Current position
I take what you might - slightly pretentiously - call a full-stack approach to digital leadership, working across strategy, project management, and development and deployment of applications. As Academic Director for Digital Education and Innovation in a university school with over a thousand staff and ten thousand students, I identify and prioritise strategic needs, work with stakeholders to plan digital solutions, and support procurement or, increasingly, supervise the in-house development of digital tools.
I also lead in development of policy and guidance at university level, as co-chair of the university's AI Education Committee, school Digital Lead, and member of various committees around digital education, accessibility, hybrid learning, and the university's AI Accelerator. I have worked on policies for permitted use of AI in assessments, declaring AI use, AI Guardrails, building chatbots, and similar areas.
I also talk about digital issues or potential in a range of areas, with recent keynotes on AI in education at the UKCCC conference, Portman and Tavistock Teaching and Learning Day, as well as talks on specific AI projects at the University of Kent's AI seminar series, Learning at City St George's conferences, and internally running sessions for the University's Office for Institutional Equity and Inclusion on LLM and bias; the Law School, on using AI for legal simulation training; and various school- and departmental-level meetings. I am also in the process of writing a series of seven articles on AI in Healthcare Practice and Training for the British Journal of Cardiac Nursing.
Skills
Like many academics, I am an enthusiastic amateur developer, with skills in a range of tools.
I have used a variety of lab-based and online psychological testing software, including eye-tracking, but have tended to prefer to build things myself. Most of my early research was written in ActionScript in Macromedia (later Adobe) Flash, which combined JavaScript-like code with frame-based animation. I eventually moved over to developing studies in JavaScript with HTML/CSS, using PHP and MySQL for storing and retrieving data. I now use AI assistance (generally Codex / ChatGPT app) to support development work.
I analyse data and sometimes conduct modelling in R, which I find simultaneously infuriating and impressive.
I use tools such as Unreal Engine for game-like experience design on desktop or VR, basic Blender where needed, am confident with video and audio editing and a range of digital audio workstations for music production. I can create portable tools on an Arduino, with relays to control physical devices, as well as addressing breakout boards and low-voltage relays to control physical and mechanical devices from a laptop or the web.
I also understand the basics of the AWS ecosystem for professional deployment, and associated best-practice concepts around data security and scalability. I deploy and manage node servers on Heroku, and use pipelines in Google Cloud.
I run local LLMs using LMStudio.
Background
I have been working with computers since the age of ten, starting with programming on the Spectrum and BBC microcomputers.
Artificial intelligence
I have had a deep interest in artificial intelligence for over thirty years. I built my first chatbot as a young teenager, creating a bespoke ELIZA clone based on a DOCTOR script and adapting the code from the - for me - highly influential book Exploring Artificial Intelligence on your Spectrum+ and Spectrum. I also wrote slightly unsatisfying natural language block-manipulation programmes based on the famous SHRDLU.
As a natural sciences undergraduate at Cambridge, I was lucky enough to take a final-year psychology module on connectionism - my favourite module - where I studied concepts like multilayer perceptrons; back propagation, error correction and gradient descent; and other key features underlying the development of today's large language models (even if it was a bit of a backwater at the time). I also completed my undergraduate dissertation on the Rescorla-Wagner rule, which applies the same conceptual approach to error correction that underlies model training for LLMs. I remember in supervisions asking about how to implement backprop networks on my own computer, but a 486 SX 25 struggled to do the compute.
I returned to this challenge 25 years later, building a classic example of written-number categorisation learning to build intuition as to how AI systems work. I also developed and trained my own simple LLM, and have been fascinated to see how structured language gradually emerged through training (albeit in fairly rudimentary form, with echoes of Beckett prose...).
In terms of my stance on AI, from a technical perspective I find it incredibly exciting and satisfying to see AI able to do the things I dreamt of it doing as a student. At the same time, I am deeply unsettled about the way in which GenAI is emerging into the world, and the implications for power, society and the world we share. I know from cognitive dissonance that these are hard ideas to hold concurrently and that there is the temptation either to decide that AI isn't actually any good, or that its impact on the world will probably be fine, but I think it is important to hold the conflict in mind.
Digital research methods
I have enjoyed building a variety of digital proof-of-concept research tools. I conducted some of the earliest research using online recording of participant response times, collecting data from over 10,000 BBC viewers who clicked through to my website and completed a task-switching study coded in Macromedia Flash. I developed this strand of work more formally, publishing on the reliability of browser-based response-time collection, and extending this to the use of what would eventually become smartphones. I have since used that expertise in a range of projects, including the UK Women's Cohort Survey, in which participants completing a paper-based survey could visit a website and complete a response-time test, writing a code that contained their results on the survey response sheet. Most recently, I led the design and implementation of a series of child-friendly tablet tasks for the BICYCLE project, where participating families were mailed tablets to their homes and had their children complete onscreen activities while on a Zoom call with one of the researchers.
I have also used game peripherals like the Wii Balance Board as input devices for behavioural studies, although I never found any interesting effects beyond the insight that participants enjoy these kinds of experiments.
Educational support tools
I also have a background in building digital tools for education. Most of these have been relatively small scale, and have included text-message-based in-class responses to quizzes, before the advent of smartphones; real-time feedback tools where students can use a set of emojis on their smartphones to respond to in-class questions, or indicate anonymously to their lecturer that they're confused or flagging.
Most notably, I supervised the development of Quodl, a web app for in-lecture quizzing and learning support, which was used with over 2,500 students across City St George's and beyond, with very positive feedback. It was runner-up in the Guardian University Awards in 2017, and led to the creation of a spinout company, though it wasn't possible to secure enough funding to take it forward.
Digital projects
Showing all 9 projects.