Thought-Experiment: If you read a clone of your own writing style, would you recognize it instantly, just as if you were looking in a mirror?
While brilliant parody is a human ability, I wonder if any AI exists which could beat a human at this game, writing on a topic in a given author’s voice, complete with characteristic digressions and caveats.
We’ve come a long way since Zork, a game with a simple English language “parser” that got me interested in technology. With a large enough writing sample, perhaps 20+ years of blogging like mine, I think an AI could do this, but it would likely require AI understanding of natural language far beyond what is known to currently exist.
It was a dream of mine long ago, to create an AI clone of the way I think. Once that’s done, I can retire from blogging and let the AI me surf, find new strange news I haven’t already covered, then write about it, adding some research and some mildly witty closing comment or observation at the end.
The state of the art seems far behind this but is still very interesting. In 2016 Google had a paper about a system that could flow and transition from one sentence to another. In the example below, the two bold sentences are the ones humans provided, and the sentences between were filled in (written) by the AI.
Recurrent neural network language models (rnnlms, Mikolov et al., 2011) represent the state of the art in unsupervised generative modeling for natural language sentences. In supervised settings, rnnlm decoders conditioned on task- specific features are the state of the art in tasks like machine translation (Sutskever et al., 2014; Bahdanau et al., 2015) and image captioning (Vinyals et al., 2015; Mao et al., 2015; Donahue et al., 2015). …
He was silent for a long moment. He was silent for a moment. It was quiet for a moment. It was dark and cold. There was a pause. It was my turn.
More relevant is this Writing Style fingerprint recognition system. It doesn’t write, but the details give us insight into how it might be done in part.
Trend Micro Writing Style DNA is a new layer of protection against BEC attacks, which uses AI to “blueprint” a user’s style of writing, employing more than 7,000 writing characteristics. When an email is suspected of impersonating a high-profile user, the style is compared to this trained AI model and a warning is sent to the implied sender, the recipient and the IT department.
Amazing, I think.
Related: Aug 2016, Now computers can clone your handwriting with creepy accuracy. – link
Handwriting? This was something done long ago around the time of John Hancock when there were no smartphones able to type out our spoken English … and no Internet if you can believe such a thing.