Operating AI

Bridging the Gap Between Technology and Business

Ulrika Jagare

• Publisher: Wiley
• Year: 2022
• ISBN:
  9781119833215
• Edition: 1
• Topic: Computer Science
• Subtopic: Artificial Intelligence (AI) & Semantics
• Index: Computer Science

A holistic and real-world approach to operationalizing artificial intelligence in your company

In Operating AI, Director of Technology and Architecture at Ericsson AB, Ulrika Jägare, delivers an eye-opening new discussion of how to introduce your organization to artificial intelligence by balancing data engineering, model development, and AI operations. You'll learn the importance of embracing an AI operational mindset to successfully operate AI and lead AI initiatives through the entire lifecycle, including keyareas such as; data mesh, data fabric, aspects ofsecurity, data privacy, data rights and IPR related to data and AI models.

In the book, you'll also discover:

• How to reduce the risk of entering bias in our artificial intelligence solutions and how to approach explainable AI (XAI)
• The importance of efficient and reproduceable data pipelines, including how to manage your company's data
• An operational perspective on the development of AI models using the MLOps (Machine Learning Operations) approach, including how to deploy, run and monitor models and ML pipelines in production using CI/CD/CT techniques, that generates value in the real world
• Key competences and toolsets in AI development, deployment and operations
• What to consider when operating different types of AI business models

With a strong emphasis on deployment and operations of trustworthy and reliable AI solutions that operate well in the real world—and not just the lab— Operating AI is a must-read for business leaders looking for ways to operationalize an AI business model that actually makes money, from the concept phase to running in a live production environment.

CHAPTER 1
Balancing the AI Investment

Making a strategic decision to invest in AI is not just like any other decision. It's not only the financial aspect of the investment you must consider but the transformational power of AI for your company that needs to be understood. AI has the potential to fundamentally transform the business you are doing, and for some businesses it's even a question of survival to embrace this technology as fast as possible. Few examples in the market today have shown that businesses can achieve the expected values by just “doing some AI experimentation on the side.” However, a surprisingly large number of companies either don't know what investing in AI means or truly believe that investing in AI means hiring a bunch of data scientists to build AI models. This attitude needs to change for more companies to gain the expected return on investment (ROI) from their investments.

The reality of today is that AI is reshaping entire industries, making it possible to achieve previously impossible levels of scale through operational efficiencies and continuous learning as well as innovation. The reason for this is that AI automates the extraction of insights from data, detecting patterns in a way that would take weeks, months, or even years for humans to do—if at all.

AI can be used to automate internal business processes and make them more efficient, as well as develop new and enhanced products and services in the commercial dimension. It can be used to predict what a customer is most likely to buy and to automatically detect manufacturing inefficiencies or fraudulent behavior. A retailer can use AI to predict the volume of traffic in a store on a given day and use that prediction to optimize its staffing. A bank can use AI to infer the market value of a home, based on its size, characteristics, and neighborhood, which, in turn, lowers the cost of appraisals and expedites mortgage processing. Autonomous vehicles are another interesting area of applied AI. There are not only AI capabilities built into an autonomous vehicle, but the vehicle also includes sensors that capture and encode data about the world. This could be seen as a “brain” that reasons and makes decisions. It seems that the more use cases and business segments AI is applied to, the more ideas arise in terms of where and how it can be used.

And we're just getting started with AI.

The fact is that modern data management and software capabilities have progressed far enough to allow any organization to capture and use its data to build, train, and validate even the most complex predictive AI models. Many companies have successfully embedded predictive models in their core business capabilities, which has empowered them to build game-changing products and services that would otherwise have been unachievable. And in doing so, they've proven that artificial intelligence is changing the business landscape forever.

However, although most of the business opportunity comes from adopting AI at scale, only a minor part of the enterprises tends to invest in AI across multiple business areas. One possible explanation for this could be that many business leaders are still exploring AI to better understand its benefits in their specific context. Just knowing that AI can solve problems that were previously unsolvable, and that AI can answer questions enterprises didn't even know to ask, isn't enough to go all in. On top of that, there are also misguided delusions that AI can solve anything, and when it becomes apparent that it can't, confusion arises about the true business value of AI. Hence, experience shows that achieving business success from AI requires experimental and incremental approaches to adoption, but it should also be acknowledged that introducing AI at scale is a transformational and challenging task for most large enterprises.

There is no silver bullet to succeed with your AI investment, but there are some fundamental aspects that should be driving your objectives and realization plans, and that includes a balanced approach to AI. In this chapter you will find out what that means and why it is important for your business. You'll learn why it's vital to approach your AI solution from an operational perspective from the get-go. The chapter will begin by defining AI and by sorting out what AI is in relation to other related concepts such as machine learning (ML), automation, and robotics, just to mention a few.

