Artificial Intelligence for Managers

Leverage the Power of AI to Transform Organizations & Reshape Your Career: Leverage the Power of AI to Transform ... & Reshape Your Career (English Edition)

Malay a Upadhyay

• Publisher: BPB Publications
• Year: 2020
• ISBN:
  9789389898385
• Topic: Computer Science
• Subtopic: Artificial Intelligence (AI) & Semantics
• Index: Computer Science

Understand how to adopt and implement AI in your organization Key Features

• 7 Principles of an AI Journey
• The TUSCANE Approach to Become Data Ready
• The FAB-4 Model to Choose the Right AI Solution
• Major AI Techniques & their Applications:
• - CART & Ensemble Learning
  - Clustering, Association Rules & Search
  - Reinforcement Learning
  - Natural Language Processing
  - Image Recognition Description
  Most AI initiatives in organizations fail today not because of a lack of good AI solutions, but because of a lack of understanding of AI among its end users, decision makers and investors. Today, organizations need managers who can leverage AI to solve business problems and provide a competitive advantage. This book is designed to enable you to fill that need, and create an edge for your career. The chapters offer unique managerial frameworks to guide an organization's AI journey. The first section looks at what AI is; and how you can prepare for it, decide when to use it, and avoid pitfalls on the way. The second section dives into the different AI techniques and shows you where to apply them in business. The final section then prepares you from a strategic AI leadership perspective to lead the future of organizations. By the end of the book, you will be ready to offer any organization the capability to use AI successfully and responsibly - a need that is fast becoming a necessity. What will you learn
  
• Understand the major AI techniques & how they are used in business.
• Determine which AI technique(s) can solve your business problem.
• Decide whether to build or buy an AI solution.
• Estimate the financial value of an AI solution or company.
• Frame a robust policy to guide the responsible use of AI.
• 
  Who this book is for
  This book is for Executives, Managers and Students on both Business and Technical teams who would like to use Artificial Intelligence effectively to solve business problems or get an edge in their careers. Table of Contents
  1. Preface
  2. Acknowledgement
  3. About the Author
  4. Section 1: Beginning an AI Journey
  a. AI Fundamentals
  b. 7 Principles of an AI Journey
  c. Getting Ready to Use AI
  5. Section 2: Choosing the Right AI Techniques
  a. Inside the AI Laboratory
  b. How AI Predicts Values & Categories
  c. How AI Understands and Predicts Behaviors & Scenarios
  d. How AI Communicates & Learns from Mistakes
  e. How AI Starts to Think Like Humans
  6. Section 3: Using AI Successfully & Responsibly
  a. AI Adoption & Valuation
  b. AI Strategy, Policy & Risk Management
  7. Epilogue About the Authors
  Malay A. Upadhyay is a Customer Journey executive, certified in Machine Learning. Over the course of his role heading the function at a N. American AI SaaS firm in Toronto, Malay trained 150+ N. American managers on the basics of AI and its successful adoption, held executive thought leadership sessions for CEOs and CHROs on AI strategy & IT modernization roadmaps, and worked as the primary liaison to realize AI value on unique customer datasets. It was here that he learnt the growing need for greater knowledge and awareness of how to use AI both responsibly and successfully.Malay was also one of 25 individuals chosen globally to envision the industrial future for the Marzotto Group, Italy, on its 175th anniversary. He holds an MBA, M.Sc. and B.E., with experiences across India, UAE, Italy and Canada.A Duke of Edinburgh awardee, Malay has been driving the subject of responsible AI management as an advisor, author, online instructor and member of the European AI Alliance that informed the HLEG on the European Commission's AI policy. At other times, he remains a Fly that loves to travel and blog with Mrs. Fly. Blog links: www.TheUpadhyays.com
  

Section - II

Choosing the Right AI Techniques

Welcome to Section 2 of the AI journey. In this section, we will enter the AI laboratory to look at how and where AI models are created. We will also look at the different AI techniques without becoming too technical. This section is important because by knowing them, you will be able to:

• Tell whether a software uses AI, or what kind of AI it uses
• Differentiate powerful AI solutions from weaker ones
• Determine what type of AI techniques can solve your specific business or organizational problem
• Understand some of the most popular Machine Learning and Deep Learning techniques that are being used, such as: how it analyses different types of information, how it makes predictions, how it recognizes images like your face, how it communicates with your customers, or how a robot learns to behave like humans!

