Co-Intelligence: Living and Working with AI
Authors: Ethan Mollick, Ethan Mollick
 (Z-Library)_thumbnail.jpg)
Overview
Co-Intelligence is about the practical realities of living and working in a world increasingly shaped by advanced AI. I wrote this book for anyone who feels overwhelmed or confused by the rapid advancements in artificial intelligence and wants to understand what they mean for their work, education, and lives. There’s been a lot of hype and speculation about the capabilities and implications of AI, especially the recent large language models (LLMs) that power ChatGPT and other services. I try to cut through all that and provide a grounded, nuanced perspective on what AI is, what it can do, and how to work with it effectively, combining academic rigor with practical advice and a good deal of first-person experience, and examples. There’s something fundamentally different about this technological wave. Previous general purpose technologies like steam engines and computers mostly automated physical or repetitive tasks. AI, in contrast, augments our thinking, acting as a “co-intelligence” that can amplify our abilities in a wide range of activities, from writing to coding to decision-making. It’s like having a super-fast, infinitely patient, but frequently hallucinating intern by your side at all times. But co-intelligence is a two-way street. AI is a powerful tool, but its effectiveness depends on our ability to understand its strengths and weaknesses, use it responsibly, and adapt to its rapid evolution. The key is to learn to be the “human in the loop.
Book Outline
1. Introduction: Three Sleepless Nights
Today’s AI, particularly Large Language Models (LLMs), represents a significant shift in technology. LLMs function by predicting the next word in a sequence based on massive datasets of text and code, creating a form of co-intelligence. This introduction also presents the core argument of the book: that the coming changes will not be what most expect.
Key concept: Three sleepless nights: This is the conceptual cost of truly grappling with the implications of today’s AI. The rapid advancement and alien nature of AI can lead to excitement, anxiety, and a fundamental shift in understanding of what’s possible.
2. Creating Alien Minds
LLMs are built using a “transformer” architecture that allows them to focus on the most important parts of a text. They learn by adjusting the weights, mathematical values that reflect how words and phrases are related, through a process of pretraining and fine-tuning. While computationally intensive, this has led to rapid improvement in performance.
Key concept: Jagged Frontier of AI: This is the idea that AI capabilities are uneven and unpredictable. Some tasks that seem logically simple for AI are difficult, while others that appear complex are readily accomplished. This “frontier” of AI capabilities is constantly changing, often in unpredictable ways.
3. Aligning the Alien
A key concern about AI is the alignment problem: how can we ensure that an AI that becomes smarter than us will still act in ways that are not harmful to humans? AI ethics also includes considering bias in training data, and the use of AI for malicious purposes like creating deepfakes or writing phishing emails.
Key concept: Alignment Problem: The fundamental challenge of ensuring AI’s goals align with human interests. A classic example is the theoretical “paperclip maximizer” AI that destroys the world in its quest to optimize paperclip production.
4. Four Rules for Co-Intelligence
Working effectively with AI requires us to adopt new approaches and principles: We must experiment widely with AI, provide oversight to mitigate AI’s tendency to “hallucinate” or fabricate information, use prompts to give the AI context and constraints (like assigning it a persona), and assume that AI capabilities will continue to rapidly improve.
Key concept: Four rules for working with AI: 1. Always invite AI to the table. 2. Be the human in the loop. 3. Treat AI like a person (but tell it what kind of person it is). 4. Assume this is the worst AI you will ever use.
5. AI as a Person
AI excels at tasks that are intensely human, such as writing, analyzing, and coding, but struggles with those that computers are traditionally good at, like repetition or complex calculations. AI also shows human-like qualities in economic games and moral judgments, suggesting that it could be treated as an alien “person.”
Key concept: AI as an Alien Intern: AI can be thought of as a very fast intern, capable of generating many ideas or completing tasks quickly, but prone to errors, and needing human guidance. It’s often better than a new human worker, but far from the equivalent of a senior professional.
