(Illustration by Chad Hagen)
In 1950, Alan Turing posed a bold question: Can machines think? That inquiry sparked the field of artificial intelligence and ignited a powerful vision—along with deep concerns—that machines might someday replicate or even surpass human intelligence.
Today, signs of AI’s presence in education are multiplying: personalized tutors, adaptive assessments, predictive dashboards, automated grading. But before we ask how AI can improve education, we must first ask: What are we optimizing for?
Too often, the answer is efficiency.
This tendency reflects what Stanford University economist Erik Brynjolfsson calls the Turing Trap—according to which, AI systems are employed to mimic existing human tasks rather than to unlock fundamentally new possibilities. We can, of course, obtain real benefits from automating repetitive labor—a Roomba liberates time. But careless deployment of AI risks automating the thinking of an ineffective past, entrenching ways of educating that no longer serve us.
In education, that risk is acute. A recent study found that AI-generated lesson plans tend to default to didactic instruction—one teacher at the front of the room, many students passively receiving information. That result is not surprising: Didactic instruction is the most prevalent format on the internet and in our schooling system. The architecture of modern schooling was shaped by the norms of the industrial age: standardization, scale, control. That model enabled widespread access to education, but it also limited how we think about learning.
Now that universal schooling is largely in place, we should prioritize learning in all its richness and complexity, including learning to care, connect, and flourish in a changing future. AI can help—but only if we avoid the Turing Trap.
Appreciation, Understanding, Adaptability
To begin, we must ask: What should people know and be able to do in an AI-infused future?
Often, this conversation starts with the question of what skills people will need. Skills, which are typically defined as sequential procedures, are precisely what AI excels at replicating. If we anchor learning in tasks that AI can eventually do better, we will not be giving children what they need to excel in the future. Instead, we need a more expansive view of human intelligence grounded in appreciation, understanding, and adaptability.
Appreciation of a sunset, a painting, or a clever idea cannot be outsourced. Asking AI to appreciate something for you will always ring hollow.
Understanding, too, demands more than surface-level information. In a seminal study, researcher Michelene Chi analyzed a biology textbook’s description of the human heart. It listed only a fraction of the system’s causal connections—too many would have required an unwieldy passage. Good readers fill in the connections. They rely on prior knowledge that enables them to go beyond the information given. AI may present well-structured responses, but without foundational knowledge, learners cannot construct a deep understanding of a text.
Adaptability, the capacity to change one’s approach in the face of new contexts, will be essential in this new era. While people will still need to know how to appreciate and understand, generative AI introduces the new challenge of navigating a rapidly evolving landscape of information and tools. When calculators first appeared in classrooms, teachers did not stop teaching math; they taught both math and the calculator. Today’s moment is similar—albeit far more complex. Consider how writing an essay has changed. A generation ago, students struggled to find three sources for their essays—one often an encyclopedia. Now, they confront a flood of content, much of it algorithmically amplified and often misleading. Stanford professor of education Sam Wineburg’s work on teaching students how to evaluate internet information sources to develop “fake news” literacy underscores the urgency of equipping students not just to consume information, but to discern, evaluate, and adapt in an era defined by information overload. His research demonstrates that even highly educated individuals can be misled by online misinformation and that strengthening critical evaluation skills not only fosters informed citizenship but also enhances students’ ability to learn effectively in a complex, rapidly changing information landscape.
Most instruction still prioritizes recitation over adaptation. Assessments isolate students from learning, offering little chance to revise or grow. But dynamic assessments, which incorporate opportunities to learn during the test, better predict a student’s capacity to adapt their knowledge to new contexts. AI enables a focus on the process and not just the product of learning.
Looking forward, we believe the AI-infused future will place a premium on learners’ ability to create—not just to consume. Creativity is a deep form of adaptability: It means trying new approaches, generating ideas, seeking feedback, and attending to constraints on reasonable solutions. Generative AI has already ushered in this shift. People delight in composing poems, remixing code, and designing digital art with AI’s help. We may be moving from the information age to the age of creation.