Understanding the AI life cycle is key and will be sorted out in relation to defining some of the operational fundamentals you need to address in order to succeed with your AI investment. I'll also clarify the importance of operating AI in the context of the AI life cycle.

Finally, this chapter will address why you need to put more effort on making your AI model operational than you put on developing your AI model. Understanding and accepting this is a first major step toward embracing an operational mind-set to AI, which is vital in order to succeed with your AI investment.

Defining AI and Related Concepts

Artificial intelligence (AI) refers to the ability in a computer program and in robots to emulate humans' and animals' natural intelligence. This refers primarily to cognitive functions such as the ability to learn from experience, to understand natural languages, and to solve problems, but also to tasks such as planning a sequence of activities and generalizing between situations. As more and more companies start to realize how AI can benefit their specific business, the uses of AI expand by the minute. Some examples of areas where AI is currently being applied are:

• Voice and face recognition
• Language translation
• Chat bots
• Digital assistants
• Image recognition
• Recommendation engines
• Self-driving cars

In relation to AI, it's worth mentioning the term data science. Data science can be defined as an interdisciplinary field that uses scientific methods, processes, algorithms, and systems to extract knowledge and insights from structured and unstructured data and that applies knowledge and actionable insights from data across a broad range of application domains.

Artificial intelligence and data science are unfortunately often used interchangeably in the industry, which sometimes causes confusion since there are some differences between the concepts. Whereas data science is a broader term than AI and should be seen as a comprehensive procedure, AI is instead a set of modeling techniques that a data scientist uses to develop models. It's also worth noting that contemporary AI used in the world today is artificial narrow intelligence. Under this form of intelligence, computer systems do not have full autonomy and consciousness like human beings; rather, they are only able to perform tasks that they are trained for. However, some prioritized objectives within AI research include machine reasoning, knowledge representation, machine planning and learning, natural language processing for communication, computer vision, and the ability to move and manipulate objects. Keep in mind, though, that artificial general intelligence (AGI) and artificial consciousness (or singularity, as it is also referred to) are still in a conceptual stage, and their real-world use is far from mature. The theory of technological singularity refers to a hypothetical point in time at which technological growth becomes uncontrollable and irreversible, resulting in unforeseeable changes to human civilization. However, most researchers in the industry can't agree on when AGI will be ready. Some estimate somewhere between 2040 and 2050 at the earliest.

Machine learning (ML) is the use of computer algorithms that improve automatically through experience based on patterns and deviations in data. ML is seen as a subset of AI. Machine learning algorithms build a mathematical model based on sample data known as training data to make predictions or decisions without being explicitly programmed to do so. Machine learning algorithms are used in a wide variety of applications where it's difficult or insufficient to develop conventional algorithms to perform the needed tasks.

These basic algorithms for teaching a machine to complete tasks and classify like a human date back several decades. But if ML isn't new, why is there so much interest today? Well, the fact is that complex ML algorithms—for example, using neural network techniques—need a lot of data and computing power to produce useful results. Today, we have more data than ever, and computing power is pervasive and cheap. The past few decades have seen massive scalability of data and information, allowing for much more accurate predictions than were ever possible in the long history of ML. Machine learning algorithms are therefore now better than ever and widely available in open source software. However, for more simple ML models the big-data revolution was more important than easy access to computational power.

Here are some common usage scenarios for ML:

• Predicting a potential value
• Estimating a probability
• Classifying an object
• Grouping similar objects together
• Detecting relations
• Finding outliers

There are many different types of ML algorithms, and each class works differently. In general, ML algorithms begin with an initial hypothetical model, determine how well this model fits a set of data, and improve the model iteratively. This training process continues until the algorithm learning is optimized or the user stops the process. Learning can be supervised, unsupervised, or semi-supervised (see Figure 1.1).

Figure 1.1: Overview of supervised, unsupervised, and semi-supervised learning

Supervised learning (SL) needs structured and labeled data to run and is best used for classification of data or for regression analysis, or both. Classification refers to the problem of identifying in which category (sub-population) an observation (or observations) belongs to. Regression analysis in the context of ML usually refers to building a prediction model.

Unsupervised learning (UL), in contrast, uses unlabeled data that the algorithm tries to make sense of by extracting features and patterns on its own. This technique is mostly used for clustering and anomaly detection. Clustering in this context refers to the task of grouping a set of objects in such a way that objects in the same group (called a cluster) are more similar to each other than to those in other groups (clusters). Anomaly detection is the identification of rare items, events, or observations that raise suspicion by differing signif...