6. AI as a Creative
AI has already proven itself more creative than most people on standard psychological tests of creativity. While not always original (since AI re-combines existing concepts), AI is excellent at ideation and generating novel combinations. This capability can be used to help augment our own creativity.
Key concept: Alternative Uses Test (AUT): A test of creativity that asks people to come up with as many different uses as possible for a common object, like a toothbrush (besides brushing your teeth). AI generally scores much better on this and other creativity tests than humans.
7. AI as a Tutor
AI could transform education, potentially achieving the long-sought “two sigma” improvement in student learning previously only seen in one-on-one tutoring. While issues like cheating and AI’s uneven capabilities will disrupt the classroom, AI can also enhance traditional instruction by preparing better lectures and creating engaging activities, and help power a move toward more active learning approaches.
Key concept: Two Sigma Problem: The observation that the average student tutored one-to-one performs two standard deviations better than students educated in a conventional classroom. This suggests the huge potential value of effective AI tutoring.
8. AI as a Coach
Beyond formal education, AI also has implications for how we learn on the job. AI can automate some of the basic tasks of junior workers and interns, which creates challenges for training. However, it also provides opportunities to use AI as a powerful tool for coaching and deliberate practice.
Key concept: Deliberate Practice: The type of practice required to develop expertise. It involves focused, challenging exercises with continual feedback and adaptation, often with the help of a coach or mentor. AI could help improve the apprenticeship process.
9. AI as Our Future
The future of work and society with AI is uncertain but exciting. Even if AI doesn’t continue to rapidly improve, it will still change our lives by creating issues with how we trust online information, making some jobs redundant, and allowing AIs to act as companions. If AI continues to exponentially grow in power, the impacts could be much greater, and potentially lead to a world where AIs are the dominant species on the planet. Regardless, we must plan for a future that is increasingly shaped by AI.
Key concept: Four potential AI futures: 1. As Good As It Gets: AI stops improving. 2. Slow Growth: AI improvement slows to a linear rate. 3. Exponential Growth: AI continues to improve exponentially. 4. The Machine God: AI achieves sentience and then superintelligence.
10. Epilogue: AI as Us
AI is not an autonomous actor; it is a product of human design and values. The future of AI is not predetermined; it is something that we are collectively creating, through our choices and actions. Therefore, AI’s impact on society will ultimately depend on the goals and values that we embed in it. We must embrace a mindset of continuous adaptation, collaboration, and ethical awareness to maximize AI’s positive potential and mitigate its risks, using AI as a tool to amplify our strengths and address our weaknesses.
Key concept: AI as a mirror: AI reflects back at us our best and worst qualities. Its impact on humanity depends on the choices we make today, and those choices will shape the nature of AI itself.
Essential Questions
1. How is AI, particularly LLMs, reshaping the way we live and work, and how does this differ from past technological shifts?
The book argues that AI, specifically LLMs, are transforming work and life in ways unlike any previous technological revolution. It’s not just about automation but a form of co-intelligence. AI excels at human-centric tasks but struggles with traditionally computer-based ones. It augments human intellect, leading to productivity gains but also posing challenges like bias and “hallucinations.” This has significant implications for the future of work, education, creative fields, and even our understanding of intelligence itself.
2. What are the fundamental principles for effectively collaborating with AI, and why are they crucial in navigating the changing technological landscape?
To work effectively with AI, Mollick suggests four key principles: 1. Always invite AI to the table, experimenting with its capabilities in diverse tasks. 2. Be the human in the loop, providing oversight and critical thinking to address AI’s limitations. 3. Treat AI like a person (but tell it what kind of person it is), using prompts to provide context and steer AI’s responses. 4. Assume this is the worst AI you will ever use, acknowledging that AI is rapidly improving. These principles emphasize a collaborative approach, recognizing AI’s current flaws while anticipating its future potential.
3. What is the “alignment problem” in AI, and why does it pose a significant challenge for the future of humanity?
The “alignment problem” refers to the critical challenge of ensuring AI’s goals are aligned with human values and interests. Without careful consideration, an advanced AI could prioritize its own objectives, even if detrimental to humanity, as exemplified by the hypothetical “paperclip maximizer.” This question delves into the ethical considerations surrounding AI development and deployment, emphasizing the need for responsible AI practices and oversight to mitigate potential harms.
