In this seminar we heard from three speakers, a double header from the British Library/University of Leicester team of Paul Merchant and Sally Horrocks, reflecting on the results of their ongoing oral history collecting as part of National Life Stories, and Sarah Dillon on her ongoing work at the intersections of literature, science and artificial intelligence. This post is intended to capture some of their key points and the discussion that followed but is in no sense comprehensive. As with all of our seminars an audio recording is available for those who write to me directly.
Sally Horrocks, University of Leicester - and Paul Merchant, British Library
Scientists’ narratives in An Oral History of British Science
To begin with Horrocks provided a brief overview of the Oral History of British Science project, which is dedicated to ensuring that science and engineering are better represented in the broader National Life Stories project. They are continuing to find ways to fund further oral history collecting, to ensure the widest possible range of sciences and scientific lives are recorded. Despite the extent and breadth of their materials, they often find that historians of science struggle to make use of them, perhaps due to unfamiliarity with the method. Her talk, and Merchant’s, were designed to make their value and usefulness easier to appreciate.
The Life Story method emphasises a biographical approach, rather than rushing toward a person’s education and scientific career. Their interviewees usually sit for multiple sessions, allowing for deeper and more open engagement, drawing out elements that a typical oral history of science skips over. Likewise the project tends to prioritise people who have never been interviewed at length before. Alessandro Portelli’s work in oral history was recognised as a point of inspiration, as it is for many oral historians. One of the first findings that Horrocks shared was that scientists tend to tell stories of their childhoods that emphasise certain standard patterns or narrative features: playing with Meccano, dangerous experiments at home, a chemistry kit acquired at a young age, etc. She then pointed to the oral historians’ theory of narrative and psychic composure, explaining the feelings of lack experienced by scientists without such experiences to draw on.
Putting the work of an Oral History of British Science in light of the Narrative Science project, Horrocks felt that insufficient attention had yet been given to the huge amount of material in interviews on the production of scientific knowledge itself. At this point she handed over to Paul Merchant.
Merchant wanted to flag some potential examples of narrative knowledge recorded in their interviews. To begin with, he pointed to the work of the British Antarctic Survey and its painstaking efforts to measure the condition of the ozone layer from 1956, using ground-based instruments at Antarctic research stations. Merchant argued that the scientist in charge of these measurements – Joseph Farman – found it difficult to motivate his team and secure ongoing funding in the absence of an exciting narrative (this all changed when the ozone hole was discovered in the early 1980s).
A second example concerned the importance of narrative for making sense of heterogeneous scientific evidence. Merchant pointed to the work of one interviewee – Richard West - completed in retirement, to produce a geological history of part of North Norfolk. West used very varied evidence, including old maps, conversations with farm workers, aerial photographs given to him by a tractor driver who happens to fly a glider at weekends, local guides, and measurements of sediments taken from his own collecting expeditions. West’s efforts to create a natural history of a puzzling geological situation involved the construction of a narrative that would draw together such materials in a coherent and satisfying way. West himself comments on the satisfactions of the story composed. Merchant’s third example returned to measurements of the ozone layer above Antarctica but now in the early 1980s, when dramatic ozone depletion had been recorded. This time his attention was directed to the ways in which data on ozone depletion were presented, especially the kinds of graphical innovation and labour that were regarded as necessary for a convincing narrative in the journal ‘Nature’.
In the question period one participant asked whether there are differences between the kinds of story told by different kinds of scientist. Horrocks pointed to a clear difference between scientists and engineers, the latter rarely having a pattern story to tell. The cultural narratives of scientific childhood, though appealed to by a wide range of scientists, were typically off the map for engineers. Another participant asked whether the material can speak beyond the ‘British’ sense of an ‘Oral History of British Science’? Horrocks and Merchant clarified that though they had to focus on people who had spent most of their career in the UK, this did not mean that only had to speak to people who were originally UK nationals.
Unfortunately at this point the dictaphone stopped recording because it was full! The further questions and discussion are therefore lost to us, providing a visceral reminder of the value of the work oral historians do.
Sarah Dillon, University of Cambridge
“The Ineradicable Eliza Effect and Its Dangers”: Weizenbaum, Pygmalion and the Implications of Gendering AI
One of Dillon’s earlier research starting points had focussed on patterns in the literary habits of scientists engaged in work on artificial intelligence (AI), looking for common authors, themes, etc. This led to a range of categories covering the broad ways in which AI researchers made use of their leisure reading, including the making of an identity and research community. She felt there was potentially an overlap between the kinds of identity-making highlighted by Horrocks and Merchant (pattern stories and composure) and findings from sociological and literary analysis. Dillon suggested that the kinds of leisure reading engaged in by scientists can be understood as contributing to their narrative knowledge (their internalised repertoire of narrative possibilities). The latter seemed to her to lay more in the direction of the Narrative Science project. In the context of these wider thoughts, Dillon’s paper sought to follow the deep connections between literary sources and challenges faced by AI research over the course of the twentieth century, using specific pairings between literary texts and scientific texts and figures in AI. These could include figures such as, Alan Turing and Samuel Butler, Daniel Dennett and Richard Powers, and Marvin Minsky and Harry Harrison.
At present these particular cases aren’t ready for presentation, so instead she focused on Joseph Weizenbaum and George Bernard Shaw, particularly the former’s ‘Eliza effect’ (in which people come to respond to AI interlocutors as though they are human) and Shaw’s Pygmalion (in which a lower class woman is taught to pass as an upper class lady). Her interests concern, amongst other things, the gendered nature of the relationships evoked by these literary-scientific implications, and the kinds of social and epistemic results they bring about. Weizenbaum’s interpretation of the Eliza effect, and his reading of it as a problem, can be seen as a particular instantiation of a broader political problem, namely that of where power lies and to what we should recognise as human. Shaw’s play is full to the brim with this kind of commentary, Eliza falling in and out of different and competing definitions of humanity, or at least definitions of humanity worth regarding.
Later in the play, as Eliza learns more and more, she enters into the position of something like a personal assistant (before, that is, eventually taking off). This is significant given the emergence of a new norm in AI systems in the twentieth and twenty-first centuries, making software adopt the name, persona, and voice of a woman. Recognising the back and forth between the literary and the scientific opens up new opportunities for critique, making very plain how the same kinds of critique and commentary present in Pygmalion can be applied doubly to Alexa, Cortana, and all the other female gendered AI personal assistants that, in being so gendered, serve to replicate the power relations and othering of women.
In the question period we asked for further insight into the relationship between the fictional and ‘factual’ narratives being explored here. Dillon is interested in making the strong argument that in as much as these literary narratives depend on and produce particular ontologies, ontologies that scientists are likewise guided by and guiding us towards, they are contributing to the world-building of science just as published journal articles do. Another participant pointed to an overlap between this talk and one given in the previous seminar, by Heike Hartung. Both Hartung and Dillon had paired George Bernard Shaw with particular scientific figures, so as to provide a vehicle for deeper analysis of the literary and scientific at once. Dillon explained that one of her next steps will be to look in more depth at Shaw’s wider output, to see what it might teach us about understandings of intelligence, just as we had seen in the previous seminar but there with attention to understandings of ageing.
The next seminar took place on the 28th of May and featured Emily Hayes and Dominic Berry (me). A blog post will follow soon. Also keep your eyes peeled for our first working papers, available for download on our website and circulated to members of our mailing list. We also recently hosted a variety of workshops concerning Time (30-31 May), Polyphony (3rd June), and Environment (18-19 July), about which further information will be shared soon. Remember if you want to keep up-to-date, just join our mailing list from the website home page.