The steady march to create computers that can do everything we can continues, with a development by Alex Graves, a Junior Fellow in the Department of Computer Science at the University of Toronto, making it possible for them to write as if with a human’s virtual hand. Given enough text to work with, the recurrent neural networks used can be trained to copy a writer’s particular style.
Originally publicized by Google’s Research division in a blog post and social media update, but then removed, the article explained that Graves had used a long short-term memory recurrent neural network to analyze someone’s handwriting and then use prediction to determine how that writing would be continued. With that completed, technically the computer could continue to write as that person, to a high level of accuracy.
Looking at a simple piece of information and extrapolating that out to much more complex scenarios has a large number of uses, with handwriting synthesis being a prime example. However, the actual replication of a human’s handwriting is not necessarily a useful tool in and of itself, at least not for conducting something above-board. Theoretically, of course, it could be used to falsify documents or to trick people into believing someone they know intimately enough to recognize their handwriting, has contacted them.
Of course this was possible for nefarious humans before now — it just required a lot of time and dedication. Now, however, at least in theory, someone with access to one of these sorts of neural networks could pretend to be anyone from whom they were able to gather enough handwriting samples to source from.