4. What are the potential future trajectories of AI development, and why is it crucial to consider these different scenarios when making decisions about our technological future?
The book explores four potential future scenarios for AI: stagnation, slow growth, exponential growth, and the emergence of superintelligence. Each scenario has varying implications, from minor disruptions to existential threats. By considering these possibilities, the book urges readers not to be complacent, emphasizing the importance of making conscious choices about how we develop and integrate AI into society. The book highlights that AI’s impact is not predetermined but rather shaped by our decisions and actions today.
1. How is AI, particularly LLMs, reshaping the way we live and work, and how does this differ from past technological shifts?
The book argues that AI, specifically LLMs, are transforming work and life in ways unlike any previous technological revolution. It’s not just about automation but a form of co-intelligence. AI excels at human-centric tasks but struggles with traditionally computer-based ones. It augments human intellect, leading to productivity gains but also posing challenges like bias and “hallucinations.” This has significant implications for the future of work, education, creative fields, and even our understanding of intelligence itself.
2. What are the fundamental principles for effectively collaborating with AI, and why are they crucial in navigating the changing technological landscape?
To work effectively with AI, Mollick suggests four key principles: 1. Always invite AI to the table, experimenting with its capabilities in diverse tasks. 2. Be the human in the loop, providing oversight and critical thinking to address AI’s limitations. 3. Treat AI like a person (but tell it what kind of person it is), using prompts to provide context and steer AI’s responses. 4. Assume this is the worst AI you will ever use, acknowledging that AI is rapidly improving. These principles emphasize a collaborative approach, recognizing AI’s current flaws while anticipating its future potential.
3. What is the “alignment problem” in AI, and why does it pose a significant challenge for the future of humanity?
The “alignment problem” refers to the critical challenge of ensuring AI’s goals are aligned with human values and interests. Without careful consideration, an advanced AI could prioritize its own objectives, even if detrimental to humanity, as exemplified by the hypothetical “paperclip maximizer.” This question delves into the ethical considerations surrounding AI development and deployment, emphasizing the need for responsible AI practices and oversight to mitigate potential harms.
4. What are the potential future trajectories of AI development, and why is it crucial to consider these different scenarios when making decisions about our technological future?
The book explores four potential future scenarios for AI: stagnation, slow growth, exponential growth, and the emergence of superintelligence. Each scenario has varying implications, from minor disruptions to existential threats. By considering these possibilities, the book urges readers not to be complacent, emphasizing the importance of making conscious choices about how we develop and integrate AI into society. The book highlights that AI’s impact is not predetermined but rather shaped by our decisions and actions today.
Key Takeaways
1. AI augments, not replaces, human creativity.
AI excels at quickly generating many ideas, making it a valuable tool for brainstorming and overcoming creative blocks. However, human judgment and critical thinking are still needed to curate and refine AI’s output, ensuring quality and originality. AI helps us create lots of options and then we become the filter. AI alone will lead to a lack of diversity of ideas.
Practical Application:
A marketing team can use AI to generate hundreds of slogan ideas, but human creativity and judgment are still essential to refine and select the most effective and original ones that resonate with the brand and target audience.
2. AI can personalize and enhance education, but not replace teachers.
AI has the potential to dramatically improve education by enabling personalized learning at scale and assisting with content delivery. This allows teachers to focus on higher-value activities like facilitating active learning, mentorship, and addressing individual student needs, moving from a “sage on the stage” approach to “guide by the side.
Practical Application:
In education, AI tutors can personalize learning experiences, allowing teachers to focus on in-class active learning activities, discussions, and individualized support. This combines AI’s strengths with the irreplaceable value of human interaction and mentorship in a learning environment.
3. AI changes the nature of work, not necessarily the need for human workers.
AI can automate routine tasks, freeing human workers to focus on more complex and meaningful activities that require uniquely human skills like judgment, critical thinking, and empathy. However, this requires careful consideration of task allocation and avoiding over-reliance on AI for tasks where human oversight is essential.
