Artificial Intelligence - critical summary review

The authors explore the profound economic and business implications of artificial intelligence, with a particular focus on Machine Learning (ML). They draw parallels between AI and past transformative technologies like the steam engine, electricity, and the internal combustion engine. These technologies revolutionized industries by spurring complementary innovations and reshaping business practices, and AI is poised to have a similarly transformative impact.

They argue that while AI's potential is vast, it remains underutilized in many sectors. This book is a pivotal addition to the Harvard Business Review Press series that focuses on essential business topics. This volume specifically addresses the rapid advancements and implications of artificial intelligence (AI) in today’s complex and uncertain global marketplace.

Edited by Thomas Davenport, this book offers a curated selection of twelve articles from HBR that explore various facets of AI. Davenport’s introduction sets the stage by emphasizing that while AI is a transformative technology, it remains poorly understood by many. He argues that AI is booming in certain business segments but hasn’t yet fully revolutionized the industry. The volume provides a comprehensive overview of AI's current applications, ranging from improving decision-making to automating routine tasks and enabling advanced robotics.

Separating hype from practical potential

The opening chapter outlines the essence of AI and ML, explaining that machine learning represents a significant departure from traditional software development. Unlike conventional methods, where systems are explicitly programmed with specific instructions, ML systems learn from examples and improve their performance over time. This ability to learn from data, rather than relying on pre-programmed rules, marks a fundamental shift in how technology is developed and applied.

ML systems can now achieve superhuman performance in various domains, such as image and speech recognition, and are increasingly used in industries ranging from finance to healthcare. The authors highlight the recent advancements in AI, particularly in perception and cognition. In perception, AI systems have made remarkable strides in areas like voice recognition and image analysis. For instance, speech recognition technology, as used in virtual assistants like Siri and Alexa, has become significantly faster and more accurate.

Similarly, image recognition systems have improved dramatically, enabling applications such as facial recognition and automated tagging on social media. These advancements have been driven by deep learning techniques, which utilize large datasets and sophisticated algorithms to enhance accuracy and performance. However, Brynjolfsson and McAfee caution against the overhyping of AI’s capabilities. They point out that many business claims about AI’s potential are exaggerated and that AI systems often fall short of their promises in real-world applications.

For example, while AI can excel in specific tasks, such as recognizing images or detecting fraud, it struggles with tasks that require general intelligence or understanding context. This limitation is partly due to the narrow scope of current AI systems, which are typically trained for specific tasks and do not generalize well across different domains.

The chapter also addresses the practical implications of AI for businesses. It notes that while many companies have started integrating AI into their operations, the technology’s full potential is still largely untapped. The key to leveraging AI effectively lies in overcoming barriers related to management, implementation, and business imagination. The authors emphasize that successful AI adoption requires more than just implementing technology; it involves rethinking business models and processes to fully exploit the capabilities of AI.

Furthermore, Brynjolfsson and McAfee discuss the various types of machine learning, including supervised, unsupervised, and reinforcement learning. Supervised learning, where models are trained on labeled data to predict outcomes, has seen significant success in applications such as fraud detection and personalized recommendations. Unsupervised learning, which involves discovering patterns in unlabeled data, is still developing but holds promise for uncovering new insights. Reinforcement learning, used in applications like game playing and optimization tasks, is also emerging as a powerful tool for solving complex problems.

From small data sets to big impact

The chapter "Putting Machine Learning to Work" outlines three key advancements that benefit organizations adopting ML. Firstly, AI skills are increasingly accessible due to the proliferation of online educational platforms and talent marketplaces like Udacity, Coursera, Upwork, and Kaggle, which help both in-house training and hiring skilled experts. Secondly, ML infrastructure has become more affordable and accessible through cloud services provided by tech giants such as Google, Amazon, and Microsoft.

This competitive environment drives down costs and expands capabilities. Lastly, the chapter emphasizes that organizations don’t always need massive data sets to benefit from ML. For instance, Sebastian Thrun and Zayd Enam demonstrated how analyzing chat room logs improved sales performance by 54%, illustrating that useful insights can be gleaned from relatively small data sets.

