|Funding for:||UK Students, EU Students|
|Funding amount:||£16,777 p.a. subject to confirmation and eligibility status|
|Placed On:||8th December 2018|
|Closes:||17th January 2019|
Location: Engineering, Bristol
Funding amount: £16,777 p.a. subject to confirmation and eligibility status
As AI agents are deployed in complex open-ended environments and regularly interact both with other autonomous systems and with humans, we will need to find ways to judge their trustworthiness. This will require them to explain or justify their decisions and actions to meet fundamental standards of transparency.
Providing such explanations is straightforward if an autonomous system’s decisions are based on a manageable number of rules that can be easily understood by humans. However, optimal decision making for complex environments takes account of multiple interactions to minimise multi-dimensional cost functions, making the results extremely difficult to explain, due to the very high number of alternatives considered. Human understandable rules tend to be low dimensional and consider only small numbers of interactions at a time. This identifies the great importance of the trade-off between the performance of a system and how easy it is to understand the rules governing its behaviour.
This PhD project will explore the trade-off between understandability and optimality for autonomous navigation and collision avoidance. As a point of departure, we will study an existing system of rules such as the maritime collision regulations (COLREGS) or the Rules of the Air. These are both established rulesets governing navigation which are easily understandable and sufficiently flexible to cover a broad range of possible scenarios. They have evolved over time to minimise the risk of collision. However, both sets of rules are designed only for pairwise interactions between agents and hence there are interesting questions concerning their scalability and robustness in multi-interaction scenarios. As the project evolves, it will investigate an appropriate modelling framework (e.g. a language) for expressing the behaviour of autonomous agents and their governing rules. That language must be formal enough to support computer operations, natural enough to be understandable to humans, and sufficiently flexible to allow rules to generalise across different but related scenarios.
How to apply:
Please make an online application at www.bris.ac.uk/pg-howtoapply. Please select <Engineering Mathematics> on the Programme Choice page and enter details of the studentship when prompted in the Funding and Research Details sections of the form with the name of the supervisor
A First or high 2:1 Honours degree in Mathematics, Computer Science Engineering or a related discipline.
Scholarship covers full UK PhD tuition fees and a tax-free stipend at the current RCUK rate (£14,777 in 2018/19) plus an industrial top-up (subject to contracts).
Please contact Prof Jonathan Lawry, J.Lawry@bristol.ac.uk or
For general enquiries, please email email@example.com
Type / Role: