Screenshot of the model developed by the researchers. Photo: Aalto University

Science and technology

Researchers at Aalto University have achieved a breakthrough in text input technology by developing the first AI model capable of simulating human-like typing on mobile phones. This innovative model, a significant milestone in modern text input simulation, enables the analysis of various user groups, from young to old and those who type with one finger, to determine factors that facilitate or impede mobile typing.

Professor Antti Oulasvirta, leading the research team, explained the complexities of mobile typing that the model addresses. "Typing on a phone requires dexterity, visual perception to press the right keys, proofreading, error correction, and memory usage. Features like auto-correction can assist some users, while for others, it complicates the typing process," he said.

The AI model mimics human actions by moving its 'eyes' and 'fingers' and using its 'memory,' replicating typical human errors and their corrections during typing. "We created a simulated user endowed with human-like visual and motor systems. We trained it millions of times in a keyboard simulator until it acquired typing skills applicable in various real-world scenarios," Oulasvirta shared about the research process.

Aiming for Better User Interfaces

Developed in collaboration with Google, this predictive model for text input aims to evaluate and optimize mobile keyboard designs, complementing the often costly and time-consuming user testing with real people.

Antti Oulasvirta heads a research group at Aalto University that focuses on computational models of human behavior, particularly in how people interact with user interfaces. "We can teach computer models without needing large-scale human observational data to make predictions. With interfaces all around us, the essence of our research is to foster a more functional society and smoother everyday interactions," Oulasvirta highlighted.

The findings will be presented at the CHI conference in mid-May, the premier international forum for publishing and discussing research on human-computer interaction. The publication is already available online.