In his new book on AI Co-Intelligence: Living and Working with AI, Wharton professor Ethan Mollick modestly declares that “while we didn’t think of ourselves as having a special skill in prompting, we found that we were very good at making AI dance to our tune.” Mollick may be right. While he was early to experiment with tools like ChatGPT and then to share his experiences via academic papers and his substack blog One Useful Thing before most people had tried ChatGPT, today anyone can get smarter and more productive using generative AI.
That’s the point of this slim but engaging volume. Because generative AI is trained on billions of words from the Internet and has mastered the patterns and structures of the English (and other) languages, we can all use these tools as a thinking companion for a variety of tasks. We should “try inviting AI to help you in everything you do,” he argues. This “co-intelligence” can “improve our own decision-making, helping us to reflect on our own choices” and the more we use it, the faster we will discover all the intriguing, ingenious, and exciting uses for these digital companions.
In contrast to older tools like the lever or the loom, this is the first set of tools that augment how we think, rather than how we move in the world. The concept of "co-intelligence" represents a significant departure from the way we typically approach human-machine interactions.
In traditional software and automation, we interact with our tools in a highly structured and predictable manner. When we use a calculator or a spreadsheet, for example, we input specific commands or formulas, and the machine generates outputs based on a predetermined set of rules. We know exactly what to expect from the machine, and there is little room for surprise or creativity. In contrast, the AI tools Mollick describes in Co-Intelligence exhibit a level of unpredictability that more closely resembles human interaction. When we engage with a language model like ChatGPT, we don't always know what kind of response we'll receive. This element of surprise is what makes the interaction feel more like a conversation with a knowledgeable and creative partner than a simple input-output transaction.
If you are a reader of Mollick’s excellent and timely blog, you won’t find anything new in here. The book is a highlights reel from One Useful Thing--and that’s a compliment--boiled down to a 40K portable primer on large language models, what they are, and how they are likely to impact education, work and society written in a lively, first-person account. That impact will be momentous but the shape it will take and how fast we will arrive at that future are both uncertain. The book offers a greatest hits of the issues: hallucination, bias, the death of homework. Mollick’s anodyne musings about the inconclusive impact of AI on jobs can also be found in the latest consultant reports, think tank white papers, other research, or, even better, in his own Harvard Business Review article on how AI improves worker productivity.
Where Co-Intelligence bristles with energy is when Mollick shares his own sense of wonder at inhabiting the AUT or astonishing universe of AI tech. Mollick repeatedly cuts and pastes into the manuscript both his prompts and ChatGPT’s replies. It’s a not insignificant chunk of the book, which was unabashedly written with such prompts as “Make this better, in the style of a bestselling popular book about AI.”
It worked.
The reader is a bystander to Mollick’s conversations (or is it an inner monologue?) with the team of AI personas he created to help him write the book. From the pompous, critical editor to the eager collaborator to the ordinary, albeit confused reader, Mollick has created a cast of characters who work alongside him.
Some readers might find the frequent excerpts from ChatGPT to be annoying, especially given the small volume’s $30 price tag. But I loved reading about Mollick talking to his imaginary buddies in the way that a parent enjoys seeing their child blossom as he talks to his magical but invisible companions. The book reminds us that as immigrants to this new AI frontier, we will get the most from these tools when we stop thinking of them as the better calculators and word processors they are and, instead, treat them like the reincarnation of our childhood friends with whom we think aloud about our creative endeavors. Today, for example, I asked Claude a question and it anthropomorphically responded: “One thought that occurred to me as I was reading….”
When we were kids, we used our conversations with our imaginary playmates to talk through our experiences. As we grew older, they faded away as did our imaginations. As AI tools become more advanced and attuned to our needs, they may become an increasingly integral part of our lives, not just as tools but as collaborators and partners in our personal and professional growth. (check out this example of LinkedIn founder Reid Hoffman talking to his AI-self.)
The challenge, then, is to embrace the potential of AI to bring out our imaginations through more conversation, allowing AI to reinvigorate our creativity and productivity. But there is also a risk. AI can make us lazy and careless when we depend on it too much. We might lose ourselves and our agency to the tools.
As we step into this new era of co-intelligence, we have the opportunity to reconnect with a part of ourselves that many of us have lost touch with: our childhood imagination. Mollick's book serves as a valuable companion, reminding us how not, as Henry David Thoreau said, to become “the tools of our tools” but, instead, how to use them to rediscover our own creative potential. In the end, the book does not have much space to offer novel prescriptions about societal improvement. For Mollick, the true power of co-intelligence lies in our willingness to embrace it as a catalyst for personal improvement and individual creativity.
