AI is learning how to teach: Will bots replace human teachers?

Artificial intelligence is transforming learning, writes Carl Hendrick. For the good? Not necessarily. "The question is whether that transformation will make us smarter or render us dependent on machines to do our thinking for us."

Soon, AI may prove to be "more effective at producing learning," he writes. "The optimistic vision is compelling: every child receives expert, tireless, infinitely patient instruction calibrated precisely to their needs," narrowing achievement gaps. "Teachers are freed from the grinding mechanics of delivery and assessment to focus on what machines genuinely cannot do: motivation, relationship, the human dimensions of learning."

But he worries: "If instruction is algorithmic," will human teachers be freed for higher-order tasks? Or just replaced by cheap, efficient bots?

"The story of EdTech has largely been one of expensive failure," Hendrick writes. There's evidence that AI tutoring can be highly effective, if structured to supplement human teaching. However, "many uses of AI in education actively harm learning." Research confirms what teachers suspect: "When AI does the thinking, students stop doing it themselves."

Younger AI users show the greatest decline in critical thinking ability, a recent study showed. Students may think they're understanding, but they're mistaken.

ChatGPT will happily write your essay, solve your equation, explain the concept you should be puzzling through yourself. It is designed to be helpful, not to promote learning, and those are fundamentally different objectives.

To promote learning, AI must be designed "to scaffold rather than solve, to prompt retrieval rather than provide answers, to increase rather than eliminate cognitive load at the right moments." It must be trained to be "strategically unhelpful."

AI tutors have the potential to produce faster learning, Hendrick predicts. They will be "better at applying the known laws of learning, ones we have known for 100 years; explicit instruction, timely feedback, discriminating between varied examples, adaptive pacing, retrieval practice spaced out, and integrating new knowledge with old."

However, "fast learning is not always deep learning," he writes. "Efficiency is not understanding."

Michael Horn and Diane Tavenner talk about AI in education with Anthropic's Neerav Kingsland on "Class Disrupted."