New tech teaches computers to type like humans

An overview of the modeling approach. Credit: Danqing Shi et al.

Researchers at Aalto University have developed an innovative predictive typing model that can mimic how different people type on their phones, whether they use one hand or two, or whether they are younger or older.

This new model, which will be showcased at the upcoming CHI Conference on human-computer interaction in May, could revolutionize the way we optimize our phone keyboards.

Typing on a phone involves more than just hitting the right keys.

According to Professor Antti Oulasvirta of Aalto University, it requires manual dexterity, good eyesight, and a bit of memory work.

We need to be able to press the correct buttons, notice when we make mistakes, and remember what we were trying to say. While autocorrect features help some people type faster, they can make typing more difficult for others.

To tackle this, the Aalto research team, in collaboration with Google, created a sophisticated machine-learning model.

This virtual “user” has its own set of “eyes and fingers,” and it uses these tools to type sentences on a simulated keyboard.

The model goes through the same process humans do: making typos, noticing them, and then correcting them.

“We built a user model that mirrors the human visual and motor functions,” Oulasvirta explains. “After millions of simulations, it learned to type in diverse real-world scenarios.”

This approach is not just about building a better autocorrect system; it’s a leap toward improving how keyboards are designed in the first place.

Traditionally, new keyboard layouts are tested by observing real people—a process that’s both expensive and slow.

By simulating how people type under different conditions, this new model allows designers to test and refine keyboard designs much more efficiently.

Beyond just improving phone keyboards, Oulasvirta’s team at Aalto is focused on enhancing user interfaces in general.

By understanding how people interact with technology when carrying out tasks, they aim to make everyday digital interactions smoother and more intuitive.

This work is part of a broader effort to use computational models to predict human behavior without the need to watch hundreds of people.

The ultimate goal, says Oulasvirta, is to create user interfaces that are so well-tuned to human habits and needs that they help make society function better as a whole.

In a world where we all spend a lot of time on our devices, these improvements could make daily life easier for everyone.

Source: Aalto University.