The field of artificial intelligence (AI) has existed since the 1950s when visionaries imagined being able to turn computers into thinking machines that could mimic and even surpass the ability of humans to learn. Why did it take almost six decades for the dream of artificial intelligence and machine learning (ML) to emerge? There is a convergence of technologies that have come together to lead the market towards the reality of machine learning, artificial intelligence, and, finally, to augmented intelligence. In this chapter, we discuss the technology infrastructure that has emerged to support augmented intelligence.

What are the fundamental changes that have allowed artificial intelligence and machine learning to experience a rebirth? It begins with having the right infrastructure in place to manage data. How can an organization manage high volumes of complex data in order to build and maintain machine learning models? To be successful, it is mandatory to manage all of the required data in a systematic manner so that data being leveraged is well understood and accurate. Therefore, the journey to augmented intelligence begins with managing data. Simply put, you can’t attempt to do true augmented intelligence until you have a lot of the right data. Big data in this context has to be approached in a methodical manner. Why is now the time right for the dramatic transition to the combination of machine learning and human intelligence? We are in the midst of an era of convergence in which advances in technology have accelerated our ability to manage and understand data. Later in this chapter, we provide an overview of the type of machine learning techniques that are being applied to this well-managed data. But first, we need to understand the underlying hardware systems and platforms needed to put the data to use for augmented intelligence.

Many advanced thinkers in the mid part of the 20th century anticipated that computing would become a utility. They imagined that like electricity, computing could become a simple utility that a business would plug into in order to unleash unlimited power at a fraction of the cost. The idea of Internet-driven services were common in academic circles through the use of the Advanced Research Projects Agency Network (ARPANET)—a government-funded project to help the military and academic researchers exchange information over a network.

True commercial cloud services emerged in 2002 when Amazon introduced Amazon Web Services (AWS), followed by the launch of Elastic Compute Cloud (EC2) four years later. Ironically, Amazon began selling its web services when it implemented more cloud services than it needed to operate its online commerce platform. It became clear that the ability to add compute and storage capability on demand was a game changer. The architectural model of the cloud services combined with a new pricing model was instrumental in the transition to commercial machine learning and artificial intelligence. It wasn’t just that the price per CPU hour of computing was low and the price to store data was a fraction of previous financial models. The key issue was that the underlying architecture was designed to support unanticipated scaling and management of data.

A multitude of vendors understanding the potential revolution of the cloud began to build hardware enablement—newer powerful chips and graphic processing units (GPUs) turbocharged the performance of machine learning models that needed to use massive amounts of data to solve complex analytics for machine learning models. Techniques such as GPUs enabled near real-time processing of images, videos, and complex data.

The question then became, how much could you accomplish if there are virtually no limits on computing power or storage? The answer was simple: Without physical constraints, anything was possible. This transformation to cloud computing has opened the door to innovation that is changing both the maturation of data science and the ability of businesses to accelerate their growth. Combining the power and economics of cloud computing with the innovations in machine learning and artificial intelligence has created a revolution that will touch every industry, every business and business process, and every human activity. All human endeavors and processes will be impacted by this renaissance. The real question is, how can we harness this innovation to support the power of the human brain to make better-informed decisions that can change everything?

As we discussed in Chapter 1, artificial intelligence is the umbrella concept that incorporates methods, tools, languages, and algorithms. With cloud infrastructure, it is now possible to provide sophisticated tools that can transform what had only been possible through laborious manual efforts. The efficiency and pricing structure of the cloud has transformed the way data can be used to move to a new generation of AI. With this platform in place, it is not surprising that machine learning and artificial intelligence are experiencing a renaissance. So, with this power in place, how do we use cloud services to transform our ability to harness the power of data.

In the past, data analysts were locked into analyzing data at a superficial level: How many widgets did I sell last month? How much did they cost to make and how much profit did my business unit make? As businesses grew more complex with more lines of business, more business partners, and more innovation, it become more difficult to simply ask direct questions and gain the level of insight needed to be competitive. Business leaders wanted to understand the impact on the future. Would it be possible to use data to anticipate the future and determine new directions and new business models never seen before? Leaders wanted to understand hidden patterns in their vast stores of data and information about their products, services, and customers. They also wanted to begin to put their data to work in order to plan for the future. What will customers expect to purchase next year? In other words, how can a business anticipate the future?

