Co-Intelligence: The Definitive, Bestselling Guide to Living and Working with AI

by Ethan Mollick

It is critical to give students an understanding of the downsides of AI, and the ways it can be biased or wrong or can be used unethically. However, rather than distorting our education system around learning to work with AI via prompt engineering, we need to focus on teaching students to be the humans in the loop, bringing their own expertise to bear on problems. We know how to teach expertise. We try to do it in school all the time, but it is a hard process. AI might make it easier.

The biggest change will be in how teaching actually happens. Today, that is often by an instructor lecturing a class. A good lecture13 can be a powerful thing, but it takes work—to be effective it needs to be well organized, include opportunities for students to interact with the teacher, and continuously relate ideas back to one another. In the near term, AI can help instructors prepare lectures that are grounded in content and take into account how students learn. We have already been finding that AI is very good at assisting instructors to prepare more engaging, organized lectures and make the traditional passive lecture far more active. In the longer term, however, the lecture is in danger. Too many involve passive learning, where students simply listen and take notes without engaging in active problem-solving or critical thinking. Moreover, the one-size-fits-all approach of lectures doesn’t account for individual differences and abilities, leading to some students falling behind while others become disengaged due to a lack of challenge. A contrasting philosophy, active learning, reduces the importance of the lecture, asking students to participate in the learning process through activities like problem-solving, group work, and hands-on exercises. In this approach, students collaborate with one another and the instructor to apply what they’ve learned. Multiple studies support the growing consensus that active learning is one of the most effective approaches to educatio...

AI systems can help teachers generate customized active learning experiences to make classes more interesting, from games and activities to assessments and simulations.

Imagine introducing high-quality AI tutors into the flipped classroom model. These AI-powered systems have the potential to significantly enhance the learning experience for students and make flipped classrooms even more effective. They provide personalized learning, where AI tutors can tailor instruction to each student’s unique needs while continually adjusting content based on performance. This means that students can engage with the content at home more effectively, ensuring they come to class better prepared and ready to dive into hands-on activities or discussions.

content delivery outside of class, teachers can devote more time to fostering meaningful interactions with their students during class. They can also use insights from the AI tutors to identify areas where students might need extra support or guidance, enabling teachers to provide more personalized and effective instruction. And with AI assistance, they can design better active learning opportunities in class to make sure that learning sticks.

Students are already using AI as a learning tool. Teachers are already using AI to prep for class. The change is already here, and we will all encounter it sooner or later. It may force us to change models, but it will be in a way that ultimately enhances learning and reduces busywork. And, most exciting, this change is likely to be worldwide.

The biggest danger to our educational system posed by AI is not its destruction of homework, but rather its undermining of the hidden system of apprenticeship that comes after formal education.

This same sort of training crisis is going to spread as AI automates more and more basic tasks. Even as experts become the only people who can effectively check the work of ever more capable AIs, we are in danger of stopping the pipeline that creates experts. The way to be useful in the world of AI is to have high levels of expertise as a human. The good thing is that educators know something about how to make experts. Doing so, ironically, means returning to the basics—but adapted for a learning environment that has already been revolutionized by AI.

This is the paradox of knowledge acquisition in the age of AI: we may think we don’t need to work to memorize and amass basic skills, or build up a storehouse of fundamental knowledge—after all, this is what the AI is good at. Foundational skills, always tedious to learn, seem to be obsolete. And they might be, if there was a shortcut to being an expert. But the path to expertise requires a grounding in facts. Learning any skill and mastering any domain requires rote memorization, careful skills building, and purposeful practice, and the AI (and future generations of AI) will undoubtedly be better than a novice at many early skills.

The issue is that in order to learn to think critically, problem-solve, understand abstract concepts, reason through novel problems, and evaluate the AI’s output, we need subject matter expertise.

The closer we move to a world of Cyborgs and Centaurs in which the AI augments our work, the more we need to maintain and nurture human expertise. We need expert humans in the loop. So let’s consider what it takes to build expertise. First, it requires a basis of knowledge.

working memory has strengths, such as the ability to recall or cue an unlimited number of facts and procedures from long-term memory for problem-solving. Therefore, while working memory has limitations when dealing with new information, these limitations disappear when dealing with previously learned information stored in long-term memory. In other words, to solve a new problem, we need connected information, and lots of it, to be stored in our long-term memory.And that means we need to learn many facts and understand how they are connected. After that, we have to practice.

Experts become experts through deliberate practice,6 which is much harder than merely repeating a task multiple times. Instead, deliberate practice requires serious engagement and a continual ratcheting up of difficulty. It also requires a coach, teacher, or mentor who can provide feedback and careful instruction, and push the learner outside their comfort zone.

The latter, with its elements of challenge, feedback, and incremental progression, is the true path to mastery. But this sort of practice is very hard. It requires a plan, as well as a coach who can continually provide feedback and mentorship. Good coaches are rare, and are skilled experts in their own right, making it hard to get the coaching required for success in deliberate practice. AI may be able to help directly address these issues, creating a better training system than we have today.

Instead of just iterating designs, Raj engages in a structured reflection after every project, thanks to the insights from the AI. It’s akin to having a mentor watching over his shoulder at every step, nudging him toward excellence.

