I once, probably 4-6 years ago, used exports from Slack conversations to train a Markov Chain to recreate a user that was around a lot and then left for a while. I wrote the whole thing in python and wasn't overly well-versed in the statistics and math side but understood the principle. I made a bot and had it join the Slack instance that I administrate and it would interact if you tagged it or if you said things that person always responded to (hardcoded).
Well, the responses were pretty messed up and not accurate but we all got a good chuckle watching the bot sometimes actually sound like the person amidst a mass of random other things that person always said jumbled together :D
I had a similar program designed as my "AWAY" bot that was trained on transcripts of my previous conversations and connected to Skype. At the time (2009) I was living in Taiwan so I would activate it when I went to bed to chat with my friends back in the States given the ~12 hour time difference. Reading back some of the transcripts made it sound like I was on the verge of a psychotic break though.
I'm sure that Gilfoyle's prank in Silicon Valley had its basis in a true story, like most of SV's episodes. i.e. I'm sure this has been done quite a few times ever since Mark V. Shaney.
Well, the responses were pretty messed up and not accurate but we all got a good chuckle watching the bot sometimes actually sound like the person amidst a mass of random other things that person always said jumbled together :D