AI model will change how we learn and how we work

In 2016, we ran a telemedicine eye exam company, and we tried to use AI to predict what eyewear lens power a person will need. The assumption back then is that lens power is the most straightforward mathematical problem AI can solve. If you wear prescription glasses, you will notice the prescriptions are just a bunch of numbers. Those numbers represent the sphere (nearsighted or farsighted), cylinder (astigmatism) and axis. Mathematically, there are 6 million different combinations to make a pair of lenses.

This is a real world problem tailor made for AI to solve, pure math. Back in 2016, we had a team of vision scientists, computer engineers, and imaging scientists. We spent 6 months trying to solve the problem. 90% of our effort was spent on manually annotating data to feed to the AI model. We end up with a model with a hundred patient data. It was a promising start, but it never achieved any clinical relevance back then. It takes too much work. In 2023, we tried again. It took a few days to train the AI model with a few thousands de-identified patient data. The model can predict lens prescription with 0.98 correlation to real world results. Granted, more clinical validation needs to be done to make any health related prediction. I am blown away by the quantum leap on the current AI stack. The current AI foundation technology will change how we learn and how we do our work. I have seen it.