Here's a quick thought about this article on Why Google Wants to ‘Can’ Your Messages. Apparently, Google has filed a patent for a service that would offer suggested text for responses to text messages, emails and social media interactions. As the author points out this is more of an incremental step from existing scripts and tools that can do this for you. But the difference here is you would not need to design any rules for the responses. Google would combine all the information they have about how you respond to communications, down to the way you put words together, to provide the prompted text.
The author goes into various pros and cons of this and is generally in favour of it. There isn't any technical description of how it might work so you'd have to read the patent more closely.
This line from the article really is worth repeating though:
Plus, they’re (Google) really good at algorithms.Um, yep. They are indeed good at...algorithms.
I imagine they're planning to use the same technology powering their Google Now product and the Deep Learning methods build into their new image search in Google+ images. They can learn a model of series of likely words that occur in your responses to previous messages, run it against grammar models and tie it in to your GPS location, your calendar etc.
The Real StoryThe first thing I thought of when I started reading this was that a successful service like this could be a huge problem for Google's competitors. Google could provide a service that automatically clicks 'like' on Facebook and wishes your friends happy birthday or more. The they could have a service that automatically retweets or replies on Twitter. If they did that then those services would lose one of the only things they really have, information about their users' actual behaviour.
This is a ways off yet, but if a significant portion of the behaviour of normal users on Facebook, for example, was not from those users but from a Google bot, then all the analytics and data mining Facebook would actually be trying to learn Google's model of a user rather than the behaviour of that user itself. It's brilliant actually if it works, Google would be inserting themselves into the datastream as a service to their users, meanwhile they'd gain more and more control over the incoming datastream to the competitors.
Of course, Facebook would respond with their own bots. I can see it now, an ever escalating arms race of intelligent auto-reply agents between Facebook, Twitter, Google, Microsoft, etc.
So enjoy the current era of natural, eloquent human interaction on the Internet while it lasts.