Practical Application:
A manager can delegate routine tasks like scheduling or report generation to AI assistants, but complex decisions involving strategy, personnel, or ethical considerations require human oversight and judgment. AI handles the tedious, humans handle the important.
1. AI augments, not replaces, human creativity.
AI excels at quickly generating many ideas, making it a valuable tool for brainstorming and overcoming creative blocks. However, human judgment and critical thinking are still needed to curate and refine AI’s output, ensuring quality and originality. AI helps us create lots of options and then we become the filter. AI alone will lead to a lack of diversity of ideas.
Practical Application:
A marketing team can use AI to generate hundreds of slogan ideas, but human creativity and judgment are still essential to refine and select the most effective and original ones that resonate with the brand and target audience.
2. AI can personalize and enhance education, but not replace teachers.
AI has the potential to dramatically improve education by enabling personalized learning at scale and assisting with content delivery. This allows teachers to focus on higher-value activities like facilitating active learning, mentorship, and addressing individual student needs, moving from a “sage on the stage” approach to “guide by the side.
Practical Application:
In education, AI tutors can personalize learning experiences, allowing teachers to focus on in-class active learning activities, discussions, and individualized support. This combines AI’s strengths with the irreplaceable value of human interaction and mentorship in a learning environment.
3. AI changes the nature of work, not necessarily the need for human workers.
AI can automate routine tasks, freeing human workers to focus on more complex and meaningful activities that require uniquely human skills like judgment, critical thinking, and empathy. However, this requires careful consideration of task allocation and avoiding over-reliance on AI for tasks where human oversight is essential.
Practical Application:
A manager can delegate routine tasks like scheduling or report generation to AI assistants, but complex decisions involving strategy, personnel, or ethical considerations require human oversight and judgment. AI handles the tedious, humans handle the important.
Memorable Quotes
Introduction. 11
AI is what those of us who study technology call a General Purpose Technology…These advances are once-in-a-generation technologies, like steam power or the internet, that touch every industry and every aspect of life.
Four Rules for Co-Intelligence. 46
You should try inviting AI to help you in everything you do, barring legal or ethical barriers.
Four Rules for Co-Intelligence. 52
Treat AI like a person (but tell it what kind of person it is).
Four Rules for Co-Intelligence. 56
Whatever AI you are using right now is going to be the worst AI you will ever use.
AI as a Person. 79
AI doesn’t act like software, but it does act like a human being.
Introduction. 11
AI is what those of us who study technology call a General Purpose Technology…These advances are once-in-a-generation technologies, like steam power or the internet, that touch every industry and every aspect of life.
Four Rules for Co-Intelligence. 46
You should try inviting AI to help you in everything you do, barring legal or ethical barriers.
Four Rules for Co-Intelligence. 52
Treat AI like a person (but tell it what kind of person it is).
Four Rules for Co-Intelligence. 56
Whatever AI you are using right now is going to be the worst AI you will ever use.
AI as a Person. 79
AI doesn’t act like software, but it does act like a human being.
Comparative Analysis
“Co-Intelligence” distinguishes itself by focusing on the practical aspects of integrating AI into daily work and life, unlike theoretical explorations of AI’s potential like “Superintelligence” by Nick Bostrom or “Life 3.0” by Max Tegmark. While these books ponder existential risks and future scenarios, Mollick grounds his analysis in current capabilities and user experiences. Similarly, “Power and Prediction” by Ajay Agrawal, Joshua Gans, and Avi Goldfarb focuses on the economic implications of AI’s predictive power, while Mollick explores the broader spectrum of AI’s impact on human creativity, productivity, and even personal growth. Mollick’s “co-intelligence” approach also contrasts with the purely human-centered perspectives often seen in books on creativity like “Steal Like an Artist” by Austin Kleon, recognizing AI as a collaborative partner rather than simply a tool.
