Venue: Edinburgh Futures Institute   [Google Maps directions]  

Access: Via Lauriston Place only 

Room: 0.01, located on Level 0, accessible via lift (press 0 button) or central stairwell 

Quiet Room: 0.10, on Level 0 of EFI accessible via lift or stairs 

There are also individual seating areas available on each floor with charging capacity. 

WiFi – See more information on how to connect to Guest WiFi (VisitEd) via the link below:

Visit-Ed Guest WiFi Service | Computing | Information Services

Course organiser Prof Frauke Zeller will lead student team representatives through reflecting on learnings from their respective case study projects which spanned London Stock Exchange Group, Publishing Scotland, and Mentor360 organisations. Dr Iain Rodger of Mentor360 will share company perspective from having had a CDT team assess their product through the course.

Panellists: Osman Batur, Sarah Immel,  Bríd-Áine Parnell and Dr Iain Rodger, Mentor360, Staff Machine Learning Engineer and Technical Lead 

Moderator: Professor Frauke Zeller, Co-Director of Design Informatics and this course organiser

 The session will focus on a series of public engagement activities centred on the theme of deepfakes and synthetic audio/video. These activities invite members of the public to interact directly with modern speech/video generation technology – such as cloning their own voices and attempting to distinguish between real and synthetic speech samples.  

Presenters: Ariadna Sanchez, Jinzuomu Zhong, Artemis Deligianni, Giulio Zhou  

Room: 3.35 

Moderator: Daniel Shorr

Talk 1:  The Next Frontier: Transparent, Explainable, and contestable AI for Social Good 

Presenter: Amanda Horzyk

Abstract: Want to stay ahead of the curve? In today’s saturated black-box AI market, it is time to challenge the status quo and envision new architectures of trust. Gain fresh insights drawn from global expert and diplomatic convenings and interdisciplinary sessions at the intersection of law, policy, and ethics in AI. Under the influence of political, legal, and technical pressures, our practices and collective agenda continue to evolve. This talk explores how these forces drive the emerging mandate for transparency, explainability, and contestability—principles that have become close to non-negotiable in responsible AI development. System design mirrors a decision-making pipeline: a sequence of architectural and organisational choices, each carrying its own value assumptions. How we make these choices determines what we optimise for and who participates in shaping outcomes. The session introduces a holistic framing of explainability that responds to today’s regulatory, societal, and technical momentum.   

Talk 2:  AI Break-Ups & Lobotomies: The Need for Standards around Discontinuation of AI Companions.

Presenters: Alexandra Gillespie and Dr Laura Bishop of British Standards Institute, AI and Cyber Security Sector Lead  

Abstract: AI companions sit at the intersection of a human need for connection and the increasingly compelling AI-powered chatbots available. Even chatbots that were not initially designed for companionship have become confidants and caregivers. Sometimes, these relationships form without the user’s immediate attention or intention. As use continues to rise, so too does the potential fallout of discontinuation. Bans, modifications, or business failure could end the relationships of millions of users all at once. Already, instances of companies updating their services have sent users into spirals of grief and confusion. The rapid investment and parallels to the “dot-com-bubble” mean that the discontinuation of many services is likely, if not inevitable. Pulling on literature from a variety of disciplines, we outline four types of discontinuation and four spheres of consideration relevant to the formation of standards around AI companion discontinuation.   

Room: 2.55

Moderator: Professor John Vines

Presenters: Neel Rajani, Rayo Verweij, Kimberley Paradis 

Neel Rajani will explore the risks of continual learning in large language models. 2026 has been been declared by some to be the year in which the Continual Learning problem in LLMs is solved. Frontier labs are pushing for this objective both for sustained capability acquisition, and aggressive user personalisation, but it would introduce new risks that we are not even remotely prepared for. Even in the existing paradigm, where the parameters of an LLM remain constant after deployment (aside from sporadic model updates), safety and alignment testing are extremely brittle and deeply flawed. But when the parameters of a model continue to update throughout deployment, specific to every user individually in real time, your “guardrails” fly out of the window. In this talk, he will explain why he is worried about Continual Learning, and why you should be too.  
  
In “Let Us Find a Better Word for De-Anthropomorphisation”, Rayo Verweij will discuss the dangers of ascribing human-like qualities to large language models. In addition, he will share early work on interventions that open up the technology inside LLMs to non-experts and allow them to construct more accurate mental models of how they operate. Finally, he will enlist the audience in his quest to find a more accessible term for “de-anthropomorphisation” – but be warned, there may be a level of audience participation.  
  
Kimberley Paradis will present on how language technologies are used in everyday contexts such as health and wellbeing, often in response to gaps in existing support systems. These uses highlights how these technologies are taken up in practice, and how participatory and feminist approaches can inform more grounded and accountable forms of design.  

