Exploring How AI Can Gain Deeper Understanding of Human Behavior: Researchers Investigate
Technology is an enormous resource for humanity, but it is still underutilised. According to Aalto University professor Antti Oulasvirta one of the biggest reasons is that computers have a poor understanding of human behaviour, thought processes and experiences.
‘Humanity would get a great deal more out of technology like AI if computers had a better understanding of their users,’ he says.
Oulasvirta studies the interaction between humans and computers and creates computational models of human behaviour. These models can be used to predict, for example, how easy it is to use a mobile device, but can also explain behaviour, like why some feel that AI-assisted text input is difficult while others do not. The modelling of behaviour has taken a giant leap over the past decade, while machine learning has made significant steps forward.
The European Research Council awards the Advanced Grant to experienced researchers with a track-record of significant research achievements for exceptional or unusual scientific openings. Oulasvirta has previously received the ERC’s Starting Grant for promising researchers in 2015.
‘In the previous ERC project, we were able to improve usability and the user experience in about 5-25 per cent of cases. The new methods will allow us to achieve a great deal more,’ says Oulasvirta.
The aim of the new ERC project is to produce new types of models, which can be applied extensively and easily to different use cases, such as designing a driver’s dashboard in a car. When this can be tested with an AI user model in a 3D simulation, the successful deployment of the design and technology is more likely. Models can be used to explain behaviour in different situation or to optimise user interfaces such as websites and menus.
‘Our vision to be able to make the work of computers easier in the future by giving AI means to better interpret and predict people’s behaviour. This involves machine learning and simulations of the functioning of the human body and the thought process. With this simulation, AI can find models for explaining its observations and choose the best way for acting with people,’ Oulasvirta explains.
‘It is not enough for us that our user model works, but if we succeed, we will also have the opportunity to talk to the user model within the operating environment.’
How can computers be taught that people are all different?
‘We are combining two previously unlinked approaches. First, we will use theories of human behavior and cognition, which describe people through psychology and computational principals, in our research. For the first time ever, machine learning will allow us to predict a large number of variables that describe behaviours and teach behavioural models in environments without needing observation materials from a large group of people to make predictions.’
New lessons from the United States
The ERC project is spurred on by a sabbatical year at UC Berkeley, where Professor Oulasvirta has focused on the more in-depth use of methods from robotics. He feels that the nine-month visit has been extremely successful and is returning to Finland with new insights in May. He has focused on the interaction between robots and humans, and the methods used for learning about human preferences and objectives in robotics.
‘I have learned a great deal about algorithms and how they are made in practice. Computational models require a large amount of manual work, but also theory.’
In addition to the five-year-long ERC project, his work in Finland will continue at the Finnish Center for Artificial Intelligence FCAI, as a coordinating Professor for Interactive AI research program. FCAI gathers top Finnish AI expertise across Aalto University, the University of Helsinki and VTT Technical Research Centre of Finland. FCAI’s research aims to develop AI that can both learn and plan and works in collaboration with people in resolving complex problems.