Co-Intelligence: A Practical Guide to the AI Revolution – book review

A Practical Guide in a Sea of Speculation

Ethan Mollick’s “Co-Intelligence” offers a concise and practical introduction to the current state of artificial intelligence, specifically large language models (LLMs). As a professor at Wharton and a respected voice in the AI community (I strongly recommend his social media presence and newsletter), Mollick brings his expertise to bear in this accessible guide for those looking to understand and leverage AI in their daily lives and work.

This is not a book focused on the philosophical (or possibly apocalyptic) issues that have come with the current explosion in capabilities. The book provides a grounded yet fun exploration of the practical applications of today’s AI tools.

The Rise of AI: An Accessible History

Mollick kicks things off with a concise history of AI and LLMs. He paints a picture of an alien intelligence that acts more like a human than traditional software. These AI ‘minds’ excel at tasks we consider quintessentially human, like writing and rhyming, but often stumble on “machine” tasks like arithmetic and logic. The first piece of great advice is to spend 10 hours getting to know an LLM. You’ll discover their strengths, and weaknesses and quirks—like their strange affinity for the number 42.

Mollick’s humour is a delightful aspect of the book. Early sections juxtapose AI’s superhuman abilities with funny failures, often in rhyming sentences. One memorable example of “jailbreaking” ChatGPT to provide a formula for making napalm had me laughing out loud  – Yarrr matey! 

This is followed by a brief overview of the major issues such as alignment – and why they are not the focus of this book. More blogs on this soon. 

Four Principles of Working with AI

The heart of the book introduces AI as a co-intelligence—a powerful tool to augment our capabilities. Backed by academic evidence Mollick presents four key principles:

1. Always Invite AI to the Table: Start by asking how AI can help with any task—you might be surprised by the results.
2. Be the Human in the Loop: AI can be an incredible co-intelligence, but it doesn’t actually know anything. They are prediction machines designed to please the user over being accurate, much like superhuman interns with zero real-world experience.
3. Treat AI Like a Person (but tell it what kind of person it is): Generic prompts get generic answers. Specify who you want the AI to emulate—an expert recruiter, Nobel-winning scientist, or four-star general—and you’ll get more tailored responses.
4. Assume This Is the Worst AI You Will Ever Use: We’re not at the top of the AI progress curve. You don’t have to subscribe to the vision of AGI or Superintelligence to accept that the tools will become more powerful, quickly. Stay open to new developments.

AI as a Creative, as a Coach, as a Coworker and as a Tutor

The following sections are useful but focused on areas where the author, an academic, has a lot of experience. This is actually an important point. A co-intelligence is best used by experts in their own fields. You are the expert in your specific field. In a more recent newsletter Mollick quipped “we are all in R&D now” indeed you are best placed to leverage the new technology in your day to day life and work, there will be surprising results. Short term I think these ‘mundane’ uses will be the major driver of changes to our working lives. You will need to try yourself. Helpfully, he does provide custom GPTs to be used with ChatGPT. I found some more useful than others. A sign of things to come in publishing. 

The Future of Work: A Balanced View

Worried about AI taking your job? Mollick offers a nuanced view. Yes, things will change, but it’s not about replacement—it’s about augmentation. AI proficiency has a jagged edge; some tasks are well-suited to AI, while others are not. Finding the right balance through testing is key. Mollick’s research supports this, showing that those who augmented their tasks with AI performed better than those who fully automated. Tasks to delegate to AI should be tedious, repetitive, and boring—though by no means unimportant or easy.

Practical Tips and Final Thoughts

There are some prompting tips—nothing new if you follow the space, but useful nonetheless. Notably there is an accurate prediction that advanced prompting will soon be automated, this is already released by Anthropic and is the focus of many GPTs. 

While “Co-Intelligence” might not dive as deep as AI enthusiasts hope, it’s an excellent primer for anyone looking to quickly understand and harness the power of AI today. It’s like a collection of Mollick’s top-tier social media posts and blog articles, all in one place. This may have been a deliberate decision to help the book stay relevant; intriguingly there is not much in direct comparisons of the major tools either; as they evolve so rapidly it would date the book. I would say it’s good for those who follow his work and AI closely, great for those that don’t but who are interested in upskilling quickly.