Recap: FEMINIST MACHINES
A lecture evening on art, AI, data sets & bias
On December 12th, Creative Coding Utrecht organized with the support of Utrecht University, Het Utrechts Archief, HKU, Creative Industries Fund NL, a lecture evening called Feminist Machines. The speakers Peter van der Putten, Rick Companje, Caroline Sinders, and Hannah Davis gave lectures around the topics of data, datasets, AI and Machine Learning. The talks gave insights and reflected on where and how these technologies are situated socially and historically, while providing various ways on how to critically think about their stance. The speakers gave several examples that critically and creatively engage with these technologies in order to make them more accessible, open, and understandable. The lectures took place at Het Utrechts Archief, an old court of justice, now housing an exhibition centre. The old purpose of the location carried a symbolic meaning to the lecture night in which the aim was to question technologies of today and their social impacts while suggesting ways to improve their use.
The Fascination with AI
The first speaker of the night was Peter van der Putten, an assistant professor at Leiden University, with a background in cognitive artificial intelligence and data mining. Van der Putten started his talk by explaining what Artificial Intelligence is to dismantle the mysteries around it. He started his talk by asking, “Why are we fascinated with AI, and where does this fascination come from?” To tackle these questions, Van der Putten wanted to place AI in a societal context by giving some examples. Van der Putten thinks that our fascination with AI derives from our fascination with ourselves. He uses the mirror as a metaphor to explain this fascination. Artificial Intelligence, according to him, gives us a reflection of ourselves; it tells us more about who we are.
We are driven with the curiosity to understand ourselves better with the questions such as “Where does our intelligence come from, how do we learn, what makes us human?” This questioning is central to our thinking and has been an interest of ours for a long time. To exemplify this, Van der Putten discussed the work of French philosopher Julien Offray de La Mettrie, the author of the Man and the Animal, and Man a Machine. According to Van der Putten, rather than merely arguing that “man is a machine,” Julien Offray de La Mettrie, in his book, was interested in looking at how the metaphor of a machine can help us to understand how we function as humans. Another example Van der Putten put forward to highlight our interest between the questions of intelligence and machines was Ada Lovelace, who wrote the first computer program long before the computer was invented. Occupied with the notions of creativity and intelligence, she predicted that the machines of the future will be able to play music, compose symphonies. Various scenarios around machines and their intelligence were present back then as is the case with the utopian and dystopian scenarios emerging around technologies now.
Another prominent figure in the analysis of the intelligence of the machines is Alan Turing. With his famous Turing test, he fostered thinking about the intelligence of machines, and he thought about the development of intelligent systems. In this line, Van der Putten explains that for the development of intelligent systems, relying mainly on rules, logic, and reasoning would not be enough since it does not spare any room for unpredictable behavior and learning. Learning is an essential aspect of intelligibility. This point bridges the discussion to the idea of neural networks, systems that are ideated based on the functioning of the brain, pioneered by Warren Sturgis McCulloch around 1943. The discussion on the neural network models based loosely on how the brain works, followed with a focus on the social implications of AI. Van der Putten asserts that considering only the utopian or dystopian views around Artificial Intelligence may leave us blind to what is happening around these technologies.
However, focusing on questions that concern the social impacts of the design of technologies might benefit a more significant portion of society. We might speak of these technologies as part of a dystopian future scenario, yet this misses an essential point of considering their current societal impacts. Reviewing current and potential uses of AI and critically engaging with the issues of bias have an important place in this. In this regard, Van der Putten touched upon the issues of bias in datasets, and he asserts: “It is not necessarily the algorithms or machines that are bad, but data might be biased. We need to understand the methods of AI and make our own, in creative and playful ways.”
The talks of Caroline Sinders and Hannah Davis were connected to this suggestion and further provided with various ideas for creative, artistic ways to engage critically with these technologies.
The application of AI by Het Utrechts Archief
The second speaker of the night was Rick Companje. Companje is a creative coder and has a background in computer science and Media Technology at Leiden University. Currently, he works at Het Utrecht Archief. Within the archive, a large compile of handwritten documents is stored and making use of AI; the handwritten letters are analyzed and converted into computers to be stored. Companje described the process of transcribing the handwritten letters through the use of AI. The computer was used to transfer the handwritten letters automatically, and Het Utrecht Archief’s The Rampjaar 1672–1673 exhibition makes the techniques that were used for the transcription of hand-written letters visible for the audience.
Feminist DataSet
Caroline Sinders is an artist, designer, and researcher. In her work, she is interested in looking at and understanding how people engage with and make use of various systems and platforms. In the last few years, Caroline Sinders has been very involved with the question of how to make AI more accessible to people. As a design researcher, she thinks that this is possible through design, especially considering its impacts on products and eventually on society. She uses design as a broader term to think about Artificial Intelligence, Machine Learning, Machine Intelligence, and the processes that lead to the development of systems, which often include the steps of planning, focusing on structure, intentionality. Referencing Mimi Onuoha and Mother Cyborg from their book The people’s guide to AI, she uses the metaphor of salt to describe the functioning of AI. Artificial Intelligence is not a product itself (as salt is not a meal itself) with other ingredients, code, algorithms, data, it becomes transformative for the design of the products that we engage with on a daily basis. Based on this, Caroline Sinders argues that data is as necessary as code or algorithms since data is used for training the algorithms.
