Book review: Ethan Mollick. 2024. Co-intelligence: Living and Working with AI

In Ethan Mollick’s book Co-intelligence: Living and Working with AI, Mollick discusses how the world is changing at a rapid pace and how it takes sleepless nights just to grasp the way AI is transforming our lives. The author adeptly engages the audience by illustrating how generative AI is reshaping the way we work and raising concerns about the extent of the changes AI might bring as it advances, as well as, the potential consequences. Mollick argues that similar to other general-purpose technologies that have brought significant changes, this will do the same. However, unlike previous technologies that took years to be adopted, this one is spreading much faster due to its free access, ease of availability, and incredibly useful. Throughout the book, the author presents both the advantages and drawbacks of integrating this new wave of AI, particularly generative AI into daily life. He asserts that while past general-purpose technologies primarily impacted mechanical and repetitive tasks, this one might work as a co-intelligence, urging readers to fully embrace it now rather than waiting for others to direct its use. Mollick emphasizes that AI is influencing nearly every aspect of life. Moreover, he highlights what he calls “the alien in the room”—the larger issue of whether AI can replace humans or match human creativity. The book proceeds with the sentiment, “No one really knows where this is all heading,” and explores both the potential implications and benefits throughout.

The book is spread over two parts and organized into nine chapters, with the first chapter titled “Creating Alien Mind” introducing the evolution of AI over the years, beginning with chess computers in the 1770s and moving to toy experiments in the 1950s. The book presents how much of the initial work involved theoretical approaches, which later led to computer scientists actively experimenting with algorithms. Further, it extends to how investment in AI surged in the 2010s with the advent of machine learning techniques for analysis and prediction. The author mentions, “AI was a poor (albeit marketing-friendly) label for what this sort of software did, as there was very little about these systems that actually seemed intelligent or clever, at least in the ways humans are.” The author describes how initially AI showed significant progress in industries like e-commerce, with companies such as Amazon, but it also had notable limitations such as, for instance, these systems could only make predictions based on past data and struggled with new, unknown situations, as they relied on supervised learning techniques with labelled data. Another limitation also mentioned in the book was lacking human-like understanding. The book highlights how a major breakthrough in AI came with the research paper “Attention is All You Need.” The transformer model introduced in this article provided new ways for computers to better process human communication, enabling them to understand context more effectively. The chapter explains the technical aspects of how these transformer models work in an accessible manner. It then transitions to an introduction to different types of large language models (LLMs) and the phases of learning behind them. The author points out the limitations in training these large models, noting that the vast sources of data needed are finite and may soon be exhausted. The author also discusses the emergence of reinforcement learning, which incorporates humans in the loop, as the next step in training. The book provides intriguing examples that showcase the capabilities of LLMs and highlight how they can appear both strange and fascinating. Since the book primarily focuses on LLMs, it details their evolution and how their capabilities have grown over time. The author in the chapter concludes by arguing that we have created an “alien mind” that behaves like a human but is not human, raising the question of whether this alien is friendly, which ties into the alignment problem which is further discussed in the next chapters.

In the second chapter titled “Aligning the Alien,” the author discusses the potential issues AI may bring in the future and how to align the advancement of AI ethically. The book clearly explains how AI could lead to AGI (Artificial General Intelligence) and eventually to artificial superintelligence, and how AI could become dangerous due to hidden and unexpected motives. The author emphasizes that well-aligned AI, however, could use its power to serve humanity rather than destroy it. The author also notes that alignment research is no easy task, as it involves multiple aspects, some of which are inherently problematic, as humans themselves also struggle with alignment. The chapter presents statistics showing how some research suggests AI could lead to humanity’s destruction, but the possibility of reaching AGI is still uncertain. The author argues that while the potential harms are alarming and deserve attention, the focus should be on the present, as AI is already affecting our lives. If we are not actively aware of these advancements, we may be forced to accept decisions made by others. The chapter further explores the legal and ethical issues AI may raise and how they are already coming into play. It discusses how reinforcement learning from Human Feedback can reduce biases, but how some companies use unethical practices in training AI. The book presents how the capabilities of AI could also be misused to harm society in unethical ways. The author asserts that it is the duty of government officials to enact sensible regulations and calls for society to adopt a unified approach to solving these problems, ensuring clarity about the future we want to create.

The author proceeds to the next chapter titled “Four rules for Co-Intelligence,” where he introduces his concept of co-intelligence by proposing four principles for co-intelligence. According to the author, the first principle emphasizes always inviting AI to the table. He states that regardless of the biases it may bring, we should be willing to work with it to understand its true potential. In his second principle, he states the “human in the loop” approach, meaning that humans should remain involved in continuously making judgements about AI’s role in life as people must be able to handle the hallucinations and biases AI may produce and learn how to adapt and overcome them. The third principle as per author is to treat AI like a human. The author himself uses anthropomorphism in the book—treating non-human entities as human-like. He argues that this perspective will provide more clarity, suggesting that we should think of AI as an “alien human” rather than a “human-like machine.” He states that this approach helps us relate to it better, as it aligns with the human tendency to relate to non-human things as if they were human. Additionally, the author explains that we can instruct AI to behave like the persona we need, treating it like a human intern that requires guidance and supervision. The final principle is to assume that the AI you are using is the worst version you will encounter. The author argues that as technology progresses daily, AI will only advance, so it’s important to be aware of the boundaries it may push and understand it better.

