Is There a Place for a System for Edge Cases?

In the next five years, unless there are a very, very specific set of use cases that have been premeditated, my guess is that won't be feasible. Even in scenar s where you spend a lot of time cleaning oup the data, training models off line, to actually evaluate whether those models are safe, working uneffective, takes a lot of energy. So the fact that on the fly a system culd automaticallya learn something and make it available to a provider and it would just be safe, effective, is very, very unlikely.