Ethic stakes in the use of AI. An outlook from the African continent.
Coded Bias by Joy Buolamwini, directed by Shalini Kantayya
Digital artistic creation has followed the development and the major evolutionary phases of AI, from the algorithm (Aaron program) to conceptual intelligence (CAN) passing by deep learning (GAN). Artistic work that integrates artificial intelligence requires multiple interactions and involves several actors at the time of conception but also after the creation. This leads to philosophising about the concepts of work of art and intellectual property or authorship and to questioning the links between Man and Machine.
In ten years, artificial intelligence has considerably modified creative potential until she became capable of generating original creations, whether musical, architectural, pictorial or even written. Is the artist becoming the curator of AI? While a few years ago we were still wondering whether the curator should sign an exhibition or not, what about the artist who would be the curator of an AI if he or she does not sign? It is coming back to the question of ownership, but also to the question of authorship, since we have to take into account the person at the source of the logarithm, who is a human being inspired by his own environment.
The myth of the artist becoming "a machine harnessed to another machine" or the redundant science fiction fantasy of the machine taking control over humans is now obsolete and it prevents us from addressing the real problems induced by the untimely deployment of automated decision that affects all aspects of our daily lives, through questions of gender, race and class. But machines are not as intelligent as they seem to be, and if they are dangerous it is not because they might replace humans or take their jobs, but because this fear masks other issues.
The first problem concerns the work carried out in so-called 'click farms', by underpaid and unprotected workers in the South countries who are asked to carry out ultra-repetitive micro-actions produced in very short times and which are so mechanical that they engender the perpetuation of fallible paradigm, which potentially include errors or biases. These biases, from stereotypes to discrimination, will send out data that will be re-used and will generate other discriminations by endlessly repeating the flaws of the human model...
At the source of the machine lies a second problem: cognitive biases that can be unconsciously induced by its designer, himself being the product of a culture and a way of thinking. As well as the Internet, which, despite its original social claims to connect the world and cultures, remains the product of a very specific and localised culture (the one of Silicon Valley) and propagates a form of paradigm; the person at the source of an AI artwork conceives and therefore expresses himself from a given place at a given time. Joy Buolamwini's research, presented in the film "Coded Bias", directed by Shalini Katanyya, sheds light on the discriminatory and racist biases that she personally experienced while working on the theme of facial recognition. Indeed, her black face was not recognised by the system, which nevertheless took her into account when she wore a white mask.
It turned out that these (facial recognition) algorithms perform better on the male faces than the female faces.They perform significantly better ont the lighter faces than the darker faces. - Joy Buloamwini -
Coded Bias
In September 2020, the French collective Obvious produced Facets of AGI, a new artwork composed of a series of African masks created with artificial intelligence and more specifically, the GAN technique. A "Generative Adversarial Network" (GAN) is a Machine Learning technique that relies on the competition of two networks, "generator" and "discriminator", in the same program. One part of the algorithm is trained to recognise thousands of African mask patterns from databases to generate or produce visuals of new masks, and then the other part of the algorithm will compare these new masks with those in the database to judge the authenticity of the object and determine whether or not it is part of the dataset.
Obvious then selected a Ghanaian sculptor, Abdul Aziz, to produce the masks. The algorithmically created series was praised in two press articles and nobody raised any questions about the creation process. Ancestral knowledge from the African continent is copied and used to recreate, in another territory, models inspired by them, which will be returned to their territory of origin to be produced before being given a name in an African language from another culture linked to a third territory that has absolutely nothing to do with the other two. The creators make no secret of the fact: "We have invented a universe where an artificial intelligence would take the place of traditional deities and be the object of a cult...". In addition to statements bordering demiurgic fantasy, the question of cultural appropriation seems to be on the agenda, especially if we take into account the territories and actors involved. Let us remember, for this purpose, Eric Fassin's definition of cultural appropriation: "cultural appropriation is when a loan between cultures is inscribed in a context of domination". In this case, it is important to remind that the collective also sold another work made by an artificial intelligence for $432,500 at Christie's, New York, in 2018.
A fourth issue can be raised about "data mining", more commonly known as big data. To train a neural network (deep learning), it will be given a multitude of examples and, depending on the difference between the result obtained and the expected result, the system will update its coefficients. These numbers of data on which the deep learning technique feeds are literally "excavated" before being transformed into useful information by establishing correlations between relational databases. In addition to the questions of data protection, there are also the questions of sorting this data with the risk of "cultural netflixisation". Indeed, in an interactionist vision of the circuit of an artwork, if the logarithm at the source of the proposals is already faulty, what about the consumers' analysis and then, what about the contents' proposals that will result from it... The popularity of Netflix is not only due to the novelty of a mode of dissemination of culture, but also to the control of the chain from production to reception, in this case closer to consumption.
On 11 March 2021, the digital work "Everydays: the First 5,000 Days" by the American artist Beeple was sold for 69.3 million dollars by the auction house Christie's, a record in this new market. This sale was made possible by a new authentication technology that uses the "blockchain" (cryptocurrencies) to detect copies and allows works to be marketed in the form of an NFT token (non-fungible): "any virtual object with a theoretically incontestable and inviolable identity, authenticity and traceability". Culture is also an issue of sovereignty which can be seen in particular in the competition from the GAFAs (Google, Apple, Facebook, Amazon).
During the Agora 3 organized organised by the senegalese multimedia centre Ker Thiossane for the 8th edition of Afropixel Festival -Power to the Commons, researcher and curator Tegan Bistow came back to the importance and challenges of data sharing, through the Masakhane project, whose mission is to strengthen and stimulate NLP researches in African languages, for Africans, by Africans. Despite the fact that 2000 of the world's languages are African, these languages are poorly represented in technology. Colonialism has obviously not helped in the preservation and integration of these African languages. But also the value given nowadays to ancestral or vernacular knowledges. The result is a technological space that does not understand African words, languages and therefore cultures and their history. Her research on vernacular knowledge is in the same vein and raises awareness of the importance of its preservation, particularly by emphasising the need for financial interests in intangible heritage. Data sharing therefore becomes an important stake. Beyond technology, the machine can become a real partner for the cultural actors or artists to encourage awareness of these issues.
From facial recognition to data processing, through questions of intellectual property, work automation, extractivism, cognitive biases, cultural appropriation or instances of domination, artificial intelligence raises major social debate points that artists question. Art therefore puts the question of AI's appropriation at the centre and insists on the need to use it as a tool for social co-construction, and it can open our eyes to the ethical problems posed by its development by denouncing the societal drifts of its use. It is in this sense that it is essential, because it continues the work of deconstructing the paradigms of AI and other technologies that are part of our daily lives by returning to the simple but essential questions: who is talking, to who, to say what and from where?