The chapters included are:

• Chapter 4: Inside the AI Laboratory
• Chapter 5: How AI Predicts Values and Categories
• Chapter 6: How AI Learns & Predicts Behaviors and Scenarios
• Chapter 7: How AI Communicates and Learns from Mistakes
• Chapter 8: How AI Starts to Think Like Humans

Learning these is important to understand clearly and know what AI can or cannot do, and to appreciate how AI is built to perform tasks.

CHAPTER 4

Inside the AI Laboratory

One of the most interesting things about artificial intelligence is that most of the basic tasks are preprogrammed in standardized codes that we can use and reuse. That does not mean that we do not need to customize the AI model, but there is a reason China introduced AI textbooks for preschoolers! (1). In this chapter, we will see what that secret place is where AI is created, and the blueprint that most algorithms follow.

Structure

In this chapter, we will discuss the following topics:

• Labeled and unlabeled data
• Supervised, unsupervised and semi-supervised learning
• AI modeling platforms
• Components of AI models
• Steps to creating AI models
• Training and testing AI models
• Overview of major AI techniques

Objective

After studying this chapter, you should be able to:

• Understand how much control you have over a given AI model's results and how much supervision it needs
• Appreciate the essential steps and elements that are involved in AI modeling by your engineers
• Understand how AI results are tested and performance is improved
• Get a sense of the major AI techniques and the problems they help solve

4.1 Data and models

Most code snippets used to build AI are standardized and only require tweaking to suit the specific problems that they are trying to solve. Almost 70 percent of the time is usually spent preparing and pre-processing the data to be analyzed by the model rather than in writing the code itself (2). The data can be labeled so that it identifies what each piece of information in the dataset represents. For example, in the following figure, the label TEMPERATURE at the top of the first column indicates what the numbers in this column are:

Figure 4.1: Labeled data

As you can see in the headings in the next table, it can also be unlabeled:

Figure 4.2: Unlabeled data

Likewise, machine learning models can be built to be supervised, unsupervised, semi-supervised, or reinforced. While supervised learning can give us predictions based on patterns, unsupervised learning can give us segmentations.

If we knew exactly the types of drinks that Maya has on any particular evening—water or tea—the ingredients that are used, and the recipe to prepare them, we would use supervised learning, where we would feed in labeled data and give specific instructions to the robot's AI model to decide which drink to serve on which day, and how to prepare it.

However, if we had no clue of the types of drinks Maya may choose to have, or what conditions influenced her choice or drank on any particular day, we may use unsupervised learning to let the model figure out patterns in Maya's preference and take decisions on its own. The data that would be used in this case can be unlabeled.

Usually, we use semi-supervised learning, where we know what Maya will have in terms of either water or tea, but we let the model decide the specific nuances of when she might prefer one over the other. The data is labeled to a small extent, with the rest remaining unlabeled. AI can give us richer insights with unsupervised methods, but we would have to trust it blindly. That might be okay in certain cases, such as recognizing the image of a person, where we don't necessarily need an explanation on how the AI arrived at its conclusion. However, a manager would need to explain why they made a certain business decision and would, therefore, need context to ensure that they are making informed choices rather than blindly following AI. it would require AI to be supervised, which may limit the depth of analysis. A semi-supervised approach balances the two extremes, which can be useful in certain use cases.

Finally, reinforcement learning is extremely important as it implies teaching the robot to learn through constant feedback or reward for the right step. That, of course, brings its own set of risks and ethical dilemmas. We will see this in detail in its dedicated chapter.

4.2 AI Modeling

Where are AI models created? Coders use particular languages to write codes on a platform of choice, the way you'd write in English or French on a blackboard or a paper. Two of the popular coding languages are Python and R. These languages are written on a platform like Anaconda. Python is better for more advanced analysis like deep learning, while R is better for neat visualizations and simpler machine learning techniques. Both have amazing libraries and packages to do things efficiently.

What do we mean by 'library'? A data scientist friend of mine gave me the perfect analogy to explain this term in the context of AI. Think about the Matrix series of movies. Do you remember when Neo was learning Kung-Fu or when Trinity had to learn to fly a helicopter? The operator just uploaded libraries 'Kung-Fu' and 'helicopter' to the Matrix environment, respectively.

The library is a term used to describe tools that are pre-equipped to do a job. All we have to do is provide some inputs to get an output. For example, Numpy is a library that knows all mathematical calculations. Matplotlib knows how to plot numbers and graphs. And Pandas is used to import and manage data sets. Remember, AI codes are not born with the knowledge of even basic things like mathematics. Teaching them is like teaching a kid everything from scratch. Libraries help automate a lot of that. So, to perform mathematical calculations, all we have to do is import the Numpy library. Our robot then does not need to learn how to count the hours to determine how long Maya has been at work.

How are AI models created? There are three parts to it:

1. Pre-processing the data that AI will analyze
2. Writing the actual code
3. Training and testing the model for efficiency

To recap from previous chapters, any dataset has to be prepared and pre-processed to make it ready and usable for AI codes. After all, data can be incomplete. For example, the data we feed Maya's robot may have information missing on what Maya drank on some of the days last year. Data can also be of different types. For example, Maya's age over the years would be a numerical value, but her mood may be categorized as happy, neutral, or sad. The weather, on the other hand, would be numerical, too, but in a different unit. Data can also be biased. If, say, Maya was working last year but is now pursuing her doctoral studies and living a student lifestyle, her drinking patterns may change. In such a case, information about her drinking last year would be biased to a working lifestyle and not applicable to her current situation. Data scientists and engineers take all this information to clean up the data to ensure that the algorithm provides as accurate a result as possible. Inaccuracy is one of the main reasons for AI's slow adoption.

Once the data is pre-processed, we write codes depending on which AI technique or techniques we are planning to use and then modify it to suit our specific business problems.

Finally, to determine whether the AI model works, most engineers follow the 80-20 rule, particularly in the case of supervised learning. They first train the machine on 80% of historical data and then test it on the remaining 20%. For example, let's say we have the data on which beverage Maya drank over the last 50 workweeks. Her robot would look at all the inputs (that is, the time she came home, the weather outside, the day, and so on) and the corresponding outputs (that is, what she drank) in 40 of the 50 workweeks. That's 80%. It would then study these to analyze what's going on and find patterns. It would look at only the inputs in the remaining 20% of the cases, that is, the remaining ten weeks—and try to predict what Maya drank in those ten weeks. Those guesses will then be compared with what Maya drank in those ten weeks to see how accurate the robot was:

Figure 4.3: Testing the initial model

If the accuracy is low, the AI model (both the code and the input data) will be adjusted further until the accuracy improves:

Figure 4.4: Testing the retrained model

It is why being able to see how a machine learning software determines an output is important for any AI user so that we can check the biases or inherent errors that may be in play. If we can't understand why a machine decides what it decides, how would we know whether to trust its decision? After all, the robot can learn to warm a glass of cold water by simply leaving it standing at room temperature, which will take quite a while. Instead, we can teach it to heat the water on the stove to bring it to room temperature even quicker. The effectiveness of AI naturally increases with how well trained it is, how good the data is, and how rigorous the codes are.

Having acclimatized to the AI laboratory, let us now look at all the different techniques the robot will have to apply to serve Maya her drink successfully.

4.3 Techniques overview

There are many AI solutions out there today, but how many genuinely work? How many are even genuinely AI? As we saw in Section 1, and as we will see in this section, artificial intelligence can be very simple automation of mathematical or statistical computation. It can also be something very complex and truly intelligent. A 2019 report based on a survey by London venture capital firm MMC claimed that 40% of European start-ups classified as AI companies didn't use AI in a way that was material to their businesses. (3). The AI label helps companies capitalize on the AI buzz, which often leads to the overuse of this term by their communications department or PR.

By the end of this section, you will have understood how to sense strong AI solutions from weaker ones. In this section, we will take Maya's robot as our guinea pig once again, and learn all the things it could ideally do to serve Maya the right beverage. There are many more AI techniques out there, but we will cover some of the most prominent ones in the next few chapters. These will include:

• How AI predicts numbers or categories: We will look at classification and regression techniques, as well as more advanced decision trees and ensemble learning methods to see how the robot can predict whether Maya would like water or tea.
• How AI understands and predicts behaviors and scenarios: Clustering will show us how the robot figures out the different types of beverage there are to choose from in the first place. Association rule learning will then allow it to figure out connections between clusters and Maya's behavior to see if she would like to have something else with her drink. Search algorithms will then explain how the robot can choose the best option with the lowest cost. We will also share a brief word on Monte Carlo Simulation, which is normally used for Risk Analysis.
• How AI communicates and improvises: Reinforcement Learning will show us how the robot can learn from its mistakes on the go. That includes the robot learning to walk up to Maya. With Natural Lang...

Table of contents

• Cover Page
• Title Page
• Copyright Page
• Dedication Page
• About the Author
• About the Reviewer
• What Some Students Have To Say
• Acknowledgement
• Preface
• Errata
• Table of Contents
• Section I: Beginning An AI Journey
• Section II: Choosing the Right AI Techniques
• Section III: Using AI Successfully and Responsibly
• Epilogue
• References