UK Pavilion at Expo 2020 Dubai

Shaped like a giant wooden conical musical instrument, the Poem Pavilion gathers words donated by each visitor and uses an advanced machine learning algorithm to generate the cumulative collective poem which illuminates its twenty-metre diameter facade.

It is the first time a UK pavilion has been designed by a woman since the inception of Expo in 1851. Devlin conceived the building to express Britain as a cultural gathering place, a meeting and melding of ideas and languages from across the globe. Twenty five million visitors are projected to pass through Expo during its six month run, and each will be invited to donate a word at the ‘mouthpiece’ of the pavilion, then enter within the heart of the instrument where they will be surrounded by donated words glimmering in illuminated Arabic and English, underscored by a soundscape gathered from multicultural choirs across the UK. As visitors emerge through the facade of the pavilion, they will pass through the twenty-metre diameter composition of collective text: a new poem generated every minute.

Devlin collaborated closely on the designs with structural engineers, Atelier One, environmental design consultants Atelier Ten, executive architects Veretec and creative agency Avantgarde. Guided by the engineers’ expertise in sustainable construction, the team chose cross laminated timber as the pavilion’s prime material. Sourced from sustainably managed European forests in Austria and Italy and championed by engineers as a sustainable alternative to concrete and steel, the cross laminated timber is crafted around LED tiles engineered in Belgium and manufactured in China, installed by local British and UAE teams, while the poetry generating algorithm has been developed by creative technologists in California. The pavilion has been conceived as an expression of cultural inclusivity and its execution is a feat of European and international collaboration.

Devlin’s exploration of machine generated poetry began with PoemPortraits at the Serpentine Gallery in 2016 in response to curator Hans Ulrich Obrist’s invitation to consider the idea of a ‘social sculpture’. In 2017 she continued her explorations at the Victoria & Albert Museum, turning their annual artist-conceived Christmas tree into a ‘collective carol’. In 2018 visitors fed words into the mouth of one of the lions in Trafalgar Square and watched the collective text projected up the length of Nelsons column.

The text generated by the Poem Pavilion uses a machine learning model called GPT-2, a large language model defined by 1.5 billion parameters. GPT-2 was originally trained on a broad spectrum of internet text, and for this project it was fine-tuned on a diverse and carefully curated selection of over five thousand poems—comprising over two hundred thousand lines of poetry refined over months of iterative feedback from a diverse team of poetry curators.