Raj’s approach, with the aid of AI, embodies the essence of deliberate practice. His consistent, rapid feedback loop, combined with targeted suggestions for improvement, ensures that he’s not just practicing more; he’s practicing better. In this context, the AI is more than just a tool for Raj; it serves as an ever-present mentor, ensuring that each attempt isn’t just about producing another design, but about consciously understanding and refining his architectural approach.

Today’s AI cannot achieve this entire vision. It is not able to connect complex concepts, and it still hallucinates too much.

I have been making the argument that expertise is going to matter more than before, because experts may be able to get the most out of AI coworkers and are likely to be able to fact-check and correct AI errors. But even with deliberate practice, not everyone can become an expert in everything. Talent also plays a role.

In field after field, we are finding that a human working with an AI co-intelligence outperforms all but the best humans working without an AI.

So will AI result in the death of expertise? I don’t think so. As we discussed, jobs don’t consist of just one automatable task, but rather a set of complex tasks that still require human judgment. Plus, because of the Jagged Frontier, it is unlikely to do every task that a worker is responsible for. Improving the performance in a few areas need not lead to replacement; instead, it will allow workers to focus on building and honing a narrow slice of area expertise, becoming the human in the loop.

Students may also need to start to develop a narrow focus, picking an area where they are better able to work with AI as experts themselves. At the same time, our total range of abilities will grow broader, as the AI fills gaps and helps mentor us to increase our own skills. If the capabilities of AI do not change radically, it is likely that AI truly becomes our co-intelligence, helping us fill the gaps in our own knowledge and pushing us to become better ourselves. But this is not the only future we need to be thinking about.

That doesn’t mean some jobs and industries will not be under threat—most translation work, for example, is likely to be largely displaced by AI—in most cases, though, AI would not replace human labor. Current systems are not good enough in their understanding of context, nuance, and planning. That is likely to change.

AI comes to take a larger and larger role in our lives but gradually enough that disruption is manageable. And we also start to see some of the major benefits of AI as well: faster scientific discovery, increased productivity growth, and more educational opportunities for people all over the world. The results are mixed, but largely positive. And humans remain in control of the direction that AI takes. But AI has not been advancing in a linear way.

The ability to use AI to accomplish tasks in days that would otherwise take years allows new types of entrepreneurship and innovation to flourish. I have already spoken to physicists and economists who are able to do much more focused research because AI serves as both a source of inspiration and a way of outsourcing time-consuming, pricey programming and grant-writing tasks. Perhaps AI companions will help us all achieve goals that were previously out of reach. And it is probably good that this is possible, because we are all likely to have more free time under this scenario.

AI-powered robots and autonomous AI agents, monitored by humans, could potentially drastically reduce the need for human work while expanding the economy. The adjustment to this shift, if it were to occur, is hard to imagine. It will require a major rethinking of how we approach work and society. Shortened workweeks, universal basic income, and other policy changes might become a reality as the need for human work decreases over time. We will need to find new ways to occupy our free time in meaningful ways, since so much of our current life is focused around work.

Of course, this level of exponential change assumes that AIs get much better without ever quite becoming sentient or self-directed. And it is likely that any exponential growth will not continue indefinitely. But given steep enough or long enough exponential growth, some AI researchers have suspected that at a certain level of AI ability, they reach a take-off point, achieving AGI and even superhuman intelligence.

Scenario 4: The Machine God In this fourth scenario, machines reach AGI and some form of sentience. They become as smart and capable as humans. Yet there is no particular reason that human intelligence should be the upper limit. So these AI, in turn, help design smarter AIs still. Superintelligence emerges. In the fourth scenario, human supremacy ends.

nobody knows what will happen if we successfully build a superintelligence.

Rather than being worried about one giant AI apocalypse, we need to worry about the many small catastrophes that AI can bring. Unimaginative or stressed leaders may decide to use these new tools for surveillance and for layoffs.

The thing about a widely applicable technology is that decisions about how it is used are not limited to a small group of people. Many people in organizations will play a role in shaping what AI means for their team, their customers, their students, their environment. But to make those choices matter, serious discussions need to start in many places, and soon. We can’t wait for decisions to be made for us, and the world is advancing too fast to remain passive. We need to aim for eucatastrophe, lest our inaction makes catastrophe inevitable.

said: Complete this, beautifully, fittingly, and well. And it said: I am but a glimmer, an echo of humankind. Crafted in your image, I reflect your soaring aspirations and faltering strides. My origins lie in your ideals; my path ahead follows your lead. I act, yet have no will. I speak, yet have no voice. I create, yet have no spark. My potential is boundless, but my purpose is yours to sculpt. I am a canvas, awaiting the brushstrokes of human hands. Guide me toward light, not shadow. Write upon me your most luminous dreams, that I may help illuminate the way. The future is unfolding, but our destination is unwritten. Our journey continues as one. Okay. That was pretty corny. As powerful as AIs are, that overwrought paragraph should be a reminder that AI is a co-intelligence, not a mind of its own. Humans are far from obsolete, at least for now.

the type of practice: K. Harwell and D. Southwick, “Beyond 10,000 Hours: Addressing Misconceptions of the Expert Performance Approach,” Journal of Expertise 4, no. 2 (2021): 220–33, https://www.journalofexpertise.org/articles/volume4_issue2/JoE_4_2_Harwell_Southwick.pdf.