Predicting Human Behaviour Through AI

A team from Columbia collaborated to produce an algorithm to give machines a more intuitive sense for what will happen next by leveraging higher-level associations between people, animals and objects.

Our algorithm is a step toward machines being able to make better predictions about human behaviour, and thus better co-ordinate their actions with ours. Our results open a number of possibilities for human-robot collaboration, autonomous vehicles, and assistive technology.
— Carl Vondrick, Assistant Professor of Computer Science at Columbia

Researchers state that it is the most accurate method up to date. After analysing thousands of movies and shows such as The Office and Desperate Housewives, the algorithm articulated human interactions into patterns and learnt to predict hundreds of activities such as high-fives and handshaking. When it is unable to predict the specific action, it finds a higher-level concept that links them, in this case, “greeting.”

Attempts have been made regarding predictive machine learning which initially attempted to create pixel-by-pixel representation of future outcomes. This machine learning system relies on abstract imaging of cues deemed important to social interactions. The algorithm decides whether to classify the action as a hug or even a non-action like “ignore” however, when uncertainty is high, these machine learning models are unable to find any commonalities between the possible options. After hours of watching the footage, the results told, the computer accurately guessed 43% of the time whereas its human computer parts guessed correctly 71% of the tine.

PhD students, Didac Suris and Ruoshi Liu, looked at the longer-range prediction problem from a different angle. In school, students learn the familiar rules of geometry such as straight lines are straight, and most machine learning systems obey these rules. However, other geometries have counter-intuitive properties where straight lines bend. Both Suris and Liu used these to build AI models that organised high-level concepts to predict humans' behaviour in the future.

Not everything in the future is predictable. When a person can not foresee exactly what will happen, they play it safe and predict at a higher level of abstraction. Our algorithm is the first to learn this capability to reason abstractly about future events.
— Didac Suris

The mathematical framework allows the machines to organise events by their predictability in the future. For instance, swimming and running are forms of exercises, the machine can categorise these activities on its own. This technique could enable computers to size up a situation and make nuanced decisions. It is a critical step in building trust between humans and computers.

Trust comes from the feeling that the robot really understands people. If machines can understand and anticipate our behaviours, computers will be able to seamlessly assist people in daily activity.
— Ruoshi Liu

Whilst the new algorithm may permit accurate predictions, the next step is to verify the accuracy of these predictions outside of the lab. If the system can work in different environments, there are many possibilities to deploy machines and robots that may improve out safety, health and security.

Article Thumbnail Credit: