What we’re looking for

On this page we provide a few pointers on what we’re looking for in applicants to our CDT and why we’re looking for this. We do not expect all applicants to be able to cover these in their applications – however, we’re sharing this at application stage for potential applicants to consider if the CDT is a good fit for their career aspirations.

Alignment with our Skills Domains

First and foremost, as per the Application Documents page, we are looking for applicants that clearly align with our five Skills Domains.

We created these Skills Domains when developing the proposal for this CDT through conversation with our external partners and based on our scoping of cutting edge developments related to NLP. The Skills Domains represent what we feel are some of the core areas of expertise and gaps in research and practice that need to be addressed in order for NLP applications to be responsible designed and adopted and be trusted by societies.

We think it’s very unlikely – in fact, impossible! – for any single applicant to be able to cover all of these Skills Domains. Instead, we expect that applicants will be able to identify one of these Skills Domains as their primary area of focus during their training. This means they might form research questions related to that domain, have a primary supervisor who is an expert in that domain to guide them, and take advanced courses at the University of Edinburgh related to this.

However we also expect all students to also demonstrate an interest in expanding their skills and knowledge beyond this primary domain into one or two of the others. This, we believe, is part of the added value to students in studying for a CDT – that they can expand their expertise into highly relevant subjects or disciplines that are adjunct to their core areas of expertise. This means potentially having a second supervisor that has expertise in a different Skills Domain. We also expect that, as students develop their expertise in different areas during their studies, new exciting research questions might start to develop that bridge these areas – for example, questions that look the implications of emerging AI regulations on cutting edge NLP modelling techniques.

Diverse cohorts of students that are interested in collaborating with each other

We will be seeking to develop cohorts that have diverse and complimentary disciplinary backgrounds and alignments with the different Skills Domains. This means that we are looking to avoid recruiting cohorts that all have similar skillsets and background training. Instead, we’re hoping there will be fruitful exchanges between students on different perspectives related to the design and development of responsible NLP systems.

Over the course of the programme there will also be opportunities built into the training to exchange skills within the cohort, and to work together on applied projects. This includes credit-bearing courses that will support the development of skills related to interdisciplinary collaboration, applying these skills with CDT partners, and in teaching other core concepts related to ethical and responsible NLP and AI. On the back of this training, we also expect some students will collaborate together on projects directly related to their individual PhD projects, and be able to use their own contributions to these projects within their final thesis’.

While the training programme of the CDT will give you the time and resources to develop these skills, we will be looking out for an interest and desire to collaborate with others at application time.

Benefitting from the integrated training of the CDT

A key differentiator from the experience of studying for a PhD on a CDT compared to a more traditional PhD is the integrated training element. Here, students will study on 180 credits worth of courses throughout the first three years of their PhDs. This give the opportunity to go more deeply into refining specific research skills, or to dive into advanced courses to develop more specialist expertise, or to be more expansive and gain experience in relevant subjects outside your background training. This also means students get the chance to develop new skills as they progress through their studies depending on where their research takes them. This is also why you have four years for an integrated PhD, to extend the amount of time students have for this more advanced training and to develop their research project alongside this.

This also means that we do not expect students to have all the fundamental skills required for studying responsible and trustworthy NLP at application stage. We also do not expect well defined research proposals at this stage – as, quite simply, these will change over the course of your studies!

If you are less interested in studying for credit-bearing courses as part of a PhD, or have a specific project you’d like to conduct and consider this very well defined already, then it’s like a CDT training programme is not for you. Instead, consider applying for a more traditional PhD programme, such as those hosted by the Institute for Language, Cognition and Communication.

Seeing the bigger picture of NLP and AI developments

Regardless of what discipline or background training a student has, we like to see applicants and the students that will eventually be training with us on the CDT as advocates for seeing the wider implications and “big picture” of developments in NLP and AI. This includes demonstrating an awareness of the contemporary and emerging debates around the ethics of NLP and associated data, technologies and systems, and understanding in a nuanced way the debates around the potential harms these technologies could potentially generate for specific social groups, populations, and the planet.

At the same time, it’s important to note we’re not inviting applications that critique these technologies for the sole intention of critique! We are looking for novel solutions, tools, methods, and practices that will help us resolve the complexities of these technologies. We also do not expect students to resolve these issues alone – which is why we focus a lot of our training on collaborations between students, getting input from experts from outside the University of Edinburgh, and in sharing what we find out for others to build on and learn from.

A desire to see ethics and responsibility applied in practice

Related to the above point, we also wish to see perspectives on ethical NLP and AI applied in practice. Over the last decade there have been a wide variety of theories, concepts, and guidelines related to the ethical development and oversight of AI systems, but there is also recognition that it is often hard to then apply or adhere to these in practice. We’re fortunate that our CDT is co-located with a UK wide programme focused on bridging these “divides” between technical and non-technical perspectives on responsibility and ethics. We’d therefore welcome students that are interested in tackling these challenges, and have radical ideas that lead to improved applications of responsible AI and NLP principles into the design and development of these systems and their underpinning techniques.

Awareness of recent research related to responsible and trustworthy NLP and AI

We’d expect some awareness of recent debates and research of relevance to the focus of the CDT, and encourage applicants to signpost these in their application documents to demonstrate this awareness. Some places to start connecting with cutting edge research on topics related to the CDT include ACM FAccT conference, AAAI / ACM AIES conference, the different ACM SIGCHI conferences and journals, the annual meeting of the ACL and its transactions.

However, the interdisciplinary nature of our CDT means what counts as recent and “cutting edge” research might be very diverse. We also recognise that different fields and subjects have very different publication practices – in computing related fields where conferences still often take priority then publication of work can be relatively rapid, and there are common practices to publish pre-prints of papers on platforms like Arxiv. In other fields publication can be slower – for example in the humanities and social sciences it can take over a year or even several years to have a manuscript published in a journal. In some disciplines it is also more common to publish monographs, which takes years. So do not feel restricted to just those communities noted above – and during the PhD training itself we will be supporting students to explore diverse forms of knowledge and sources of literature.

Aspirations to work on real-world problems and to develop meaningful applications that matter to real people

An important term in the official title of the CDT is in-the-world. This is an adaptation of the term in-the-wild. This phrase has been used extensively within the field of human-computer interaction to distinguish a shift from studying peoples use of technologies in laboratories or under constrained experimental conditions, to instead designing and studying experiences of technologies in “real world” situations. In-the-wild has also been used by the UK research councils as a way of distinguishing research conducted in real-world, applied and ecologically valid situations compared to that which is more fundamental or lab or bench-based.

We’ve tweaked this terminology to in-the-world, to signify how NLP technologies are already being deployed in peoples lives, and how they are having very real consequences already for how organisations in-the-world operate, and how people in-the-world experience their everyday lives. We’re therefore interested in students who are interested in studying peoples experiences in real-world and everyday settings, and in exploring how cutting edge techniques on responsible NLP can be deployed safely in real-world settings. This means getting out of the office, the lab, or the studio, and being willing to talk to real people, and collaborate with them!

An interest in translating research into practice – be this in companies, in policy, into advocacy organisations, into start ups, into social enterprises, into toolkits for others to apply, and more

Finally, we’re looking for students that have a long-term ambition to translate the research from their PhD – or research they do after their PhD – into practice. The training programme will offer students many opportunities to explore how to do this, including: CDT specific training sessions focused on configuring research projects for societally impactful outcomes; joining a cohort of students developing technology focused start-ups in the Bayes Centre; studying on EFI’s Codebase Bridge course to develop entrepreneurial ambitions; possibilities to spend time with policymaking organisations to translate PhD research into new policies and regulations; and spending time with industry partners on internships and placements. It’s very unlikely that a student will do more than one of these various opportunities – and translation into practice will be more challenging for some students than others. However, we are especially interested in applicants seeking training in this as part of their studies.

Beyond this, as part of the CDT you will also have the opportunity to contribute to an Open Source “field book”.  The field-book will be woven into the taught courses and annual workshops with all the students, and be a way of collectively documenting and sharing emerging practices and principles related to responsible and trustworthy NLP. These will be shared with partners and the wider AI community, and be a further way of translating research into real-world practice.