[thatcantbeit]

While some data science tools may be `excel-implementable` (or some equivalent) in five to ten years, there's significant risk of mis-implementation of results given how much of a mystery many methods are to the people who'd use them.

I'd compare data scientists more to CPAs. You can have software like TurboTax and Quickbooks, but CPAs don't seem to be going anywhere. Similarly, anything that's more complicated than cookie-cutter data analysis will require someone who knows how to develop, build, and debug the algorithms themselves.

Use cases of data science are often too specific for most data science to turn into button pushers. Look at the vast array of ways a neural network can be implemented. Which one of those implementations will be in the `excel package`?

[stult]

I think the counterargument is that there will be some standardization of these methodologies such that they don't require experts to implement. Similar to how it used to require a programmer to develop any webpage at all, but now most common use cases are covered by simple off the shelf solutions like square space or WordPress. Moreover, the skills will disseminate quickly to other fields. Originally only accountants used Excel and now it's an indispensable skill for most professional jobs. So worker bees of the future will know the differences between various types of NN just like the modern worker bee can write vlookups or create charts in Excel even though that would have been a mysterious skill thirty years ago. Meaning your theoretical Excel package would cover many different types of NN, with users expected to understand the difference.