Artificial Intelligence for Decision-Making: How Augmentation beats Automation

Introduction

Anything that could give rise to smarter-than-human intelligence—in the form of artificial intelligence [...] -- wins hands down beyond contest as doing the most to change the world. Nothing else is even in the same league.

~Elezier Yudkowsky (CNBC, 2012)

Artificial intelligence is evidently here to transform the world. This revolutionary power is represented by its tremendous value, as the artificial intelligence (AI) market is expected to grow to $120bn by 2025 (tractica, 2019). The fastest-growing segment of AI technologies is decision support tools that are estimated to account for 44% of the AI market by 2030 (Gartner, 2019). Given the value these decision-support systems can add, understanding them is crucial. However, two opposing approaches toward the deployment of AI for decision-making exist. The first, automation, seeks to replace human decision-makers with these cognitive technologies. The second, augmentation, seeks a collaborative partnership of humans and AI to enhance cognitive performance collectively. In practice, projects are mostly dominated by the goal to automate (Davenport & Ronanki, 2018). Researchers, on the other hand, urge to focus on augmentation, as the anticipated higher acceptance would make this AI superior (Duan, Edwards & Dwivedi, 2019). This evidence-based article aspires to bridge the gap between practice and academia by examining the role of AI for decision-making.

The article is structured into four main parts. The first provides a brief background on artificial intelligence while the second and third outlines and contrast the automation and augmentation view on the role of AI in decision-making. The fourth part showcases the implications for managers and concludes the article.

Background on Artificial Intelligence

While AI has become increasingly relevant in the age of Big Data, it has been around for a much longer time than people realize. The term AI was first coined in 1956 by John McCarthy. In the early 1980s, AI research was revived by the success of so-called “expert systems”, a form of AI that simulates the knowledge and analytical skills of human experts. These applications were almost exclusively rule-based inference systems and the benchmark of AI until the year 2000. In 1997, the win of the rule-based expert system Deep Blue against chess world champion Garry Kasparov put AI systems back into the public spotlight. Since the 2000s, AI systems have increasingly been referred to as knowledge-based systems. Starting in 2010, the increase in Big Data and cloud-computing power led to the increased affordability of Artificial Neural Networks (ANNs). ANNs mimic the neurons of the brain through interconnected layers of nodes. As the reasoning process of these applications is often difficult, if not impossible, to understand for humans, the term “artificial intelligence” regained popularity. The immense power of these neural networks was demonstrated in 2016, as the reigning world champion in Go, a game exponentially more complex than chess, was beaten by an AI (Chen, 2016). While the used techniques are often different, the terms Expert Systems, Knowledge-Based Systems, and Artificial Intelligence are mostly related to fashion rather than an underlying technical distinction (Duan et al., 2019). The emergence of Big Data, however, is much more than a cyclical upswing. Big Data enables AI to become exponentially more powerful.

AI can be categorized into three main business capabilities (Davenport & Ronanki, 2018). The first and most common is robotic process automation, where typically back-office administrative work is replaced by a system that can deliver the same result. The second entails cognitive insight, an improved version of data analytics. These systems are trained and able to work with vast volumes of data and improve over time. The third, most advanced, and least common capability is cognitive engagement. These applications are intelligent agents that engage the user. Chatbots present a well-known example of this capability. Categorizing AI into these three functionalities helps managers to identify the potential power of computers.

Automation Through Artificial Intelligence

From a manager’s perspective, AI can be utilized in the three mentioned capabilities: robotic process automation, cognitive insight, and cognitive engagement. Despite 46% of projects focusing on robotic process automation, the researchers argue that AI does not replace humans (Davenport & Ronanki, 2018). They state that “in most of the projects [they] studied, the goal was not to reduce head count but to handle growing numbers of employee and customer interactions without adding staff” (ibid., p. 6). This view, however, fails to see the long-term. The decoupling of growth and jobs through AI automation is unprecedented in history, and can indeed lead to a reduction in the number of jobs.

While most projects focus on automation, Davenport observes that especially inexperienced managers limit themselves to a sole cost-reduction through automation (Deloitte, 2017). The most advanced managers strive to focus on augmentation (ibid.).

Augmentation Through Artificial Intelligence

Augmentation of humans’ abilities through computers aims to supplement human cognitive capabilities in a symbiotic manner. Since the origins of the computer in the 1940s, parallel progressions of work emphasized the use of computing for automation as well as for augmentation. As early as 1960, Licklider published the article “Man-Computer Symbiosis”, where he described the ultimate goal to enable people and computers to cooperate in decision-making and complex situations without the inflexible dependence on programmers. Current-day researchers also believe that this symbiotic relationship of augmentation delivers superior performance for businesses. In their Harvard Business Review article “Beyond Automation”, the two leading researchers Davenport and Kirby call to integrate humans and smart machines by asking:

What if, rather than asking the traditional question: What if tasks currently performed by humans will soon be done more cheaply and rapidly by machines? We ask a new one: What new feats might people achieve if they had better thinking machines to assist them? [...] We could reframe the threat of automation as an opportunity for augmentation.

Reframing AI from automation to augmentation makes all the difference for employee morale and business success.

How Employees Need to Adapt - Davenport’s Five Ways of Stepping

Cognitive technologies might soon do a lot of the cognitive heavy-lifting in a majority of professions. This inevitable augmentation of jobs will cause the need to renegotiate employee’s relationship to AI-enabled machines. A proposed framework introduces Ways of stepping in which employees can realign their contributions in collaboration with AI (Davenport & Kirby, 2015).

Source: Miller, S. M. (2018).

Stepping in will present itself as the best opportunity for many employees. These highly qualified individuals can build a link between human and machine expertise. Using the computer’s quantitative capabilities establishes a solid base of understanding that can be expanded upon through the human's non-quantifiable expertise. Employees striving to bridge this gap need to be fluent not only in the technical aspects of the system but also in the business domain they are operating within.

Stepping aside will emerge as the best alternative for many individuals that lack the technical skills required to step in. These employees should focus on the aspects of their work that are least likely to be quantified. By being an expert in the business' interpersonal elements, they can step aside, take advantage of the smart machines, and still deliver value to the company.

Stepping up, narrowly, and forward can, by definition, only be relevant to a minority of people and hence are not covered in detail.

Human-Machine Collaboration - The Missing Middle

While the Ways of Stepping provide a holistic view of individuals’ options, they lack the detail to pinpoint specific opportunities for employees. Fortunately, others have recently studied 1500 companies and traced the most significant performance improvements to human-machine collaboration (Wilson & Daugherty, 2018).

Source: Miller, S. M. (2018).

This framework highlights the growing range of work activities best tackled by a symbiotic human-machine effort - between those tasks that remain human-only and machine-only. The utter lack of attention both from academia as well as businesses in this crucial domain complements the name - “Missing Middle”. The framework provides a detailed structure of human-AI collaboration opportunities, divided into tasks where humans are complementing machines and those where machines assist humans.

In conclusion, these two frameworks show that while the underlying technologies are the same, the implications of automation versus augmentation could not be more different.

Automation Versus Augmentation Through AI

Research claims that if companies want to achieve sustainable performance increases through AI, they should focus on augmentation over automation (Wilson & Daugherty, 2018). Davenport and Kirby support this view and elaborate upon the critical differences between the two perspectives and the resulting impact on employee morale, as they claim that employees hate automation and love augmentation.

Source: Miller, S. M. (2018).

Automation focuses on the shortcomings of humans compared to machines with the goal to replace employees and leaves them worse off through lay-offs or pay cuts (Davenport & Kirby, 2016). Augmentation, on the other hand, looks at the shortcomings of humans with the goal of improving employees’ capabilities (ibid.).

Source: Miller, S. M. (2018).

While the underlying technologies are the same, the applications could not be more different. Automation limits AI to reduce costs by replacing humans. Augmentation, however, encourages collaboration between humans and machines, creating value by building on each other's strengths. 

Conclusion and Managerial Implications

This article started, by showcasing the historical developments of AI, from the 50's all the way to the current age of Big Data. Since the beginning, two perspectives oppose each other: automation and augmentation. Davenport and Ronanki show that most AI projects have automation as the primary goal, while they should focus on augmentation. By contrasting the two, this article highlights that augmentation is more likely to be successful when embraced by companies, as it does not seek to replace employees, but rather seeks a collaborative partnership of humans and machines. The presented leading frameworks provide further insights on how augmentation can deliver superior value. Augmentation is viewed much more favorably than automation, as it creates a win-win-win situation, and provides employees with a multitude of advantages.

To illustrate the advantages of augmentation, Autodesk’s product design tool “Dreamcatcher” serves as an ideal example (AutodeskResearch). The AI tool enables the user to input the hard requirements of the product, to then meticulously calculate hundreds of possibilities from which the designer can select his preferred options. Based on the options chosen, the system adapts and puts out another set of improved proposals until the best possible design is found. This collaborative process combines the strengths of humans and AI, results in a superior product, and does not replace humans. The augmentation tool is, therefore, more likely to perform better and to be accepted quicker than the automation of decision-making.

The superior function of AI for decision-making is to augment. Managers need to re-evaluate their perspective and look for opportunities for augmentation rather than automation for AI in decision-making. Through augmentation, artificial intelligence can hands down, beyond contest, change the world.

I am a recent graduate from Maastricht University, and the presented insights are summarized from the literature review of my Bachelor's Thesis. I have genuinely enjoyed researching these findings, and if you find the topic as fascinating as myself, please share your thoughts in the form of a comment, or feel free to reach out to me!

Source for graphics: Miller, S. M. (2018). AI: Augmentation, more so than automation. Asian Management Insights. 5, (1), 1-20. Asian Management Insights. Retrieved from: https://ink.library.smu.edu.sg/ami/83