Co-Intelligence

by Ethan Mollick

Read on July 15, 2024

Rating: ★★★ ☆☆

Summary

AI is eating the world. Thirteen years ago, Mark Andreessen argued that software was eating the world, and that people weren’t ready for the opportunity that existed ahead of them. Today, AI is eating the world, and people aren’t ready for the opportunity that exists ahead of us. Knowing how to leverage the new hotness in AI (Large Language Models) is what will separate good from great. It will separate those who are actually 10x producers and those who are managed by them.

Key Takeaways

  1. Keep the four “rules” for co-intelligence in mind to leverage AI effectively:
  • Invite AI to the table. It will help you find the edges of your problem space and combat biases.
  • Be the human in the loop. LLMs will hallucinate and need your help to find the diamonds in the rough.
  • Treat AI as a person. You should tell it what it is and how you want it to behave.
  • Assume this is the worst AI will ever be. It’s only smarter, faster, stronger(?) from here.
  1. Knowledge is a commodity. Everyone can create convincing arguments for any topic in seconds. Knowing what to with the knowledge is more important.
  2. Having a personal tutor is changing how we engage with experts. Students aren’t leveraging experienced mentors because they can ask AI without sounding foolish. Experts aren’t sharing their knowledge because they can use AI to produce something better and faster.

Favorite Quotes

Humans are far from obsolete, at least for now.

- Page 212

Personal Thoughts

How this book changed my perspective

I really appreciated Mollick’s perspective on “bringing AI to the table”. This is something that I want to get better at, but Mollick framed this up nicely in a way that I hadn’t thought about in the past. He suggests one of the best ways we can use AI is to find the “edges” of our problem space. This makes a lot of sense because AI will think about our problem in a way we haven’t. Using AI to help us do things like: finding extreme edge cases, test assumptions, uncover hidden variables, understand interdependencies, and identify constraints.

Another thing AI can be really effective with is helping us understand our hidden biases. The challenge here is that you need to combine this with Mollick’s third rule of AI: “treat AI as a person”. Creating multiple different prompts that behave in ways that are drastically different than yourself is critical. Without this piece, you’re only confirming your own biases or the natural biases that are baked into the model.

This insight has made me reconsider how I approach problem-solving in team settings. I now see the potential of using AI as a sort of ‘neutral party’ in brainstorming sessions, helping to surface ideas or concerns that team members might be hesitant to voice due to groupthink or hierarchy. It’s like having an impartial facilitator that can challenge our assumptions and push us to think beyond our usual boundaries.

Practical applications

I intend to create a few different prompts that will help me in my pursuits of using AI in these news ways. One of defining the problem and getting out a structured document that outlines the information I want help with in defining the edges of the problem space as well as a few different prompts for alternative personas based upon whatever the problem is.

My first cut at a prompt like this:

You are an AI assistant with expertise in [DOMAIN]. I'm working on a project to [BRIEF DESCRIPTION OF PROJECT/PROBLEM]. Please help me explore this problem from multiple angles:

Challenges:
Identify potential challenges from the perspectives of [STAKEHOLDER 1], [STAKEHOLDER 2], [STAKEHOLDER 3], and [STAKEHOLDER 4].
Unexpected Benefits:
Suggest three unexpected benefits that could arise from this project.
Extreme Scenarios:
Describe two extreme scenarios:
a) One where the project is wildly successful. What factors lead to this outcome?
b) One where the project fails dramatically. What factors lead to this outcome?
Hidden Variables:
What are some hidden variables or interdependencies I might be overlooking?
Differential Impact:
How might this project impact different [RELEVANT GROUPS, e.g., socioeconomic, age, geographic] groups differently?
Innovative Solutions:
Propose two innovative approaches to address the main challenges identified.
Future Implications:
Discuss potential long-term consequences (both positive and negative) of implementing this project.

Please structure your response using these categories, providing clear and concise points under each.

Questions for further exploration

  • How can I best define what use cases LLMs are helpful with and which it isn’t suited for? It’s a challenging area to understand what it’s good at and what is best completed by a human.
  • What are some of the edges of the climate change problem space discovered by AI that we haven’t thought or are under considered?

Last updated: 2024-07-28