Consider the following: Google has a collection of computers with inputs. One of those inputs can be accessed via the Google web search interface. When you search a term using the Google web interface, you are requesting control over a very small part of the collection of computers at Google in order to complete your request.
When you do this, you do not control all of Google, but you are exercising C2 over a tiny portion of it.
Now consider that Google was built on technologies that can precisely reject requests for control per request based on certain conditions. Google can reject requests based on your IP, your personal identity, the contents of the request, etc.
The human brain does not have facilities for doing that. It did not evolve with the expectation that it would be valuable to reject neurological patterns that behave like sensory inputs. An example of this would be like taking a psychoactive drug, and then trying to think it's effects away. You can't do it.
The difference between a psychoactive drug and a brain to computer interface is that one of them can dynamically impose specific neurological phenomena (what you call inputs) based on the intent of someone else, and the other one will land you in jail for possession.
What does "bidirectional communication" mean if not read and write? What other directions are there?