The second part of the book talks about the real-life applications of co-intelligence. Throughout the book, the author emphasizes AI being co-intelligence and not a brain of its own, and hence, humans are not likely to become obsolete. In the next chapters, the author describes how AI can work as co-intelligence. The book then delves deeper into how AI can work as a person, a creative, a co-worker, a tutor, and a coach. First, in the fourth chapter titled “AI as a person,” the author, in the initial arguments presented in the book, attempts to break the misconception that because AI is software, it should behave like traditional software. The book argues that AI is fundamentally different from conventional software. The author emphasizes thinking of AI more as a human than as software, providing points to support this perspective. It is shown how AI can estimate metrics like willingness to pay in ways consistent with existing research. Additionally, the author highlights AI’s unique advantage of adopting different personas according to developer and user needs. The book offers compelling examples illustrating how earlier forms of AI struggled with tests like the Turing Test, but as AI has advanced, it now surpasses such tests with ease. The chapter explores the progression of AI’s abilities to perform in Turing Tests, with the author providing interesting examples of AI responding naturally when asked to act like a human, reinforcing the idea of its human-like qualities. The book also discusses the potential future of AI with capabilities like AI companions, personalized engagement, and even AI therapists. Alongside these possibilities, the author acknowledges issues such as social isolation but advocates for treating AI as “co-intelligence” while remaining aware of these concerns.

In the following fifth chapter titled “AI as a creative,” the author continues to present his arguments, such as how AI can be creative. He acknowledges that although AI has issues like hallucinations and biases, and cannot be used for mission-critical tasks that require accuracy, it can still be quite creative and aid in innovation, as its randomness can lead to new ideas. The author emphasizes that while AI excels in creativity, it has its limitations and requires humans to filter and select its outputs. He makes a compelling case for how the definition of creativity will evolve in the near future and how staying attuned to advancements will lead to more innovations and breakthroughs. Although concerns about people becoming overly reliant on AI will persist, the author argues that, like previous technologies, AI will only change what we do by giving us more power.

Further in the sixth, seventh and eighth chapters titled “AI as a co-worker,” “AI as a tutor,” and “AI as a coach,” the author presents a different viewpoint, suggesting that AI can work as a co-worker, tutor, and coach. In explaining each role, he outlines how tasks will be divided between what is suited for AI, what can be delegated to AI, and what can be fully automated by AI. He adds humour to make his points clearer and more engaging, emphasizing how we should view AI as both centaurs and cyborgs—helping humans in some cases and working alongside them in others. The author further notes that as AI advances, tasks and boundaries will continue to shift. He also discusses how people are already using AI in organizations secretly, as managers remain resistant to its use. However, he argues that this resistance cannot be stopped, and managers should embrace AI’s potential to transform the work experience. The author insists that AI will not replace jobs but will change how we perform them. In later chapters, he explores new possibilities AI could bring, such as addressing issues like student misuse of AI, while also highlighting opportunities like flipped classrooms and AI acting as a personal tutor—something hard to find otherwise. In the final chapter, he emphasizes AI as co-intelligence, arguing that AI can be an effective coach to improve our work. The author strongly contends that as AI advances, we will require more human experts, and while becoming an expert takes significant effort, AI can help us push the boundaries of knowledge.

In the final chapter titled “AI as our future,” the author presents his views on what the future with AI may look like, estimating four potential scenarios. He emphasizes that if AI reaches its peak, its limitations may cause further stagnation in its advancements. Legal concerns could also hinder its growth. The author notes that even in the current scenario, the concerns are becoming increasingly difficult to manage, and AI has already impacted many industries. So, even if it does not advance further, it will still affect our lives and work as co-intelligence. He further suggests that slow growth would provide society the opportunity to adapt, with some jobs being created while others are destroyed. He also explores the possibility of exponential growth, explaining how new problems may arise alongside the benefits it could bring. The book presents interesting facts and thought-provoking points about what the future may hold. Lastly, the author imagines a scenario where machines surpass humans, which could be both horrifying and a challenge to the limits of human intelligence. This vision encourages readers to reflect deeply on the potential future shaped by advancements in AI.

The book becomes even more engaging as the author shares his real-life experiences with AI throughout the chapters, including the original chats he had with it. This personal touch adds an intriguing element, allowing readers to better grasp the viewpoints and concepts presented. Moreover, the author offers a collection of fascinating information and innovative scenarios about the future of AI, making the book very appealing to beginners in the field. These stimulating concepts arouse interest and prompt readers to critically evaluate the possible effects of AI on society, employment, and creativity. Overall, the book serves as a useful introduction and a source of motivation for ambitious researchers and those starting their adventure in the AI field.