Why a workshop for “Women in Big Data”? There are already a lot of workshops dealing with big data. They are about big data engineering, about society, about applications. But there are not so many, if any, that cover the whole field.

This international workshop brought together women in academia and industry working on big data research to share research insights across disciplines and to discuss research and career challenges. The workshop was especially for the next generation of female data scientists and engineers. The six speakers presented their inspiring and impressive work with big data and showed possible career paths in the big data era. The workshop is part of the National Research Programme on “Big Data” (NRP 75). In addition to their roles as principal investigators in NRP 75 projects, Lydia Y. Chen (Data Scientist at IBM Research – Zurich) and Sophie Mützel (Professor of Sociology, Media, and Networks at the University of Lucerne) have initiated and organized this workshop.

In the first keynote speech Gina Neff, Professor of Sociology at the University of Oxford, asked the question “Does AI have gender?” AI actually does not have a body so it has no gender. However, media, for instance, tries to visualize AI: if you look for stock photos of AI you find a lot of images showing female robots. Also, AI has a gender in movies. In a study of 77 movies, only in three cases did AI not have a gender. Voice assistance is the third example: If the voice gives you advice than it is a male voice. If it gives you assistance then it is a female voice. So female voices are used for what we think is female. Will AI undermine women’s judgement and voice?

We must be smarter about predicting and preventing abuse. We have to be smarter about how to build systems that don’t put people at risk. To conclude her speech Gina Neff summed up what we must do: Increase representation and diversity in tech; increase diversity in training data; ensure systems preserve voice and judgement; ensure systems empower the user to respond; design for use, plan for misuse, prevent abuse; implement transparency and accountability.

In the next keynote speech Lisa Amini, Director of IBM Research Cambridge, showed why AI needs even more diversity, and vice versa. Today AI is still narrow: tasks most suitable for current ML/AI are mapping well defined inputs to well defined outputs, no long chains for reasoning, tolerance for error, tasks not changing rapidly over time, no special dexterity, physical skills or mobility. AI needs to broaden. It’s still a long way to General AI but it’s going to change how people work. To adapt and prosper, the smart worker will have to invest in human relating skills like empathy, compassion, influence, and engagement. Women consistently outperform men in coaching and mentoring, and other skills. So, AI is a big chance for women.

Lisa Amini also presented projects from the MIT-IBM Watson AI Lab. It is a real joint-venture with professional researchers from IBM and professors and students from MIT. The fields of activity are AI algorithms, physics of AI, applications of AI to industries (healthcare, life sciences, cybersecurity), and advancing shared prosperity through AI. Despite tremendous progress the challenges ahead for AI are still enormous. The rate and pace are incredible but it’s never been easier to expand one’s expertise. The success of AI for business and societal benefit depends upon diversity of ideas, approaches, disciplines, cultures, and perspectives. Lisa Amini closed her speech with a vivid appeal: Everybody in this room is a role model. When you are the only woman in a meeting you are a role model. Bring in your ideas. It is never too soon to be a role model.

In addition to the keynote speeches three speakers held shorter invited talks. Isabelle Collet, Senior Lecturer at the University of Geneva, also started with a question: Is affirmative action for the inclusion of women in IT always a “good” practice? IT has a gender problem but why is that so. Isabelle Collet showed a few possible explanations and concluded with some examples of best practice. Studies reveal that it is not a question of skills but expectation. Stereotypes are omnipresent. Even if we assume that stereotypes are just a construct, we are still affected by them. But IT is neither male nor female, so how do we do better. If you say that it’s because of self-censorship you mean that women should do better. But maybe that is not fair. Girls lose self-confidence, have a low sense of self-efficacy, and a feeling of illegitimacy because of social censorship. Society should do better. A successful inclusion strategy from the Carnegie Mellon University in Pittsburgh (USA) and the NTNU (Norway) shows us how: Get women interested, recruit them (e.g. quota), and socialize them. We have to remove the wall!

Raia Hadsell, Research Scientist at Google DeepMind, told us how she rose to where she is now and how her past came that it was not straightforward. And then she showed what deep reinforcement learning is and what they do at DeepMind. They work with neural networks that have to learn how to play computer games – one game at a time or several. Raia Hadsell also spoke about their work with navigation. It consumes a lot of memory to navigate. What DeepMind strives for is to understand intelligence. Raia Hadsell gave the attendees three pieces of advice: 1. Be a lifelong learner: keep growing and embrace new opportunities 2. Be diverse: follow your curiosity, chase a bright idea, take the risky bet 3. Be loud: don’t just ask a question! Engage, elaborate, argue

From neuroscience directly to humanities: Caroline Sporleder, Professor at the University of Göttingen, gave the last invited talk about big data in the humanities. Their biggest challenge is the availability of data. Texts are not available in machine readable format, only metadata is available (not full texts), legal reasons prohibit digitization and sharing (copyrights), digitization is expensive, difficult (e.g. fragile documents) and time-consuming, and data holders are not always willing to share or only willing to provide access via search interfaces. In addition, if you have data then you deal in digital humanities you deal with the novelty conundrum: either “what good are digital methods if you can only confirm what is already known?” or “how do you know that your results are correct?”

Last but not least, it was the turn of Jennifer Dy, Professor at the Northeastern University in Boston (USA), for her keynote speech about “Learning from Complex Medical Data, Clustering and Interpretable Models”. She is not only engaged in fundamental research in machine learning, but also in many applications like skin cancer diagnosis, climate informatics, understanding emotion, the brain, mental health, and COPD (Chronic Obstructive Pulmonary Disease). In her speech Jennifer Dy spoke about her project with this disease. Clinicians believe that COPD is a heterogeneous disease: therefore the goal of the project was to subtype the disease. Fortunately, they had the data of 10’000 patients to find the right model. But it still took some time and a team of diverse specialists to find it. The lessons learned for this project and many more are: It is important to understand the problem domain, incorporating domain knowledge helps guide learning algorithms, and interpretability is important (visualization, summarization, use language of the domain). Domain scientists and data scientists have to work together to solve real-life problems. We have a chance to make a difference.