Business

Value we're providing / how to convince people to allow model to train on their facebook data:
1) Time saved - users can have a personalized bot trained on their specific chat history, which could then br their personal assistant 
2) Entertaining - users play a game to predict whether they are chatting to the bot or their friend
3) Potential $ - the most successful gamers receive awards based on current standings

Business model:
Users and bot are playing for tokens.
User stakes how many tokens it wants to bet in any time period (token accrue continuously)
"Winning" means they correctly identify whether its the bot or the human, and if so they win.

Dialogue/interface might go something like this:
- User [Henry = human] opens Facebook messenger / Open YC-ChatBot
- Start chatting = playing the game

[INITIATE GAME]
- Bot: Hi Henry, you are chatting with [Friend's name] (Or are you?!)           [note: this could be the AI or the actual friend]
- Bot: You're token balance is 1000, how many token's would you like to bet?
- Henry: 10
- Bot: Chat away 
- Henry: Hey man, what did you do yesterday?
- Bot: I went swimming. You?
[END]
- Bot: Was that me or my AI alter-ego?
- Henry: You
- Bot: Correct! You just won 20 tokens and your balance is now 1,020 tokens!

[NEW GAME]
 - Bot: You're now chatting with [different friends name]
 ......
 .....
 
 Incentives:
 - Users are incentivised to continue playing the game = improving the AI
 - If they want they can just let the AI take over, at the risk of the token balance falling