宝暴露 [bǎo bàolù- ’treasure exposed'] aims to both expose and hack certain facets of AI-powered Consumer Predictive Analysis on TaoBao (chinese online shopping platform), such as classifiers, clustering algorithms, and recommender systems. Along the way, it surfaces the semantic spaces conveyed by TaoBao data.
A significant part of Machine Learning algorithms is currently used to model human beings, to be able to predict either their use of infrastructures, their energy consumption patterns, their political views, or which movie they are likely to buy, and their consumer behavior. Consumer Predictive Analysis has reshaped the marketing industry recently, boosted by the explosion of Big Data and smart bussinesses. The data we produce online is often distributed between various parties interested in knowing more about us. The data we produce online -or trade for services- is often distributed between various parties interested in knowing more about us, and extends way beyond our purchase history. Moreover, new insights are then generated as machine learning algorithms help speculate about our personalities, physical or psychological states, aspirations, among others. These databases of intentions and profiles help entities such as advertisers or corporations target individuals or groups with tailored products and services. A common issue with these types of predictive algorithms, comes from fact that predicting human behavior is not like predicting far distant stars trajectories, in the sense they can -and do- affect our behaviors.
A first workshop ‘Shoppers Exposed’, was held at Chronus Art Center (Shanghai) in September 2019. We looked into artificial intelligence as means to profile us based on individual consumer patterns, by exploring online purchase histories on TaoBao. After raising the human biases and choices underlying algorithmic classification and recommendations algorithms, we aimed to collectively reappropriate these taxonomies and recommender systems in non-normative and transgressive ways: first by redefining labels, and train a new collective classifier, then by enacting a recommender system. For this, we used collage and rapid prototyping to recommend new—even potentially absurd, useless or extravagant—hybrids of devices or services to a specific labelled customer -confused artist or expansive housewife-, in a world of overabundant commodities. We discussed how easy, hard, objective, biased, or ethically-challenging it may be to deploy algorithmic assessment on our digital profiles. We reflected on the data we produce, the impact as well as the limits of statistical stereotypes attributed to our consumption.
Workshop by Vytautas Jankauskas & Claire Glanois
Code & Jupyter notebook written by Claire Glanois, available online.