services, a set of APIs, SDKs and services that expands on Microsoft’s evolving portfolio of ML APIs. They enable you to easily add intelligent features—such as
Term
Description
Data Science
Interdisciplinary field that applies scientific methods and processes in order to extract insights from data.
Artificial Intelligence
Computer systems able to perform tasks that have traditionally required human intelligence, such as computer vision, speech recognition and more.
Machine Learning
A type of artificial intelligence that allows software to predict outcomes or classify data without explicitly being programmed.
Binary Classification
The process of classifying data into one of two groups. For example, determining if a flight will be delayed or on time.
Multi-Class Classification
The process of classifying data into one of three or more groups. For example, determining if a flight will be delayed, canceled or on schedule.
Regression
The process of determining the output of one value based on a number of other values. For example, determining how much a flight will be delayed based on weather, day of week, carrier and so on.
R
An open source programming language used in statistical processing.
Python
An interpreted multi-paradigm language often used in the field of data science.
18 msdn magazine