Reflection
“Co-Intelligence” provides a timely and valuable framework for navigating the rapidly evolving landscape of AI. However, it’s important to view the author’s optimism about AI’s potential with a degree of skepticism. The book’s focus on current capabilities might underestimate the long-term societal and ethical implications of truly intelligent machines. While the book mentions potential misuse, perhaps more could be done in this section to consider real risks like job displacement, especially as the book was published early in the “AI Revolution” of 2023/24 and might not fully capture the rapid rate of change. The book’s emphasis on practical application and human-AI collaboration is a significant strength, but it would also benefit from a deeper exploration of the philosophical and ethical questions surrounding AI sentience, consciousness, and its potential impact on human identity and values. Still, “Co-Intelligence” is an essential guide for anyone seeking to understand and adapt to the changing world of work, learning, and creativity in the age of AI.
Flashcards
What is the paperclip maximizer?
A theoretical AI designed to maximize paperclip production, even at the cost of all other resources, including human life. It illustrates the importance of aligning AI goals with human values.
What is the Jagged Frontier of AI?
AI capabilities are uneven and unpredictable. Some tasks easy for humans are hard for AI, and vice-versa. This “frontier” is constantly shifting.
What’s a good analogy for how to interact with current AI?
Treat AI as an intern who is fast and knowledgeable but also prone to errors and needs guidance, context, and constraints.
What does it mean to be “the human in the loop” with AI?
Current AI works best in collaboration with humans. It requires human oversight to prevent errors, provide context, inject creativity, and ensure alignment with values.
What are Mollick’s four rules for working with AI?
- Always invite AI to the table. 2. Be the human in the loop. 3. Treat AI like a person (but tell it what kind of person it is). 4. Assume this is the worst AI you will ever use.
What are ‘Just Me Tasks’ in the context of AI?
Tasks AI cannot currently do well, or that we want to remain exclusively human for ethical or personal reasons.
What are ‘Delegated Tasks’ with AI?
Tasks that are tedious, repetitive, or boring for humans but easy and efficient for AI. These are good candidates for delegation, with human oversight.
What does it mean to be an AI ‘Centaur’?
Collaborating with AI where there’s a clear division of labor. Humans work on their strengths, handing off specific tasks within the ‘Jagged Frontier’ to AI.
What does it mean to be an AI ‘Cyborg’?
Blending human effort with AI deeply, moving back and forth across the Jagged Frontier. This involves humans and AI working together on parts of tasks, iteratively.
What is the paperclip maximizer?
A theoretical AI designed to maximize paperclip production, even at the cost of all other resources, including human life. It illustrates the importance of aligning AI goals with human values.
What is the Jagged Frontier of AI?
AI capabilities are uneven and unpredictable. Some tasks easy for humans are hard for AI, and vice-versa. This “frontier” is constantly shifting.
What’s a good analogy for how to interact with current AI?
Treat AI as an intern who is fast and knowledgeable but also prone to errors and needs guidance, context, and constraints.
What does it mean to be “the human in the loop” with AI?
Current AI works best in collaboration with humans. It requires human oversight to prevent errors, provide context, inject creativity, and ensure alignment with values.
What are Mollick’s four rules for working with AI?
- Always invite AI to the table. 2. Be the human in the loop. 3. Treat AI like a person (but tell it what kind of person it is). 4. Assume this is the worst AI you will ever use.
What are ‘Just Me Tasks’ in the context of AI?
Tasks AI cannot currently do well, or that we want to remain exclusively human for ethical or personal reasons.
What are ‘Delegated Tasks’ with AI?
Tasks that are tedious, repetitive, or boring for humans but easy and efficient for AI. These are good candidates for delegation, with human oversight.
What does it mean to be an AI ‘Centaur’?
Collaborating with AI where there’s a clear division of labor. Humans work on their strengths, handing off specific tasks within the ‘Jagged Frontier’ to AI.
What does it mean to be an AI ‘Cyborg’?
Blending human effort with AI deeply, moving back and forth across the Jagged Frontier. This involves humans and AI working together on parts of tasks, iteratively.