IGGI 2020 Project Proposals
Some students want to define the topic of their PhD on their own. Others prefer to work from a starting point that is already developed. To support the latter group better, for the current round of IGGI studentship applications, we have invited a number of individual academics and companies to propose the following concrete project ideas students can apply to.
If you are interested in applying for an IGGI PhD on one of the topics below, here is what to do: Contact the named supervisor of the project. They can tell you more about the project. If they think you might be a good fit for the project (and you like it), they will ask you to develop it into your own detailed project proposal, as with a standard IGGI application. If there is a lot of student demand for a particular project, they may also ask you to develop your own alternative project proposal in addition. If the proposal involves a company and you are shortlisted for an IGGI interview, a company representative will sit in on your first interview to advise your supervisor on your fit with the project.
Either way, IGGI doesn't privilege any kind of project, whether developed by you on your own or developed in response to the proposals below: in the end, the overall strongest candidates will be offered positions.
Personality Assessment through Gaming
How can people’s personality be effectively assessed through gaming? Personality traits like the “Big Five” (openness to experience, conscientiousness, extraversion, agreeableness, and emotional stability) are used in many different contexts, like personnel selection, education or professional development, but existing survey instruments for assessing them can be disengaging or easy to second-guess and ‘cheat.’ This project is about exploring how existing or bespoke games and the data they produce can be designed and analysed to generate psychometrically reliable and valid personality assessment that can be used to compare across individuals and within oneself over time.
Exploring biodiversity conservation management using online games
Conserving native species often requires dealing with pests that threaten them. Different ways of managing pests (like genetic modifications or chemicals) find different levels of social support and have numerous, sometimes unintended consequences. We want to develop and use online games to teach players about the complex decisions involved in pest and conservation management, and to understand how they make decisions. For instance, how do players find compromises in a multiplayer world with multiple stakeholders, limited budgets, and neighbours with conflicting interests? Target audiences include policy makers, land managers, conservation groups and secondary school/undergraduate ecologists, with opportunities to visit and work with UK and New Zealand audiences through partners like Fera Science and Manaaki Whenua/Landcare Research.
Applied Games for Treating Phobia in Children and Young People
In the UK, one in eight children and young people today suffer from a mental disorder (NHS, 2018). For most emotional disorders like phobias, there are effective, evidence-based treatments that are easy and safe to follow, yet 70% of children and young people do not receive sufficiently early treatment (Mental Health Foundation, 2019) – because health care systems are insufficiently staffed. This project explores the use of mobile games to deliver mental health treatment for children who would otherwise wait for treatment. Specifically, the project will explore how to design a mobile game that delivers graded exposure treatment for youth with phobias aged 7-11, and evaluate its efficacy.
Immersive games for general intelligence
How do we cope with new problems and scenarios (general intelligence) while even the most powerful computer can only master specific problems in familiar contexts? A striking comparison between biological and artificial intelligence is the ability of newborn chicks to generalise and identify relational similarities few hours after birth, while AI struggles to generalise simple relational concepts such as “same vs different”. In this project, we will develop an immersive game for chicks to identify the crucial components of general intelligence and apply them to the next generation of artificial intelligence.
Language development through games
Can playing interactive fiction games improve students’ English language skills? International students now play an important part in many UK universities, but many struggle with the linguistic demands of their programmes and fail to achieve their full potential. Previous research has shown that reading for pleasure improves broad language skills and leads to improved educational outcomes. This project considers whether text-oriented games such as Disco Elysium, Heaven's Vault or 80 Days could be used to motivate international students to read more English text, and how playing such games could improve their language skills and educational outcomes.
High Precision Battle Predictions for RTS Games
Many strategy games feature complex battles among different factions where armies, made up of units with different qualities, fight each other with different goals, from taking over enemy settlements, defeating the opponent’s army in the battlefield or defending specific areas of interest. Being able to predict the outcome of these battles in advance is a challenging task, but it would be useful for AI and players alike. A precise battle auto-resolver could provide players with the possibility of producing believable battle results in a simulation, and AI players could use it to determine how likely is that the battle turns out in their favour.
The goal of this project is to build a model for a precise battle outcome simulator from past played battles, providing more personalised predictions based on the player’s play style and army composition. The battle result should provide details such as how many units were killed or captured on both sides and how much damage each unit took. It should also balance between overly favourable or very unfavourable outcomes, represent basic playing strategies, and take into account highly valued units, arbitrary priorities of the armies, different battle scenarios and the types of units involved.
Procedural Generation of Annotated Maps and Cities for RTS Games
One of the most important components of strategy games are the maps and cities where decisions and battles take place. These maps and cities are normally created by designers who introduce multiple constraints in order to produce the desired gameplay. For example, some maps may include different terrain elevations, high points of interest that need to be defended, or strategic features such as bottleneck areas to create challenging scenarios. Additionally, these maps are highly reliant on manual mark-up, which is later used by AI agents to drive their decision-making process during the game. All these requirements make the auto-generation of high-quality maps and cities a very challenging task.
The goal of this project is to develop a procedural generation system for maps and cities that automatically incorporates such restrictions and annotations required for the game. The proposed solution would be an AI-assisted tool that the designer can use to experiment with different rules and restrictions, in order for them to analyse how different designs can impact the final game.
Local Forward Model Learning for Sample-Efficient Sequential Decision Making in Open-World 3D Games
The latest breakthroughs in game playing AI have primarily focused on the application of Reinforcement Learning (RL) and Statistical Forward Planning (SFP) methods, such as Monte Carlo Tree Search (MCTS) and Rolling Horizon Evolutionary Algorithms (RHEA), to games. Advances in Go, Atari and Starcraft have been achieved by combining Deep Learning and different forms of model-free or model-based RL, including variants of MCTS. Model-based and SFP algorithms require access to an internal model of the game that allows agents to reason about the future, enabling a more data-efficient and flexible behaviour than those learnt with model-free Deep RL. However, in many real world applications and complex games (such as many created in the commercial games industry) this model is non existent, not available or computationally too expensive to use. There is a prominent line of research that currently aims at automatically learning these models from interactions with the environment, showing that learning a forward model provides an important boost for action decision making. However, the scenarios normally employed in the literature are simple.
Using the Malmo platform for AI experimentation in Minecraft, this proposal focuses on approximating forward models in complex games by learning local interaction functions, to then investigate the use of SFP algorithms for non-player character control in these domains. Specifically, this research aims at (1) learning local forward models in 3D environments with partial observability; (2) implement SFP methods that would plan over said learnt models; and (3) investigate how the presence of non-stationary policies of other agents affect both model learning and SFP performance.