AI, education, and humanity

The past two centuries have seen a massive expansion in education. Enrollment rates in all levels of education have grown substantially, along with accompanying growth in literacy and numeracy rates. Greater education is associated with individual human flourishing, in terms of economic success, health and well-being. Quality education is also beneficial to our societal-level economic, civic, and social outcomes.

Yet the path to quality, universal education is challenging.

• How do we enable equality of opportunity in education and close the digital divide gap?
• Are there enough qualified teachers to meet the needs of a growing student population?
• How do we give teachers time to do the most important parts of their job?
• How can we help students at different levels make progress?

New technology, steered properly, can help answer these questions.

We know a lot about how people learn and can use this knowledge to guide how we use AI technologies. For example, cognitively engaging activities lead to greater learning. Micki Chi and colleagues proposed a framework modeling how learning increases as engagement moves from passive to active to constructive to interactive experiences:

Interactive learning occurs when two individuals both make constructive contributions to a conversation around content to be learned. That engagement leads to greater learning because, as hypothesized by the model, interacting with another person deepens the level of cognitive engagement, as the two individuals question and respond to each other’s offerings and push to extend their mutual understanding of the content. This interaction can take different forms, from debate to collaborative problem solving to co-authoring a piece of writing.

Globally scaling the numbers of tutors and teachers required to have these constructive, interactive learning experiences is impossible. For decades, attempts to harness technology to create similar experiences have failed. The available technology did not enable these types of interactions. Instead, we settled for branching dialogue trees where students choose among canned responses.

New AI technologies, particularly large language models, may help more students engage in interactive learning. A student trying to solve a math problem would be able understand what step they’re on, whether they have skipped steps, and be asked to explain their rationale and next steps. The AI-powered tutor can also engage either side of a debate on any topic and then offer a summary of each sides’ points with feedback on the argumentation skills of the debater. It can take on the persona of a literary character and discuss their motivations with a learner. These language capabilities are an opportunity to significantly improve learning outcomes.

Granted, an AI-powered tutor won’t solve all learning problems. Part of the power of interaction is in the relationship with other people. Knowing an adult cares for your progress is a known predictor of learner success. Strong relationships with teachers and school staff increase motivation and academic engagement. An AI tutor cannot replace good teachers or the care of a human relationship. But it can provide support for all those learners without access or with limited access to qualified teachers at school.

We must ensure that these technologies do not deepen the digital divide and instead develop ways to get them in the hands of learners who need them the most. The current costs of using the large language models present challenges we must address to support teachers and learners in historically under resourced communities.

If we can guide our new AI to act in ways that we know facilitate learning, and if we can get these tools to learners who otherwise lack opportunity, we have an opportunity to better enable individuals and society as a whole to flourish.