‘The Ineradicable Eliza Effect and Its Dangers’: Weizenbaum, Pygmalion and the implications of gendering AI – Sarah Dillon (Faculty of English)

November 14, 2019 @ 1:00 pm – 2:00 pm
Seminar Room 2
Department of History and Philosophy of Science
Richard Staley

Preface 4 of Douglas Hofstadter’s _Fluid Concepts and Creative Analogies: Computer Models and the Fundamental Mechanisms of Thought_ (1995) is entitled ‘The Ineradicable Eliza Effect and Its Dangers’. Hofstadter defines the Eliza effect as an ‘illusion’, ‘which could be defined as the susceptibility of people to read far more understanding than is warranted into strings of symbols – especially words – strung together by computers’ (157). More widely, the Eliza effect in computer science names our tendency to unconsciously assume that computer behaviours are analogous to human behaviours, with a consequent effect on our perception of their ontological status. Hofstadter considers this dangerous in its effects because it misrepresents the capacities and capabilities of the research, and the technologies it creates. ‘The operational term here is’, he says, ‘hype’, but with an interesting caveat, ‘and yet it is’, he repeatedly says, ‘inadvertent’ (167). He acknowledges that it benefits the researchers, but he describes it as merely an ‘overly charitable way of characterizing what has happened’ (157). For Hofstadter, the Eliza effect is not mal-intentioned, but ‘like a tenacious virus that constantly mutates’, he says, it ‘seems to crop up over and over again in AI in ever-fresh disguises, and in subtler and subtler forms’ (158). Hofstadter identifies this phenomenon, but he is doing so as a scientist, in relation to its consequence for scientists and scientific research. What he does not do is think about the social and ethical consequences of the Eliza effect, and about the role of rhetoric in triggering it. In this paper, I explore the Eliza effect in this regard, from a feminist and a literary perspective. The Eliza effect gets its name from the responses to Joseph Weizenbaum’s first natural language processing software, ELIZA, which he named after the heroine of George Bernard Shaw’s play _Pygmalion_ (1913). Understanding ELIZA’s historical and literary origin stories highlights the role of gendering in triggering the Eliza effect, and its feminist dangers. This literary historical case-study can then inform contemporary debate regarding, for instance, the societal harm of the gendering of virtual personal assistants, in particular in relation to such social consequences as the objectification of women, and the replication of gendered models of power and subservience. More broadly, the paper demonstrates the role that literary narratives play in shaping the development, reception and impact of science and technology.