Artificial Intelligence in Digital Media

Social media, massive multiplayer on-line games (MMOG’s), information mashups, etc. have become everyday life for most of the younger generations. In order to improve user experience in digital media, we have studied, designed and implemented teachable software agents, which can perform tasks that require learning and human-like behaviour. Such tasks may include, but are not limited to:

Constructing artificial game characters which behave like humans do
What is learning and behavior in computational sense? Recording just patterns of movements does not remain behavior. Behavior, also in digital environments, is about networked consequences of our experiences that cannot be recorded as patterns. Semantic neural networks are one successful method to model complex systems, such as behavior, learning or natural dialogues. In other words this enables games’ characters behave like humans, which makes playing more entertaining. Results have been applied e.g. in AnimalClass game series.

Producing self-organizing media
Self-organizing media can be understood as a subset of adaptive media, which is basically designed to produce optimized user experiences. The technological challenge of this study is related to the complexity of the systems. Results have been applied e.g. in Navigator –content management system (distributed by Otava Publishing Company Ltd.).

Applying artificial labour in testing and controlling systems
Testing and controlling media environments can be challenging because of the complexity of environments, diversity in behaviour and extremely high number of interactions. Because of this, agent –based control and testing systems are one promising application area for artificial game characters. E.g. Navigator Test Bench is based on this idea. The newest application area is about personal security in social media. When the research is completed, the results can be applied in products for example in following ways:

1) Software agent has found an unexpected social network in a social on-line game, meant for children. A human moderator checks manually the actions of the person behind suspected behavior. Without computational analysis, no-one can monitor hundreds of thousands of players.

2) A user of video blog system has changed remarkably his behavior in last few days. Software agent informs human moderator that redirect the finding for further process if necessary. Once again, monitoring 100 million users is not possible without computational approaches.

Background, goals and results 
The main objectives of this research are: 1) to develop methods to better understand user behavior in digital environments and 2) to apply and evaluate the methods and theories in real media production cases. These cases are important in order to evaluate the usefulness and validity of the complex adaptive systems –based methods. The research is based on theories and methods from the complex adaptive systems, cognitive psychology of learning, Bayesian modeling, neural computing, machine learning, data mining and social networks research methods.

The results will be significant and useful to administrators, educational researchers, game developers, network operators and media producers.

The main outcomes are novel attempts to analyze and understand user behavior in networked media environments: usually behavior is only mined using computational methods (detected patterns), while explaining and understanding the user is done mostly with qualitative methods. As a completely computational method, it is possible to compare and apply the results achieved by behavior analysis methods outside the used context.

RESEARCH PARTNERS

Otava Publishing Company Ltd, Finland
Tampere University of Technology, Finland
University of Turku, Finland

More information:

Publication list (PDF)