As the industry changed with the advent of the cloud and new hardware architectures that were more agile and efficient, the ability to conduct advanced analytics grew at a rapid pace. With the technological advances, it was conceivable to rapidly collect and automatically process the data. Now, it was possible to provide more advanced analytics techniques such as data mining, data lakes, and predictive analytics, and to leverage machine learning algorithms. The promise of advanced analytics has been to use massive amounts of data to accelerate our ability to make sense of a chaotic world. The ultimate goal, as we have explained in Chapter 1, is to move from simply getting answers to questions to delve deeply into the context and meaning of complex data techniques based on codifying knowledge—augmented intelligence. But before we satisfy that goal, we have to understand the changing nature of data management technologies and architectures. Therefore, you have to approach big data in a methodical manner. Big data begins with understanding the data sources—what they mean and how they are related to each other. Once you establish the data sources, you need to understand the relationships between data elements and sources. The data needs to be clean and well understood in context. At the end, it is critical to find the context, anomalies, and patterns from all sorts of data to be successful.

Big data techniques are designed to manage a huge volume of disparate data at the right speed and within the right time frame. The goal is to enable near real-time analysis and action. Four characters typically define big data: Volume (how much data), Validity (correctness and accuracy based on intended use), Velocity (how fast that data is ingested and processed), and Veracity (the truthfulness of the data to solve a problem). The key issues that are important to understand are how much data you have and how much you need. You need to understand how fast that data needs to be processed based on the problem you need to solve. Finally, you need to consider the various types of data that will ensure that you are making well-informed decisions. In this section, we will delve into the different types of data that we need to work with in order to be prepared for augmented intelligence.

To gain the type of insights into data that we need to be able make more informed decisions requires enough data so that results are trustworthy. Data of all forms need to be mapped and understood in context. Data is only getting more complicated, with more nuances than ever. How do you map highly structured data with highly unstructured data such as written material, video and images? How do you ensure that there aren’t biases within your data sets? How can you pre-train data so that it is easier to get predictable results? All of these questions are fundamental to being able to augment our ability to use advanced analytics to make better decisions. Before we discuss the details of how companies are making the transition from big data to AI and augmented intelligence, we need to start with the basics. We divide the foundations of big data into three types: structured, unstructured, and semi-structured data.

Structured data refers to information that has a defined length and format—for example, numbers, dates, and names. A majority of traditional business applications are designed to manage highly structured data. The traditional relational database was structured so that each element was defined, including the relationships within the data sets. This structure made it possible to create understandable answers to questions inside this data.

The initial use cases of structured data were used to coming up with answers to relatively straightforward questions. However, questions have become increasingly more complicated as devices generate more and more data (such as sensor data), and companies create data resulting from human interactions with computers. Data generated by data logs, point-of-sale data, and clickstream data are often referred to as semi-structured data. Machine learning models are well suited to the structure and tagging of structured data.

For most of the history of the computer industry, the focus has been on structured data. However, this has begun to change dramatically over the past decade as we have discovered new ways of gaining insights from data without traditional structure. Over time, we have moved from the ability to analyze basic text to now being able to leverage unstructured data in new ways. Increasingly, we are able to add structure and nuance to unstructured data in order to gain an understanding of meaning buried in complex documents, as well as text, videos, images, atmospheric data, voice, and social media. Clearly, being able to understand the content and context of unstructured data is imperative for machine learning and, specifically, augmented intelligence.

Unstructured data does not have a predefined data model nor is it organized in a predefined manner. Therefore, it is not surprising that most of the information in the world is unstructured. Unstructured information is typically text-heavy, but it may contain data such as dates, numbers, and facts as well. This results in irregularities and ambiguities that make it difficult to understand using traditional programs as compared to data stored in fielded form in databases or annotated (semantically tagged) in documents.

Despite the abundance of unstructured data, machine learning algorithms don’t work well without structure. Of course, there is inherent structure even in unstructured data, but the difference is that humans have to do the hard work to understand the hidden structure of the data. Therefore, for unstructured data to work well in machine learning, you have to be able to understand the features and attributes of the data.

Turning unstructured data into meaningful information can be much more complex than dealing with the highly structured databases. In reality, there is no such thing as data without any structure. The imperative for finding meaning in unstructured data is to understand the hidden patterns within that data. This is the technique that is the foundation of Natural Language Processing, which derives an understanding of what a collection of words means based on understanding usage patterns.

You need to understand the fundamentals of machine learning techniques in order to understand how far these algorithms and models will take you. We will discuss building models in depth in Chapter 4.

The availability of massive amounts of data that can be stored and processed at incredible speeds is imperative to success with machine learning. However, there are constraints that make it difficult to get to the results that help a business achieve the type of breakthroughs that can set it apart from the competition. First, what is the nature of the data? What is the source of that data? Is the data reliable and is it the right data to solve the problem? Is there enough expertise within the data science team? The smartest data scientists are of little value if they don’t understand the business problem they have been asked to solve. These scientists may do a wonderful job selecting a good algorithm and preparing data, but they may not have good intuition about critical business problems. To be successful, the data scientist must be able to collaborate with “subject matter experts,” that is, business experts.

Therefore, before even working with the data, it is important to understand the business process. Why focus first on business process? Simply put, anything that is achieved in business must be based on a process for solving a problem. For example, are you trying to create a new process to better maintain machinery on the factory floor based on the data that machines generate? What is the optimal process for ensuring safety and efficient production? In diagnosing a disease, what is the process needed to support a doctor’s ability to understand both individual patients and their history, and to apply the accumulated knowledge of successful treatment outcomes to their symptoms? We include the details of business process in Chapter 3.

Once you understand the processes you are dealing with, you now have to understand how to prepare and manage the data so that it is helpful in solving problems. Data has to be tagged to enable machine learning. For example, “chicken” can mean an animal or it could be the first word in “chicken pox”—very different meanings although the word is identical. Tagging is one of the most complex issues that has to be solved when organizations are using a massive amount of data to understand the best next step. When you are dealing with a small data set, it is possible to have a human tag data so that the machine and the algorithms understand it. For example, using the chicken pox example, the subject matter expert can tag data based on the knowledge of the disease. There are often commercial data sets available that are pre-tagged. They will understand what terms mean and tag them so that the data can be accurately processed and provide answers and conclusions. But what happens if you are dealing with a massive amount of data that is untagged? In this case, you need to be able to apply methods that can anticipate what the tags might be. Automating the tagging process is one of the emerging innovations that will support the maturation of artificial intelligence. This automatic tagging is essential to help a data scientist build algorithms to support decision making.

Without the innovations in the field of advanced analytics over the past decade, augmented intelligence would not be possible. In the second half of the last century, tremendous progress was made through the growth in computing hardware systems. The industry evolved from a handful of vendors producing extremely expensive, massive systems only accessible to a few to smaller computer systems that were affordable to even the smallest business unit. Despite the fact that incredible progress was being made, the performance and capability of these early systems paled in comparison to what we see today. Because of the limitations in computer memory, computing performance, networking speeds, and data movement, the tasks that could be performed by technology were limited.

Smart technologists found ingenious ways to squeeze as much performance as possible out of limited resources. Relational databases provided a pragmatic way to analyze data in an accessible way to the massive number of businesses that wanted to begin analyzing the data that they housed in their systems. Data is not straightforward—especially when you move out of the world of highly structured databases. The traditional relational database was developed in an era when compute power and data storage were extremely expensive. Therefore, databases had to be optimized to collect data efficiently and provide business users with a clear way of getting answers to questions.

Several different technology advances make augmented intelligence feasible today. First, there have been advances in machine learning techniques that make it possible to leverage the insights captured in data. These advances have been combined with scalable infrastructure in the cloud that makes it possible to store up to petabytes of data, coupled with powerful compute engines that can drive the advanced analytics to gain insights from the data. Second, the rise of big data means a rise in data volumes and diversity. The net effect is that there is more data available from both structured and unstructured sources, along with techniques to integrate and analyze these diverse sources.

Machine learning and AI are emerging as the answer to many of the complex business problems being addressed by advanced analytics. There is a presumption that if you simply apply machine learning algorithms to a problem, you will gain insights that were never possible in the past. There is no debate that existing business leaders are facing new and unanticipated competitors. These businesses are looking at new strategies that can prepare them for the future. Although there are many different strategies that a business can try, they all come back to a fundamental truth—you have to follow the data. And you need to ensure that you are using the right data with the right techniques to achieve your results.

Machine learning has become one of the most important topics within development organizations that are looking for innovative ways to leverage data assets to help the business gain a new level of understanding. Why add machine learning into the mix? With the appropriate machine learning models, organizations have the ability to continually predict changes in the business so that they are best able to predict what’s next. As data is constantly added, the machine learning models ensure that the solution is constantly updated. The value is straightforward: If you use the most appropriate and constantly changing data sources in the context of machine learning, you have the opportunity to predict the future.

Machine learning is a form of artificial intelligence that enables a system to learn from data rather than through explicit programming of a set of rules. Machine learning algorithms themselves learn to divide each data observation into a category. Categories can be very varied. They can consist of a group of faces associated to a particular person’s name or the set of choices of what products to show a customer. However, machine learning is not a simple process.

Machine learning consists of a variety of types of algorithms, all of which iteratively learn from data to improve, describe data, and predict outcomes. As the algorithms ingest training data, they result in a model (a set of categories which form its predictions) for that data. It is then possible to use that model on new data that the machine learning algorithm has not seen before to make more predictions. A machine learning model is the output generated when you train your machine learning algorithm with data. After training, when you provide the model with new data input, its output will be provide new insights regarding the data.