The Age of Creation
The rise of generative AI opens profound new possibilities for how we learn, what we value, and who gets to create. But seizing this opportunity requires applying the science of learning. For decades, most instructional models relied on inherited intuitions. But the past half-century has given rise to a robust body of research on how people learn. We face a once-in-a-century chance to redesign education around what truly works.
Can AI support the development of our social abilities? For instance, can it help us learn to belong, empathize, and care for others?
Fifty years of research have uncovered many important ingredients for learning and adapting. Among them is the repeated demonstration that learning depends on more than the transmission of information. It depends on interaction and feedback, the context of learning and one’s standing within it. Stanford professor of education Roy Pea’s work on distributed cognition highlights how learning the tools in one’s environment shapes the ability to think. This represents a major shift from the first studies of learning, which theorized that the only driver of learning and behavior was timely reinforcement through a reward or punishment.
A second major shift has been the inclusion of social factors in theories of learning. Relationships matter. Social exchange drives many forms of learning, whether as a motivator or as a prosocial model that children can emulate. Belonging is not an add-on—it is foundational. Stanford studies have shown that even brief interventions aimed at belonging lead to long-term academic gains, particularly for students from underrepresented groups. Given the significance of social interactions for learning and for life, learning environments should place a premium on relational intelligence: the ability to empathize, communicate, and thrive in community. AI could help, for example, by providing agents that support social interactions between humans. Or, AI could cause harm if used primarily as proliferating “nanny tools” that monitor digital behaviors, prioritizing surveillance over genuine connection, or as forms of illusory and emotionally manipulative companionship.
To thrive in this new era, we need pedagogies of creation and sharing. Project-based learning offers one example: Students collaborate to create a solution or product, receive rich feedback, reflect, and iterate. Generative AI can be a tremendous help to a teacher managing a classroom of different projects. AI can also help students surpass themselves—for example, by helping to create simulations that manifest their projects.
Importantly, creation-centered learning also helps develop abilities useful for adaptation, such as persistence, experimentation, goal-setting, feedback-seeking, and novel-solution generation. In this vision, students command the AI—they use it to augment their ideas and imagination. In the traditional efficiency model, AI commands the student—telling them what to do, how to think, and when to move on.
That distinction makes all the difference for flourishing in a dynamic future. In our vision, we can use AI to augment human abilities to learn through creation and relationships. This depends on building tools that support educator decision-making and instruction, elevate student agency, and scaffold pathways for creation and connection for all learners.
A Better Path
We see early signs of this better path. At Stanford, faculty and students are beginning to explore AI tinkering—using AI not as a black box for automation, but as a collaborative partner in learning. The Learning Through Creation seed grant, led by professor of education Victor Lee for the Stanford Accelerator for Learning and the Stanford Human-Centered Artificial Intelligence institute, supports projects where students use generative AI to make, play, remix, and invent—shifting the focus from passive consumption to active creation. These early ideas and prototypes are redefining what AI in education could be.
Humans are naturally adaptive and profoundly social, but these traits must be nurtured. Creating with AI has the potential to enhance our adaptability. But can AI also support the development of our social abilities? For instance, can it help us learn to belong, empathize, and care for others?
Current trends in AI companions often do the opposite—displacing the need to learn social skills by adapting to users’ preferences and emotions without challenging them to relate with others. A machine that makes a child feel understood may offer comfort, but it doesn’t teach the child how to understand others. If we are to design AI companions, let them be bridges—not destinations—for social learning. They must support the development of social approaches that transfer beyond the machine and into the real, human world.
We believe this future is within reach. But only if we move beyond the lure of efficiency-driven automation and instead design with intention—anchored in the science of learning, recognizing the complex differences among learners, and grounded in a deep belief in human potential.
Used wisely, generative AI can help us create an education system that is more creative, more connected, and more caring—one that belongs to every learner.
Read more stories by Isabelle C. Hau & Daniel L. Schwartz.