“Data is all human outputs, Data is all people, even data that can feel extremely mechanical like shipping statistics, it still comes from people. There’s nothing cold or mechanical about data. It’s all people and it is all inherently human. We have to treat data as sensitive objects, as a precious material.”
Caroline Sinders showed several examples of artworks that engage critically with data, one of which is Mimi Onuoha’s “The Library of Missing Datasets.” The work provides some interesting insights into the issues of bias embedded within the datasets. In her project, Mimi Onuoha aimed to find out about the bias among society by spotting the gaps or missing data in datasets. Onuoha, working with community groups, focused on the casting issues on Broadway shows. She mapped different genders and races that were cast into roles. Her findings pointed out to a big pile of missing data and showed a leaning towards a casting of more white actors into roles. Another example from the Google search engine shows how issues of bias are present within the everyday products we use. For the search engine, Google makes use of machine learning by training data from data-sets. If we get to the bottom of it, we should pay attention to data and consider what kind of data-sets these companies or the designers, developers of our everyday technologies use, and question the intentionalities behind their preference to use these datasets as Caroline Sinders asserts.
These examples point out a significant relationship between social bias and data usage and may have substantial negative consequences, especially for minorities when technologies are used without . The examples Sinders highlight shows the importance of considering thoroughly the specific qualities within data and data sets that are used in everyday systems and products. How can we think about the various ways of reducing such biases in these datasets or create technologies that won’t end up being extremely harmful to certain parts of the society? In this regard, Sinders asks: “Is there a way to apply harm reduction to AI? How can we make AI more accessible?” With her multi-year, long project “Feminist Dataset” Sinders, works around such questions after her studies on online harassment. With Feminist Dataset workshops, she is aiming to collect and archive various Feminist materials such as feminist songs, song lyrics, podcast transcripts, interviews, essays. The project “Feminist Dataset” eventually aims to come up with a feminist chat interface.
Subjective Data, Music Composition and Machine Learning
The last speaker of the night, Hannah Davis, is an artist, researcher, and composer. Her work revolves around music generation, machine learning, natural language processing with a focus on subjective data sets, ethics, and AI. Her interest in subjective and emotional data derived from a data set she encountered about the emotional response of the voters toward the candidates after elections in the US. For a decade, Hannah Davis tracked the emotional situation of the voters, wondering whether they were representative of the actual emotions that the voters felt. Gathering this data, she visualized emotions according to some facial features, and the results gave an idea on the change of feelings in time, depending on the change in presidential candidates over the years.
In another project called Transpose, Hannah Davis focused on the translation of the text into music. Parallelly, she was interested in the exploration of emotional tone underlying certain novels, in other words, translating emotions across different mediums and by mapping prevalent emotions of novels with various components of music. Her interest in biased data and subjective data started when she came across a data-set on childbirth. The data-set she came across classified childbirth as a situation that has no emotions involved. This finding led Davis to ask questions about the behind scenes of preparation of the data sets. “Who is involved in the creation of the data sets? Who are tagging the data, and importantly, what worldview datasets are creating?” These are questions that we should keep in mind when thinking about the datasets.
She argues that datasets shape worldviews of those who are engaged with the data. Hence, how the data organizers, the companies create taxonomies and use data in various phases of machine learning is of great significance. Through subjective data, Davis argues that we have the power to shape these worldviews. Subjective data, according to Hannah Davis, means accepting that the datasets carry certain values and worldviews, data is not neutral nor objective. Additionally, when looking at datasets, we need to consider points of representation that are seen as outliers and that are left out.
“Machine learning has a tendency for the average leave out the outliers. However, outliers are where the beauty lies in the world. When algorithms are implemented in large scale situations, or situations with real-world consequences we risk alienating and harming those who are not appropriately represented.”
When constructing subjective data-sets, we should aim to look at the different components of the dataset, features of datasets such as the information on who created the data set, how the data-set was funded, by whom. The datasets reflect certain values or biases that are embedded within the society, depending on the conditions of the time. Concerning this, Hannah Davis argues that data-sets should be renewable or need expiration dates if there are no reasons to hold on to past social values. Data-sets can be used to create better worldviews and subjective data; focusing on taxonomies could help construct better worldviews and face the problematic parts of machine learning. Importantly she underlines the importance of taking into account the necessity of thinking about people’s unique and subjective experiences within the development of technologies. Hannah Davis ended her speech with a crucial takeaway and specified: “We shouldn’t let technologies, machine learning algorithms to push people toward the average or punish them for being outliers.”