The chapter also discusses practical applications of ML, including WorkFusion’s use of ML for automating complex back-office processes and the transformative potential of ML in various business aspects. For example, ML is reshaping tasks like cancer cell detection, optimizing workflows in Amazon’s fulfillment centers, and evolving business models to offer personalized recommendations. Importantly, ML often complements rather than replaces human work, improving efficiency and effectiveness without fully automating tasks.

However, the authors also highlight significant risks associated with ML, such as low interpretability of deep learning models, potential biases embedded in training data, and challenges in diagnosing and correcting errors in complex ML systems. Despite these risks, ML systems offer consistent and improvable alternatives compared to human decision-making.

They also reflect on the future of AI and ML, suggesting that while machines are advancing rapidly, there remain areas where human skills are irreplaceable, such as determining what problems to tackle and inspiring collective action. They argue that the true advantage in the AI era lies in combining machine capabilities with human ingenuity, emphasizing the need for leaders who are adaptable and willing to experiment with new technologies.

Candela's approach to AI at Facebook

Candela, a seasoned AI expert, provides a grounded explanation of artificial intelligence to demystify the technology for those outside the field. He describes machine learning algorithms as lookup tables where inputs, such as images, are matched to outputs, like labels ("horse" for a horse picture). The core idea is that the algorithm learns from numerous examples, comparing new inputs to stored ones to classify them accurately. As the data set grows, the challenge becomes managing and processing vast amounts of information efficiently.

Candela likens machine learning to approximating a massive table of data with a function to simplify and expedite comparisons. Candela’s approach to AI at Facebook deviates from the typical focus on creating the latest algorithms. Instead, he emphasizes practical application over theoretical advancements. When he joined Facebook in 2012, he inherited a robust but outdated ranking algorithm for ads, which he compared to the Soyuz spacecraft—functional but not cutting-edge.

Rather than immediately overhauling the algorithm, Candela focused on enhancing the data quality and speeding up experimentation. This strategy aimed to derive more value from existing algorithms rather than solely pursuing new ones. He argues that while improving algorithms is important, the real value lies in how effectively these algorithms impact the business. This philosophy reflects a shift from the conventional academic pursuit of algorithmic perfection to a more business-oriented approach.

A critical aspect of Candela’s approach is the placement of AML within the product development pipeline. He situates AML in Horizon 2 (H2), a space between research and product delivery, which allows for translating advanced AI research into practical applications without being too distant from the business impact or too involved in day-to-day product development. This positioning ensures that the AML team focuses on the immediate application of AI innovations rather than getting bogged down in long-term research or direct product implementation.

Candela’s philosophy highlights the importance of making AI tools accessible and impactful. He encourages a shift in focus from developing the most sophisticated algorithms to solving real business problems effectively. His approach includes fostering a culture where AI solutions are designed to be readily adopted by various teams, thereby driving broader integration and application. The success of this strategy underscores the need for AI teams to balance technical innovation with practical, business-driven outcomes.

Why delays are counterproductive

The authors challenge the notion that AI technologies are not yet mature enough for widespread implementation. They emphasize that many AI technologies, such as traditional machine learning and deep learning, are already based on well-established mathematical and statistical foundations. These technologies have been in development for decades, with deep learning tracing back to research from the 1980s. Thus, waiting for these technologies to mature further is unnecessary because the foundational principles are already robust and well-understood.

One major issue with delaying AI adoption is the time required to develop and customize AI systems for specific business needs. AI systems, especially those involving machine learning, require extensive training data and must be tailored to fit the unique knowledge domains of a business. This process involves not just coding but also significant "knowledge engineering"—a process of incorporating specific industry knowledge into the AI system. This can be particularly challenging and time-consuming, as illustrated by the example of Memorial Sloan Kettering Cancer Center’s extended efforts to develop a cancer treatment system with IBM’s Watson, which remains a work in progress despite high-quality expertise.

