I am thinking of implementing something when I get the time. It will be a text editor that collaborates with you.
You can type stuff into it, just like an ordinary text editor. But on the other side of the screen, as it were, a software robot is always running.
The robot looks for certain things you might type, and responds in smart ways. For example, you might type “make a table of this year’s Google monthly stock price.” In response, your robot collaborator will replace the text you just typed by the appropriate table.
A text editor with that sort of capability might be useful in many fields, including art, science, music, engineering and the humanities. The possibilities are endless.
Now I just need to find the time to implement the darned thing. If only I had a really smart text editor to help me out…
For some reason this reminded me of:
The product I work on has a bug database, and a source repository going back at least a decade, probably more. There are many thousands of bugs in the bug database, along with the source code modifications fixing them in the source repo.
Clearly, all of this data – the reported bugs, and the modifications fixing them – could be used as a training set. Once trained with this data, it should be possible for a system to automatically fix any future bugs.
Should work, right? What could go wrong?
Robin Sloan made something similar to what you are describing:
https://www.robinsloan.com/notes/writing-with-the-machine/
Robin Sloan’s editor is very cool!
I think, though, that I want to go in the opposite direction from deep learning. Rather than an uncanny collaborator, I want to build something that does various tasks for me in a predictable manner.