AI learns to decipher doctors' handwriting

Experts from the Mohamed bin Zayed University for Artificial Intelligence in the capital of the United Arab Emirates (UAE), Abu Dhabi, have developed a neural network that can imitate human handwriting. True, they promise to use it not for forging other people’s signatures, but, for example, for deciphering doctors’ handwriting, which it can also do.

As reported by Bloomberg, the authors of the new AI used a special type of neural network that is designed to study data sequences - the so-called transformer model. According to the developers, this model requires only a few paragraphs of handwritten text to teach a certain style. However, so far the neural network can analyze and reproduce texts only in English. She is a little worse at French, but Arabic script is not yet decipherable.

According to the developers, the tool they created will be used thoughtfully and carefully and will not lead to forging other people's signatures. Instead, the neural network, as already noted, can be used to decipher doctors’ handwriting, create personalized advertising and other purposes.

In addition, this AI can generate data to improve the handwriting recognition performance of other neural networks.