Room: 2.35

Moderator: Professor Frank Keller

Presenters: Sydelle de Souza, Harvey Donnelly, Ivan Vegner 

Abstract: Does cognition have structure? For decades, this question has driven impassioned theoretical debates and vast bodies of empirical work across disciplines — yet it has largely fallen by the wayside in recent AI research. As frontier models grow larger and more opaque, and available data runs thin, explicitly structured representations and processes offer a compelling path to doing more with less.  

In this talk and panel session, we will explore questions of structural abstraction and generalization, data efficiency, and mechanistic AI safety by bringing together research on language evolution, LLM mechanistic interpretability, and cognitive neuroscience. In so doing, we hope to reconnect modern AI with longstanding research directions, debates, and promises surrounding the structure of cognitive processes 

We aim to make this event accessible and productive for all attendees. Please notify us of any specific needs or preferences at ResponsibleNLP@ed.ac.uk regarding accessibility to ensure we can provide the necessary arrangements. We look forward to your participation and are confident that your insights will greatly benefit our discussions and help shape the future direction of our joint programmes.    

Detailed information about the accessibility of our venues can be found on the AccessAble website.    The Canopy Kitchen & Courtyard, including toilets for the Café, is fully accessible and open until 9 pm. More detailed information is  here.  

Travelling to Edinburgh from abroad can be a smooth and straightforward experience thanks to the city’s well-connected transport services.     

Edinburgh Airport (EDI) is the primary international gateway to the city. Located just 8 miles west of the city centre, it boasts a wide range of direct flights from cities in Europe, North America, and other international destinations. Alternatively, you might consider flying into Glasgow International Airport (GLA), which is about 50 miles west of Edinburgh, and offers additional routes which may be more convenient or economical.    

In case you wish to extend your visit in Edinburgh, we recommend you book your stay in advance (the University can offer a small discount, so please write to us if that is of interest). Areas like the Old Town, New Town, or near the Royal Mile are great for tourists due to their proximity to major attractions and transport links.    

Arrival and getting to the City from Edinburgh Airport    

The University of Edinburgh has a specific page t how to get to the city of Edinburgh, and details of how to navigate the city and find University buildings: How to get here | The University of Edinburgh     

In short:    

• By tram: The Edinburgh Trams run directly from the airport to the city centre, passing through major points including Haymarket, Princes Street, and St Andrew Square. The journey takes about 35 minutes. Please visit link for a timetable, tickets and a journey planner: Route Map and Stops | Edinburgh Trams.    

• By bus: The Airlink 100 express bus service runs 24/7 from the airport to Waverley Bridge (near Waverley train station) and is a cost-effective option.    

• By taxi: Taxis are available at the airport 24/7. A taxi ride to the city centre takes about 20-30 minutes depending on traffic.     

Entry to UK from abroad  

To travel to the UK for a short visit, you might need to apply for an Electronic Travel Authorisation (ETA). The process is fully online, and very easy. Please see here for full guidance here Apply for an electronic travel authorisation (ETA) – GOV.UK     

Edinburgh Transportation, if you wish to visit attractions 

The Visit Scotland website has a useful section of travel and things to do in Edinburgh, so please consider it if you are staying for 1-2 days Edinburgh – City Centre Visitor Guide – Accommodation, Things To Do & More | VisitScotland    

In short:    

Edinburgh has an extensive public transport network of buses and trams (plus reputable hop-on and hop-off city sightseeing bus tours).  However, Edinburgh’s city centre is compact and quite walkable unless you have mobility challenges, with many of the main tourist sites within easy walking distance of each other.   Edinburgh also supports cycling, with several designated bike paths and bike rental services available.    

Health and Insurance 

Always travel with valid travel insurance that covers health, accidents, and theft.    

Tourists in the UK can receive emergency healthcare services, but  it’s advisable to check what conditions apply, especially post-Brexit, for non-UK residents.    

We recommend this free self-guided tour along the Royal Mile (the Royal Geographical Society website which hosts it additionally has guides for other major landmarks around the city): 

Edinburgh Royal Mile walk: Edinburgh castle to Scottish Parliament Building 

The Castle, where this starts, is about a 20-minute walk from the Edinburgh Futures Institute and is pleasant as well.  Upon conclusion of the walk down the Royal Mile you’re deposited near Holyrood Park and Arthurs Seat which is a delightful, natural space to explore (has its own guide too linked from the website above). 

Some free audio tours of the Royal Mile (starting from the Castle) can be found here too: 

Our Tours | Explore Edinburgh Now — Discover & Walk — Edinburgh ByFoot Self Guided Walking Tours 

New Free Virtual Audio Tours – 7 Hills Tour Company | Edinburgh 

This Summit is organised by the UK Research and Innovation (UKRI) AI Centres for Doctoral Training in Natural Language Processing, dedicated to promoting research and collaboration in the field of NLP.  Our organisers can be recognised by their red lanyards. We invite you to reach out to us in case of any queries or to give feedback during or after the Summit.