You likely interact with machine learning applications without realizing it. For example, when you visit an e-commerce site and start viewing products and reading reviews, you will likely be presented with other, similar products that you might find interesting. These recommendations are not hard coded by an army of developers. The suggestions are served to the site via a machine learning model. The model ingests your browsing history along with other shoppers’ browsing and purchasing data in order to present other similar products that you might want to purchase.

Machine learning enables models to train on data sets before being deployed. Some machine learning models are online and continuously adapt as new data is ingested. On the other hand, other models—known as offline machine learning models—are derived from machine learning algorithms, but once deployed, do not change. This iterative process of online models leads to an improvement in the types of associations that are made between data elements. Due to their complexity and size, these patterns and associations could have easily been overlooked by human observation. Once a model has been trained, these models can be used in real time to learn from data. Offline models do not have this advantage and must be retrained periodically with new data.

In addition, complex algorithms can be automatically adjusted based on rapid changes in variables such as sensor data, time, weather data, and customer sentiment metrics. For example, inferences can be made from a machine learning model: If the weather changes quickly, a weather predicting model can predict a tornado, and a warning siren can be triggered. The improvements in accuracy are a result of the training process and automation that is part of machine learning. Online machine learning algorithms continuously refine the models by continually processing new data in near real time and training the system to adapt to changing patterns and associations in the data.

One of the innovations that has made a significant contribution to the acceleration of machine learning is the open source community. As a consequence, there are more resources, frameworks, and libraries that have made development easier. These open source communities result from 40 years of research by scientists to invent ML algorithms and create data sets using them. Although companies also invent new ML algorithms that they do not share, many new algorithms are shared because scientists recognize the value in sharing scientific discoveries. The scientific community in universities has invented and made publically available most of the vast collection of ML algorithms. The business community should thank them for their work as well as organizations such as the National Science Foundation, the National Institutes of Health, and the Department of Defense (DoD), which have sponsored research at universities in the United States to invent and understand machine learning.

The disciplines of statistics, data mining, and machine learning all have a role in understanding data and describing the characteristics of a data set as well as finding relationships and patterns in that data and building a model. There is a great deal of overlap in how the techniques and tools of these disciplines are applied to solving business problems.

Many of the widely used data mining and machine learning algorithms are rooted in classical statistical analysis. Data scientists combine technology backgrounds with expertise in statistics, data mining, and machine learning to use all disciplines collaboratively. Regardless of the combination of capabilities and technology used to predict outcomes, having an understanding of the business problem and business goals, as well as subject matter expertise, is essential. You can’t expect to get good results by focusing on the statistics alone, without considering the business side to pick critical business problems to solve.

The following highlights how these capabilities relate to each other. Machine learning algorithms are covered in the next section in greater detail due to the importance of this discipline to advanced analytics.

Statistics is the science of analyzing the data. Classical or conventional statistics is inferential in nature, meaning it is used to reach conclusions about the data (various parameters). Statistical modeling is focused primarily on making inferences and understanding the characteristics of the variables. Machine learning provide a way to computationally encode various statistical techniques. Data scientists take statistical ideas and encode them as algorithms and then apply those algorithms to data to make predictions.

To understand the role of machine learning, let’s start with some context. Artificial intelligence, machine learning, and deep learning are all terms that are frequently mentioned when discussing big data, analytics, and advanced technology. Artificial intelligence seeks to understand the mechanisms underlying thought and intelligent behavior. AI includes the subfields of natural language processing, vision, robotics, machine learning, and knowledge representation and reasoning. Machine learning is the sub-field that focuses on theories and algorithms to make it possible for a machine to learn a task or to make a prediction. A machine that can translate a paragraph of English into another language uses AI (both machine learning and natural language processing), and a thermostat that learns your preferences for keeping your home comfortable uses machine learning.

Depending on the nature of the business problem being addressed, there are different approaches based on the type and volume of the data. There are a number of approaches to machine learning that are relevant to the ability to create algorithms that support business problems. These approaches contain two main types: supervised learning and unsupervised learning. Two often-discussed types of unsupervised learning are reinforcement learning and deep learning.

It is clear that infrastructure—both in terms of cloud, data, and machine learning—comprises the building blocks for making artificial intelligence a reality. Cloud services provide the elasticity, scalability, and price/performance needed to manage the massive amount of data needed to gain insights into complex data models. However, without the ability to prepare data so that it is clean and accurate, it is not possible to gain significant insights into the data that is the foundation of augmented intelligence. When you combine cloud services with data management, you are now in a good position to put machine learning models to use to significantly improve our ability to gain insights and take action.