Texto em livro: Artountabilzty: Art and Algorithmic Accountability
de: PETER BOOTHI, LUCAS EVERSZ, EDUARD FOSCH VILLARONGA3, CHRISTOPH LUTZ, FIONA MCDERMOTT5, PIERA RICCIO6, VINCENT RIOUX7, ALAN M SEARSS, AURELIA TAMO-LARRIEUX9 e MARANKE WIERINGA
Given the complexity of the inner working of algorithms and the ulterior effects these systems may have on society, the European Union has begun an ‘Algorithmic Awareness-Building exercise to inform policy-making on algorithmic decisions’ challenges and opportunities. We contribute to this effort by identifying how art can have a strong voice in promoting algorithmic accountability and transparency in the public debate. After introducing algorithmic accountability and transparency concepts, we focus on the cognitive, affective, societal, educational, and ethical functions art can have in realising Europe’s goals.
Discussion and Conclusion:
Individuals’ characteristics and preferences are used to build profiles and combined with automated decision-making to make products and services more precise and effective. While there may be some benefits, algorithmic profiling and automated decision-making processes can impact individuals and society at large, for example, by invading individual privacy or rising inequality. Given the complicated processes in such algorithmic systems and the existing legal framework, there may be a bit of a responsibility gap. Increasing transparency, embedding privacy into the technologies, and raising awareness are commonly proposed to increase accountability in this area. The European Union, for instance, has begun an ‘Algorithmic Awareness-Building’ exercise to inform policy-making on algorithmic decisions, challenges and opportunities. An often overlooked area that could nevertheless contribute is the role of art in this arena. This chapter thus explored how art can have a strong voice in promoting algorithmic accountability and transparency in the public debate.
After introducing the intersection of algorithmic accountability and art, we then elaborated upon the concept of algorithmic accountability and what exactly it constitutes. Algorithmic accountability needs to follow the system’s life cycle, where there are ex-ante, in medias res, and ex-post considerations that should be taken into account. Accountability, wherever it is located in a system’s life cycle, is facilitated and operationalised through other principles such as traceability, reviewability, fairness, and the more abstract principle of transparency. The present contribution in assessing the role of art regarding algorithmic accountability focuses on the more abstract level of transparency.
We then analysed the concept of transparency and its multiple dimensions. Transparency as information makes underlying operations perceivable and detectable, which entails the idea of verifiability and the notion of explainability and inspectability. While this understanding of transparency enables accountability and auditability, this understanding should be embedded within the other dimensions of transparency. One such dimension is performative transparency, Where organisations perform transparency by outlining how it processes data, which it can do in a transparent or opaque manner. Another critical dimension of transparency related to the performativity element is the relational dimension ablllty tocuses on the more abstract level or transparency.
We then analysed the concept of transparency and its multiple dimensions. Transparency as information makes underlying operations perceivable and detectable, which entails the idea of verifiability and the notion of explainability and inspectability. While this understanding of transparency enables accountability and auditability, this understanding should be embedded within the other dimensions of transparency. One such dimension is performative transparency, where organisations perform transparency by outlining how it processes data, which it can do in a transparent or opaque manner. Another critical dimension of transparency related to the performativity element is the relational dimension of transparency. Only with the audience in mind, ie, the information provided is targeted towards the relevant audience, can transparency be effectively performed and lead to accountability. Finally, transparency is embedded within a legal, regulatory, and organisational context. While the GDPR mandates transparency and accountability, a ‘transparency by design’ approach may help in addressing shortcomings.
While attempts to increase the legibility of algorithmic decision-making to bring about accountability are critical, it is also essential to acknowledge the limits of transparency as a responsive action. As Ananny and Crawford have noted concerning calls for transparency around algorithmic decision-making systems, transparency on its own will not mean users can change or influence these systems, as transparency can promote seeing without knowing.
Next, the chapter delved into the role of art in promoting algorithmic accountability and transparency. The five functions of art — cognitive, emotional, social, educational, and ethical — were examined. The cognitive function of art in promot- ing algorithmic accountability and transparency lies in its potential to foster awareness, evoke thought, and challenge existing narratives that may be harmful or considered inappropriate in contemporary times. An important project in this area is the MoRM, which challenges participants to think about how AI systems operate critically. Art also has an emotional or aflective function. An AI produces artwork that engages with AI systems or that can, for example, foster empathy and allow individuals to feel similar to someone affected by an AI system.
In this sense, art broadens perspectives and can potentially create transparency in a more grounded, personal, and relational manner. Also, art has a social function by bringing together different stakeholders around a topic and creating novel communities. By doing so, art might interest new audiences in Al, transparency, and accountability or engage people emotionally. As such, the social function is connected to the other functions of art. The educational function of art fosters the artist’s personal development and a deeper understanding of transparency, enabling alternative design visions that can hold different parties to account and create communities of practice. The ethical function of art addresses what it means to be subject to what algorithms can understand what is correct or not, and more importantly, whether how algorithms function is ethically justifiable or not.
Art has the power to explicate power dynamics and raising awareness of societal issues and could be determinant in shaping a collective and societal understanding of AI societal consequences, increasing the throughput from transparency to accountability. Art can also bring many different stakeholders together, from young audiences to engineers, to policymakers, to vulnerable groups, subject to the discriminatory effect of such systems. Because of the multiplicity of stakeholders and art projects, artwork in general ‘works’ on several levels. First, by making use of the AI technologies, engineers, and the industry use. Second, creating meaning and understanding about the ambiguous, complex, and controversial elements found in the way AI in context is developed and works. Third, by offering itself as an object to reflect upon, being present as a way to question without answering
(being as art should be useful in its uselessness).