According to a recent report, only 22% of AI professionals in the world are female. It has been concluded that this matter requires “urgent action”. RE•WORK have organized Women in AI Dinners since we were founded with the goal to bring together leading female experts in AI to share their latest work and to encourage women around the world to explore STEM.

Last night at the Women in AI Dinner in London, we brought together a blend of industry and academia to explore the latest progressions in research and the impact of AI in business and society. The first day of the RE•WORK Deep Learning in Finance Summit took place a stone's throw away and so we were pleased to welcome some familiar faces in addition to the majority who joined us for the first time from a wide variety of sectors.  At the dinner, we heard from female experts at the forefront of cutting edge research who are making a mark in their industries, whilst also inspiring a more diverse future for generations to follow, and the subject of diversity was a strong theme which was emphasised by Sarah Jarvis from who highlighted that “AI really requires the involvement of everybody”.

The start of the evening was bubbly in more ways than one as we began with champagne and progressive conversation in the welcome reception. Our host for the evening, Magda Piatkowska, Head of Data Science Solutions at BBC News, offered a warm welcome to guests as they sat down for the evening ahead of discovering recent advancements in AI and networking over a three-course dinner. Magda highlighted to guests that “RE•WORK has a mission to raise our profiles and to connect us to other women and talent.

“Thanks to people who are sitting here in the room and also outside who are pushing the AI agenda and it’s great to have that support like the event organised here.” - Magda Piatkowska, BBC News

We were pleased to hear a brief snapshot of Magda’s work, sharing that recommendations and personalisations are the main focus of their team and that all data that flows into the app is managed by them. One of the key recommender systems that Magda has deployed is ‘Most Popular’ content. Magda told us that although working for an organisation like the BBC is such as privilege, it can be hard to manage data in such large organisations which others may be able to sympathise with. Magda shared two learnings with us, which are that “everything we do with AI is a journey, but also there is little about technology and a lot about people. Most of the problems in my day job are not to do with technology, it’s to do with people.”

“We quickly learned that we needed to join the venture of the machine with the human to make the machine better.” Magda Piatkowska, BBC News

Madga dived deeper into why it is all about people, and there are three dimensions to how AI interacts with humans. “First of all we have people in front of our AI, we have people who interact with AI to do their jobs, and the third thing is people behind the AI." This final element is what connects all of our talks tonight.

Magda introduced our first speaker, Jane Wang, who is a Senior Research Scientist at DeepMind. Jane works in the neuroscience team and, consequently, her contributions to advancing AI are inspired by recent progressions in cognitive neuroscience, with a focus on “neural networks to work quickly through deep reinforcement learning, specifically through an approach called meta reinforcement learning”. Jane shared her findings and research into why humans are better at learning than computers, but furthermore, she delved deeper into what we can do to bridge this gap through methods such as meta-learning to advance the intelligence and capabilities of machines.

To breakdown some of the common terminology used around deep reinforcement learning, Jane walked the audience through familiar scenarios as a human playing a game and how this compares to a computer, suggesting “as a typical human you would probably think back to how you played a game in the past, such as collecting coins, and that’s how you would figure out the next step.” This is not so simple for computers and one of the elements that she touched on were ‘priors’. As a human, we get to any given situation through pre-existing knowledge which can be defined as priors. Jane explained that even babies as young as 15 months old have a sense of altruism and fairness. We also discovered that "prior and biases are two sides of the same coin", and they can only be corrected if we are aware of them. Machine learning can reflect the same kind of biases that we hold as they reflect any biases present in the training data and some AI applications where bias can be present include predicting criminality, determining sexual orientation, and filtering resumes.

Jane also introduced recurrent neural networks (RNNs) and the key function of these networks is they innovate over time and as they maintain a hidden state which is an internal activity level, they are able to maintain some level of memory. She explained that “we want to give the RNN all of the ingredients so it can adjust to the current tasks and any novel tasks so that it can take the best course of action.” The challenge with RNNs is that they need to be trained on a variety of tasks otherwise it will memorise the series of actions, however, we don’t want too much variety as there would be no common structure.

Jane concluded her talk sharing that at DeepMind they have found three requirements to close the gap with meta-learning which are recurrence, variety of tasks and structure, and methods/loops of learning.

Jane’s talk inspired some food for thought, and the conversation quickly started flowing in the room. As is customary at each of RE•WORK’s Women in AI dinners, we invited guests to move seats between each course to encourage fresh conversation and to build new connections.

Next up to share a talk was Sarah Jarvis, Head of Data Science at, who explored how they have merged multiple disciplines within machine learning which enable their cloud-based VUKU platform, and it was essential that they built a diverse team of researchers and engineers to make this possible. Sarah began her talk by asking guests to raise their hands to a series of questions to understand the background and experiences of attendees to date, the final question asking guests to “raise your hand if the people you have to work with do the same job but have a different job title to yourself.” This question was met with a sea of hands. Sarah continued to share that the purpose of her talk was to discuss not the maths or machine learning aspects, but instead a very fundamental part, which is about diversity. Sarah shared her experiences of seeking to expand the team at and after discussions with a recruiter keen to establish what the ideal Computer Scientist ‘looks like’, it was quickly realised that it wasn’t qualifications that Sarah was looking for, it was qualities to build a multidisciplinary team. Sarah continued by establishing that qualities could include values, their socio-economic background, experience, and more, and that “sometimes diversity goes hand-in-hand with having different jobs” and the binding ingredient that they have discovered brings a multidisciplinary team together is communication.

“At this time in tech, the tools you’re using in building frameworks, may have been made from people who are from different background from you and this is fantastic. This is what I see at” - Sarah Jarvis,

“I’m going to make an assumption that if you work in AI, you’re going to be working on a tool intended for the real world and that is used by someone that is not yourself. So when we’re building a system, we’re not building the entire world, we’re building a representation including biases to go by. If we can have a team of people that means we have more and more viewpoints, then we can test each other and collaborate with each other and ultimately build a more robust system.” - Sarah Jarvis,

The takeaway that Sarah was keen to offer to conclude her talk was “that diversity is a valuable thing to have and it’s not an easy thing to have. It’s not about the labels, it’s about being proactive and diverse in the representations you have in your team.

In the Q&A, when asked about the challenge of attracting women to seek opportunities in STEM, Sarah highlighted that it is not just about gender, it is also about factors such as where someone might have trained, and “it’s about opening the door and inviting people to come in who wouldn’t normally walk through the door.

Our final talk of the evening was delivered by Mounia Lalmas, Director of Research at Spotify who enlightened guests on how the music platform is using AI to understand user behaviour and engagement, particularly in respect to their playlists.

Mounia leads the Personalisation team and she immediately began by sharing that “it’s not just about algorithms. It’s about: what is the business metric? What features are you trying to apply? How do you do the training?” Mounia was also pleased to share that she is “very proud about how diverse we are at Spotify”.

Mounia touched on some of the techniques used by her team to get under the skin of their users which starts with a “click”. Clustering methods allow their team to create user groups based on the type of music they listen to, be it ‘Jazz’ or ‘Sleep’ genres, and this offers their team great insight into the behaviour and habits of their consumers. In understanding their users, and personalising user experiences in applying machine learning, they have seen a 20% success rate.

Guests continued their reflective discussions over coffee and dessert, and after some exchanging of business cards and conversations concluded, the evening drew to a close.

We are continuing the RE•WORK Women in AI Dinner series across the globe this year, and our next stop will be Boston on May 21 and you can register now to receive a pre-release Early Bird ticket. We will also be returning to London with the Women in AI Dinner on June 12 during London Tech Week. We hope to see you at a RE•WORK event soon!