An artificial intelligence that mimics how people type on a smartphone – including making errors – could lead to better on-screen keyboards.
“There are some choices we make that it seems like the human mind optimises for when we type,” says Jussi Jokinen of Aalto University, Finland. “I wanted to do the same using computer software, then optimise that and see if its behaviour was similar to humans.”
Jokinen and his colleagues programmed the AI using existing knowledge about human behaviour. The AI has a number of components – one to track its “finger” on the screen, another to look at the screen and decide which keys to press, and a third to proofread and correct any errors.
A “supervisor” decides when to activate these components, and the researchers tasked it with learning how to replicate how people use smartphone keyboards. “It figured out what’s the optimum way to allocate resources and learn how to type,” says Jokinen, who presented the work at the virtual CHI conference this week.
The AI’s time between keystrokes averaged 399 milliseconds, comparable with the human average of 381ms measured in Jokinen’s previous research. Words typed per minute and the number of backspaces used to correct errors – and when in the writing process they were corrected – were also largely similar to those of people.
“A key innovation in this paper is that the behaviour emerges from the model, in contrast to replicating patterns from a fixed data set,” says Daniel Buschek of Bayreuth University, Germany. “I am excited about the opportunities that this brings for simulating user behaviour to quickly test new interface ideas.”
That is also Jokinen’s goal. “My hope is that designers can use this computational tool to very quickly evaluate ideas they have and basically ask the models, how would users type if they were given this keyboard?” He suggests different key layouts could be tested, as well as more ergonomically friendly digital keyboards.
More on these topics: