Personalized predictive keyboards

While there is a lot of excitement in ground-breaking, revolutionary, and even moon-shot AI applications that will no doubt change our world, you don't always have to chase after these fundamentally difficult problems to bring value to our world with AI.

One good example of an apt intelligent-enabled solution for a recent real-world problem is predictive keyboards on mobile devices. When touchscreen mobile devices became popular, we had to learn to type on small virtual keyboards, often on the go, with fewer fingers and with more distractions. These predictive keyboards facilitated faster typing by suggesting words and punctuation that we may wish to type, thus reducing friction in mobile communication.

The predictive capabilities of these keyboards are often built with ML and natural language processing (NLP) technologies that combine language models, custom dictionaries, and learned preferences in their prediction engines. The best prediction engines are likely built using a form of recurrent neural network (RNN) called long short-term memory (LSTM). These are neural networks that try to predict the next word given a window of previously typed words. The key to a successful prediction is in its speed and personalization. Each key press results in a prediction, and so the prediction engines must run fast on mobile hardware. These prediction engines are designed to get more intelligent as we use them; they are good examples of human-in-the-loop online learning systems.

Even though they are not saving lives by the thousands, the users of these predictive keyboards have saved trillions of keystrokes. We just love the elegance of these intelligent-enabled software solutions that have applied the right AI technologies to the right problems. We hope that, with the skills and insights you will have gained from this book, you will be able to find elegant AI applications to make our lives better, as well.