The human element of AI implementation adds another layer of complexity. AI systems are usually designed to augment rather than replace human roles, requiring significant retraining of staff to adapt to new systems and processes. For instance, investment firms using "robo-advisors" may need to shift human advisors' focus to new areas like behavioral finance, which involves different skills than traditional investment advice. This retraining process can be lengthy and requires careful planning to integrate new roles and skills into existing teams.

Additionally, the chapter discusses the importance of ongoing governance and monitoring of AI systems. Unlike traditional software, AI systems require continuous oversight to ensure they remain effective and unbiased. Algorithms must be regularly updated to reflect new data and business contexts, and safeguards must be in place to detect and prevent potential biases and fraudulent activities. Effective governance includes monitoring AI for changes in customer demographics and ensuring that systems do not become targets for manipulation.

The authors conclude that early adopters of AI have a substantial advantage. By the time a company that delays AI adoption completes the necessary preparations, early adopters will have already gained significant market share, operating with lower costs and enhanced performance. Companies aiming to stay competitive should start integrating AI technologies now, either by creating internal AI teams or acquiring startups with existing AI capabilities.

Building AI literacy across your organization

The authors argue that for organizations to effectively harness the benefits of artificial intelligence (AI), it is crucial for all employees—not just tech experts—to understand the technology. They advise that while AI is often depicted through sensationalistic scenarios or its potential to displace jobs, its most impactful applications are often mundane and aimed at improving efficiency in everyday tasks. Martinho-Truswell proposes that employees should be able to answer three fundamental questions to make the most of AI: how does it work? What is it good at? And what should it never do?

To address "How does it work?," the chapter explains that while employees don't need to have technical expertise, they should understand the basics of how AI systems process information. This includes recognizing that AI relies on large datasets and algorithms to identify patterns, contrasting this with human learning processes that simplify data. Understanding data's role in AI helps employees appreciate its capabilities and limitations.

By examining tools that employees already use, such as expense management software or recommendation systems, employees can better grasp AI’s practical applications and recognize tasks that are beyond its reach. Lastly, for "What should it never do?," the authors cautions against deploying AI in situations that require nuanced judgment or ethical considerations, such as hiring decisions or sensitive management issues. AI cannot grasp biases or the broader implications of its suggestions, which underscores the importance of setting clear boundaries on its use.

Martinho-Truswell concludes that organizations will thrive by not only investing in AI technology but also ensuring that all team members are educated about AI’s capabilities and limitations. This holistic understanding empowers employees to identify opportunities for AI integration and to apply it thoughtfully, preserving the essential human elements of their roles.

Final notes

Thomas H. Davenport and the Harvard Business Review team present a timely and essential guide to understanding and implementing AI in today’s fast-evolving business landscape. This volume is a must-read for anyone seeking to stay ahead of the curve in an age where AI and machine learning are revolutionizing industries and reshaping societal norms. Davenport, a leading expert in the field, sets the stage with an insightful introduction that dissects the common misconceptions and varied perceptions surrounding AI. He effectively demystifies the technology, illustrating that while AI is often viewed through a lens of fear and hype, it is fundamentally a tool with transformative potential for businesses willing to embrace it.

The book brings together twelve carefully selected Harvard Business Review articles, each contributing valuable perspectives on AI’s impact and potential. These articles cover a range of topics, from making faster, more informed decisions to automating repetitive tasks and enabling robots to interact emotionally. The collection offers practical case studies and research that help readers understand how to implement AI initiatives effectively and gain a competitive edge.

One of the book's strengths is its ability to distill complex AI concepts into actionable insights. It provides a foundational introduction to AI and its implications, making it accessible to executives, employees, and business leaders across various industries. The anthology’s structure ensures that readers can grasp the technology’s significance and practical applications without getting lost in technical jargon.

12min tip

If you’re a scientist or engineer who finds the concept of marketing your work intimidating or challenging, “The Giant’s Ladder,” by Elizabeth Chabe, is a must-read. It covers everything from understanding your audience to advanced strategies in product positioning and campaign planning.