‘Recent diversity reports show that women still make up only 20 percent of engineers at Google and Facebook, and an even lower proportion at Uber’, and it’s been reported that only 13.5% of people working in machine learning are female. Whilst the initial response to this is an astounding lack of diversity in the field and an unfair bias towards male applicants, the problem runs much deeper. Products are built and trained by a predominantly male team, feeding the product or system their own beliefs, creating something that they would want to use, or that they think the general public would want to use, but without the necessary input of a diverse team. The result? A biased system that provides a negative user experience to the wider population.
In the midst of the #MeToo movement, we are becoming more and more aware of biases in all industries. Everyone’s known they’ve been present for some time, but in light of media coverage, women are being given a platform to voice their experiences in attempting to excel and match their male counterparts in their industry. As a female-led team, RE•WORK are strong advocates for promoting diversity in AI, and at yesterday’s Women in AI Dinner we heard from leading female minds from Uber, Airbnb, Perkins Coie, and Facebook AI Research. These events bring together attendees of all genders from a wide range of backgrounds who are using AI to disrupt their industries.
Networking plays a central role in RE•WORK’s dinners, and attendees arrived at Restaurant Anzu in San Francisco to a champagne reception and where introductions were made, and before long attendees were deep in discussion about their experience and research in the industry and sharing their current work and research in the space.
‘I’ve always steered clear of ‘Women’ events, but this is such an inclusive and supportive atmosphere where people are focusing on the work, not the gender boundaries which is exactly what it should be.’
‘It’s a great lineup of speakers - some really great names and it’s empowering to see women in top roles at these huge companies.’
The evening was kindly sponsored by Perkins Coie, and Susan Fahringer, compére for the event and Partner at the company, opened by explaining why the dinner was such a compelling event to partner on: ‘we think fostering women in AI is very important. AI has a very strong and meaningful role in solving real world problems and women need to be directly and deeply involved in solving these problems which is why we’re glad to support this dinner. It gives other women role models and mentoring opportunities, and it will also be a lot of fun and very interesting - you can meet people that you might not otherwise get to meet this evening.’ RE•WORK would like to say a huge thank you to Perkins Coie for the partnership, and especially to Susan who provided some insightful and engaging comments throughout the evening, and highlighted the achievements and work of our speakers.
Attendees enjoy a three course meal, and between each keynote presentation were encouraged to move seats around the restaurant to maximise networking. There have been some really exciting conversations and Nehan from One Convergence said ‘I’ve only just graduated so I’m very new to all of this. This is already a great event, I’ve heard about people working in robotics, mental health and research. The applications of AI are immense and I’m really looking forward to the talks and plenty of networking this evening’.
Our first presentation of the evening was given by Negin Nejati, Senior Data Scientist at Airbnb, who spoke about the importance of spending time on defining the machine learning problem in the context of the overall product before diving for solutions. She explained that with the amount of available data that companies now have, the access to open source tools, and the low cost of computation, it can be tempting to ‘throw the latest technology at any given problem with little preparation, which can lead to overly complex solutions, suboptimal processes, and waste time.’ Currently part of the customer support team, Negin's team is making sure customers feel supported if they have a question or problem at any stage of their Airbnb journey. For example, if a customer has forgotten a password, wants to know how to become a superhost, or has problem with the wifi connection of his/her listing, her team helps by providing self-help materials to users to solve their problems as well as by building tools to enable agents to be more effective in helping the customers. Negin, who previously led the geosearch effort at Apple Maps, shared some real-world examples and case studies from her work. Currently focusing on improving customer support through machine learning at Airbnb, she gave more details about her current project:
'We are working on identifying the reason of a customer contact. Accurate user issue selection is very important as it helps us determine the material we expose to the customer or agents to resolve the issue. If we can automatically identify the issue we can free our customers and agents from mapping customer's problems to our internal vocabulary of potential issues and get to supporting our customers more quickly.' She argued how better understanding the end to end product and influencing the way the problem fits in this overall picture helped clarify and simplify the problem. In particular, she talked about two examples: 1) clarifying the definition of the "user issue" and recognizing that it encapsulates 3 different concepts: "user issue", "root cause", and “desired resolution" helped clarify the problem. 2) limiting the length of the text customers can send to explain their issues helped simplify it.
Before the dinner, we sat down to record an episode of the Women in AI Podcast with Negin where we spoke about her whole career in AI as well as her current role and what’s next for her. She explained how she’s currently working in an all female team at Airbnb, and the overall inclusive and diverse atmosphere of the company.
After the main course and the next rotation of seats, it was time to learn about how Facebook AI Research are working towards creating agents that can see, talk, act and reason. Susan introduced Devi Parikh, Research Scientist at the company opened her presentation by saying that she’s been ‘Listed on Forbes 20 incredible women advancing AI research.’ Devi opened the floor by humouring us with a thought: ‘wouldn’t it be nice if machines could understand content in images and communicate this understanding as effectively as humans?’. This kind of technology would have countless real world applications and the potential to positively transform lives from assisting the visually-impaired use, educating both children and adults, assisting an analyst in extracting relevant information from a surveillance feed, providing information to a spectator at an art gallery, or interacting with a robot. Devi spoke about how at Facebook, the advances in NLP and computer vision are allowing them to get closer to this goal. For example:
Facebook are working to build machines that are able to answer navigate in their given environment to gather the necessary information to gather questions. The first problem Devi gave us was visual question answering ‘you’re trying to build models that can automatically answer questions about an image. There are open sources to help you train models to help you answer these sort of questions about images - we’ve been hosting challenges based on these problems and we’ve been looking at answering questions about several aspects of an images and follows up on pieces of information and questions in an image. This needs the machine to understand the entire history of the conversation which is a new aspect of visual dialogue. We’re also trying to bring in action by introducing a visual agent that’s being asked a question e.g. in the kitchen, it’s asked ‘what colour is the car’ - the agent is in the kitchen not outside, so the agent needs to understand that it has to navigate outside of the kitchen into the carpark to answer the question - what it sees next depends on the previous action it took.
The floor was open to questions, and one attendee asked ‘of the data you have, what is the majority of images? Is it homes, people, animals? Devi explained that: 'most of them are home environments - synthetic environments that are meant to be scenes we’ll encounter in everyday lives to help train the models.’
The evening drew to a close with more networking over coffee and wine, with attendees exchanging business cards, and sharing ideas and experiences on how they’re transforming their industries with artificial intelligence.
'I want to thank you for coming here, and the work you're doing. You're all doing important thinking in this area and making poignant contributions in AI, and I'm so glad to see so many meaningful contributions in the area.' Susanne, Perkins Coie
KC @multihyphenate: @reworkAI Hands down one my favorite events held in the Bay Area! #womeninAI #datascience #ai #artificialintelligence #MachineLearning #MachineIntelligence