[anonmeow]

With modern ML techniques it looks feasible to recreate at least online behavior of an individual - facebook likes, comments etc. The datasets are here, in facebook/google datcenters, and DL models are already used to model conversation. It would be interesting to know just how many megabytes of logs of your online activities is really necessary to extrapolate your behavior into the future.

Facebook AI research is probably playing with such models right now.

[astazangasta]

Eh? Humans are not fixed entities, they adapt and learn. You can't extrapolate into the future based on my previous anything. Six-year-old me can't tell you who thirty-six year-old me is.

[anonmeow]

Given enough data and a good model (recurrent neural network can model arbitrary algorithms, for example) you can learn algorithmic regularities in data, including learning itself. There is a paper just about that: http://link.springer.com/chapter/10.1007%2F3-540-44668-0_13#...

If this can be scaled to human-like learning remains to be seen, but training a conversation RNN model that remembers some details of past conversations and acts on them should be possible.

[pygy_]

There' a Black Mirror episode built around that very idea.