AI in Education
Education is where I am most optimistic and most nervous at the same time. The promise is obvious: a tutor for every kid, feedback in seconds instead of a week, material that adapts to how someone actually learns. If that worked as advertised, it would be one of the best things AI ever did. The if is doing a lot of work in that sentence.
Take intelligent tutoring systems. They watch how a student performs, spot where they are stuck, and serve up the right explanation or practice problem. Useful, clearly. But the model learned what stuck looks like from someone's data, and that data has a shape – certain learning styles, certain backgrounds. Then there is the question of what gets lost when feedback is instant and automated: the slow, sometimes frustrating work of figuring something out yourself. And all of this is happening to minors, which makes privacy non-negotiable rather than a nice-to-have.
The technology covers a lot of ground – adaptive platforms, generated content, automated assessment, and the administrative work that eats teachers' evenings. The harder problem is deciding what we are optimizing for. Test scores are easy to measure. Curiosity, creativity, and the ability to work with other people are not, and a system tuned for the first will quietly trade away the rest.
Getting this right takes more people than usually get asked. The developers building the tools, the teachers who will be told to use them, cognitive scientists, parents, and the students themselves, who tend to have sharp opinions once someone bothers to ask. Otherwise the tools end up technically impressive and pedagogically hollow, and the schools that can afford them pull further ahead of the ones that cannot.
The version of this I want is simple to state. Teachers freed up to do the parts of the job that need a human. Students who get help matched to where they are. A system that widens access instead of adding another gap. It is achievable, but only if the people building it treat human-centered as a design constraint rather than a line in the pitch deck.