Predicting Human Behaviour With Deep Learning

Artificial intelligence is the future. Generally artificial intelligence is portrayed as a supreme entity powerful to influence the technology. Machine learning is one of the science behind the supreme power of artificial intelligence and deep learning is the engine that propels the science. One of the most promising application of deep learning is to enhance interaction with the computers. Deep learning is particularly tailored for this purpose since it has the intuitive ability similar to biological brains. This implies that interaction of artificial intelligence with humans will demand some capability of the same to predict human behaviour.

In the past, people's behaviour was unpredictable because there was no means of collecting and analysing data that goes into the decision-making processes. Now with the entry of sophisticated computer system and cloud which helps to store huge amount of data complement the development of deep learning.

The conventional behaviour model is to develop a statistical or analytical model of human behaviour and to assign a distribution fitting to validate the model. The deep learning approach is quite different from the conventional model. Deep learning system learns to predict the outcomes by observing the human behaviour.

Some of the recent research works in MIT like Predictive Vision is a result of intense application of deep learning. Predictive vision system on YouTube videos from TV shows will predict whether people are going to hug, kiss, shake hands or slap a hi five. MIT trained the system for 600 hours of video and the system was able to predict actions of 43 percent.

Deep learning is also used to predict strategic behaviour. In most systems the assumption is that the participants perform in a perfectly rational manner and are based from insights from cognitive psychology and experimental economics. However, in this system that is based on deep learning, the system learns a cognitive model without the need for expert knowledge. This system is able to outperform systems that are built from expertly constructed features.

Another pragmatic use of prediction of human behaviour is in the confines of a car. The US-based company Brains4Cars uses sensor fusion deep learning system based on LSTMs to anticipate driver behaviour 3.5 seconds before it happens. It uses a collection of sensors such as cameras, tactile sensors and wearable devices to make its predictions.

Considering beyond predictions, deep learning has been used to assist in the context of human negotiation. In the process the reinforcement learning and hand-crafted agenda-based policy was framed. Both were evaluated by having negotiating each other in different setting. Surprisingly, reinforcement learning model consistently outperformed the hand-crafted agenda-based model. Human evaluation of the process also proved that the reinforcement learning model turned to be more rational.

Deep learning has already penetrated into the game world. Deep Stack is an algorithm for imperfect information settings. It combines recursive reasoning to handle information asymmetry, computation on the relevant decision, and a form of intuition that is automatically learned from self-play using deep learning. Deep stack bridges the gap between AI techniques for games of perfect information-like checkers, chess and also go with imperfect information games like- poker. Deep stack plays using intuition honed through deep learning to reassess its strategy with each decision. Deep stack became the first AI capable of beating professional poker players.

Deep learning includes the study of behavioural traits of humans and using it for predictions. Leveraging deep learning in business applications is phenomenal. The ability to anticipate and predict behaviour or estimate the chance of winning in a game are advantageous tool one could possess in their business arsenal.