Co-Intelligence, Part 2
Ethan Mollick's Book Summarized
As mentioned in part one of this book review, there is no better resource on LLMs in 2024 than Ethan Mollick’s book, Co-intelligence. This is a practical summary of the second half of Co-Intelligence—Chapters Six through Nine—the half of the book more applied to the issues of education and work. Where the advice appears truncated and in need of elaboration, the fault is my own; please pick up the book for all that couldn’t be contained here.
Context for AI and Education
In education, Mollick describes how AI is reviving the historically important, unbeatable, underappreciated tutor-student dynamic. In the shadow of tutoring, he points at the productive role of homework and tests in learning, showing that cheating has been eroding both homework and tests since the emergence of the internet. Since the wide availability of LLMs, that gradual erosion has exploded. Mollick compares the resulting reactions to AI from educators to the introduction of calculators in the late 20th century, suggesting that AI will eventually be completely integrated into a new kind of learning—except in quarantined core critical thinking and essay classes.
In that light, as schools evolve with technology, there’s a need for new curricula that focus (for now) on AI literacy and basic AI prompting, but that also continue to foster broad-based knowledge to help individuals be the “human in the loop” for AI. While AI can offer personalized tutoring, encouragement, and instruction, its (current) inability to grasp complex concepts or avoid errors means that human oversight remains essential.
Context for AI and Work
For work, Mollick first draws on historical examples to show the patterns in how technology has always simplified labor. He discusses the necessary evolution of job tasks to fit new technologies, highlighting how 19th century organizational charts, 20th century assembly lines and 21st century agile software development reflected contingent and persistent human adaptation to certain new technological circumstances. Yet, humans within their social systems are more resistant to change than job tasks, and so AI’s effect on jobs themselves may be limited in the next few years.
That said, even without job displacement, AI is still set to transform the workplace, from highly compensated and creative roles to more physical jobs, fundamentally reshaping tasks, systems, and (eventually) professions. Studies suggest that AI-enhanced work is faster, more creative, and more analytical, following a spectrum from strategic collaboration (like a “centaur”) to full integration (like a “cyborg”).
While AI might level the playing field in some professions, such as law, by boosting the abilities of all practitioners, the ultimate impact will depend on how we adapt, combining human oversight with AI's capabilities to achieve remarkable possibilities.
Future Possibilities and Limitations
Looking ahead in Chapter Nine, Mollick describes a future filled with some uncertainty but largely optimistic potential. Broadly, once AI is deeply integrated in education, it may close societal and global gaps. But without careful planning, its impact on education could also disrupt the development of the next generation’s expertise, hobbling our ability to create new professionals who can benefit from AI-powered productivity.
Mollick stresses the need for experts to remain involved and warns against underestimating AI's transformative potential. AI promises, at least, steady, linear advancement; but even if AI's progress slows, the innovations we have today will likely continue to reshape society and jobs. On the other side, we may be facing a continuing exponential increase for the foreseeable future of ever faster scientific discovery and enhanced productivity.
There is a balance to be struck between using AI for its strengths and avoiding a dystopian future where algorithmic control dominates the workplace (while absurdly and effectively coaching each of us on how to ‘be a better you’). While the fears of the AI-powered dystopias, "machine gods", or other sci-fi dystopias are entirely theoretical, even a one percent risk deserves some political support to mitigate catastrophes. However, the vast majority of us should focus on realistic growth and the transformative power of AI as a partner in our personal and professional lives, staying mindful of its limitations and maximizing its benefits.
AI Integration and Usage
The bottom line is that you and your organization should be integrating AI into everyday work and tutelage right now, mindful of both the opportunities and pitfalls. With deliberate integration, AI offers rapid adoption benefits. On the other, it’s important not to become overly reliant or passive and concerns over privacy, legal issues. For organizations job security prompts users to keep their AI usage discreet, unless encouraged and rewarded for sharing the productiviy gains. Questions about ethics, such as what constitutes cheating in an AI-driven workplace or classroom, need careful consideration.
These imperatives are the core of the book’s usefulness. Through obvious familiarity with LLMs, Mollick has created a refreshing exception to the standard LLM manuals. To add to the first set of instructions from Chapters Three through Five, here are Mollick’s imperatives from the second half of the book, mostly Chapters Six through Eight:
Avoid AI on Human-Centric Tasks
Continuously explore using AI to find the tasks it currently does poorly so that those tasks can be AI-enhanced with caution or completely avoided.
Consider deeply the tasks you always want humans to do, such as personal decisions, ethical considerations, raising children, expressing values, and developing skills and thought.
Find ways to be educated in AI literacy, including the potential downsides, biases, and unethical uses of AI; for example, consider future copyright policies around AI-generated output.
Future-proof yourself by becoming the “human in the loop” by focusing on building and honing your own narrow-focused expertise that will be the kernel of human responsibility within AI-enhanced tasks.
Personal Delegated and Automated AI Tasks (Centaur Tasks)
Gradually transition to AI usage: start by inviting AI into all your tasks, but work as a centaur with discrete separation and delegation of tasks to the AI.
Identify those tedious, mentally dangerous, repetitive, boring, or time-consuming tasks (which may nevertheless also be complex or sophisticated tasks) for potential AI automation.
Make those tedious, mentally dangerous, and repetitive tasks “AI friendly” such as is done for spam filters and high-frequency trading.
Implement AI for “AI friendly” tasks in automated ways, with an eye toward even more automation potential in the future.
As typically ripe examples for automation, employ AI for note dictation, summarization, and process organization.
Personal AI Hybrid Collaboration (Cyborg Tasks)
Eventually integrate AI deeply into your day as a cyborg on tasks in which you know well how the AI responds, repeating the same process—from centaur to cyborg—with the next new tasks.
Break the autocomplete pattern in AI prompting by providing clear context and constraints.
Develop more advanced prompting skills by looking through examples of other ways to explain what you want, such as chains of thought.
As typically ripe examples for collaboration, use dialogue with AI for writing support, or to provide critiques and alternative viewpoints on your written products or ideas.
Organizational AI Integration
Get the help of the most advanced AI users in your organization to share strategies, successes, and to encourage more workers to use AI.
Involve all workers in AI adoption, recognizing that AI skills have never been a hiring criterion, so anyone may surprisingly possess them.
Remove any stigma around surfacing AI productivity gains: organize the entire workforce around sharing AI processes; reward workers commensurably when AI productivity gains are uncovered, and publicly renounce the possibility of layoffs due to AI.
Educational AI Integration
Disallow AI in classes focused on writing and essay skills; revert to 19th century handwritten essays and blue books for these skills. Seriously.
Make AI mandatory in educational tasks but hold students accountable for the outcomes and facts produced by AI, mirroring workplace standards.
Encourage students to “cheat” by having AI generate essays, and then having the students critically evaluating the AI-generated content.
Have students conduct interviews with AI before speaking to real human experts or authorities to prepare effectively.
Require AI use but make tasks “ambitious to the point of impossible; you are going to be using AI. Can’t code? Definitely plan on making a working app. Does it involve a website? You should commit to creating a prototype working site, with all-original images and text. I won’t penalize you for failing if you are too ambitious”
Use AI for at-home tutoring and save time for collective “active learning” in the classrooms.
Lean into building human expertise with a structured knowledge base and a deliberate practice plan, using AI to gradually increase difficulty and feedback.
Implement a simulated learning environment (e.g., Mollick’s Wharton Simulator) to tutor students through instruction, output creation, critical feedback, analysis, and rewarding encouragement.
These steps provide a framework for effectively integrating LLMs into various aspects of work and education, ensuring ethical usage, maximizing productivity, and fostering the essential kernel of human expertise that will carry us forward through an uncertain future.