Last night, our co-founder Tabitha Goldstaub hosted an event featuring a panel of 5 expert speakers and hundreds of the community to discuss the importance of having Women in AI.
Why does Gender Diversity Matter So Much in AI?
Dr Joanna Bryson kicked off the evening by reminding us that machines simply learn what people know implicitly. In her latest research, her team found that a machine learning technique known as word embedding, exhibits striking gender and racial biases. “This is showing we’re prejudiced and that AI is learning it.” “You have to make societies better to make word embedding better.”
— Pia Stanchina (@PIASTANCHINA) May 2, 2017
— Dr Jess Wade ??? (@jesswade) May 2, 2017
Maxine Mackintosh then highlighted the urgency of the problem by laying out scenarios in which “we will all die” as a result of gender imbalance in the world of AI. There is not enough female representation in the OpenSource community (only 11% of software developers are female) and disturbingly, research found that GitHub approved code written by women at a higher rate than code written by men, only if the gender was not disclosed.
"Not enough women in open source community > women are better programmers > worse algorithms > everyone dies" @Maxi_Macki
— Sharmadean Reid (@sharmadeanreid) May 2, 2017
Silvia Chiappa believes there are 3 potential reasons behind gender imbalance: 1. Social and cultural differences; 2. Challenges in career progression; and 3. Inherent bias in datasets. The problem of biased data sets is particularly complicated. As an experienced ML scientist at DeepMind, Silvia believes we need to alter the objective function of algorithms such that the decisions made by machines become insensitive to factors such as gender. However, since we don’t fully understand the biases, is this easier said than done?
— mahoney turnbull 马甜甜 (@mahoneyjkt) May 2, 2017
Joanna warns that taking inherent biases out would result in a misrepresentation of the world. It also implies that somebody will get to choose what this “unbiased” world looks like. Might this be too much power for one person?
What are Governments and Corporates Doing to Address This?
It is paramount to design systems that explain themselves when they make decisions. Dr Sandra Wachter argues that, whilst the EU’s GDPR policy is a step in the right direction, it is a myth that it will reveal “the rationale of all types of algorithmic decisions”. In its current form, the policy wording focuses on system functionality, not rationale. More provisions must be put in place to ensure we develop AI that is truly fair, accountable and transparent.
— Libby Kinsey (@libbykinsey) May 2, 2017
Nowadays, businesses are so overwhelmed with rules and regulations that many have lost sight of what doing the right thing means. Tracey Groves is a strong believer that the key is to get men and women to collaborate and to value and respect differences. After all, doing the right thing is good business.
— ⭕️ Pete Trainor (@petetrainor) May 2, 2017
On the other hand, are corporates thinking responsibly when designing consumer-facing AI products? A particular concern was raised around Personal Home Assistants, and the fact that major tech companies designed their bots to be females. Is this creating the image of a subservient female? Joanna reminds us that “Robots are not your friends. They are an extension of large corporations with unknown cybersecurity risks.” As a key contributor to the 5 ethical rules set out in the UK EPSRC Principle of Robotics, Joanne also highlights the risks of not adhering to the 4th principle: “Robots are manufactured artefacts. Their machine nature should be transparent.”
What can we do to Encourage Girls to Enter AI?
Our audience believes the 3 most effective ways of improving gender representation in the world of AI are: 1. Hire more women into building AI (~30%); 2. Promoting positive role models (~24%); 3. Educate girls in STEM subjects at an earlier age (~24%).
Other suggestions made by the panel include:
- Corporate diversity initiatives such as offering mentoring programmes to underrepresented groups.
- Increase parental encouragement from a young age.
- Offer additional support to women, particularly at a later stage in their career to reduce dropouts. E.g. Flexible working arrangements and providing onsite childcare.
- Change the way we think about promotions. Tailored promotion trajectories for parents?
- Breakdown stereotypes. Reframe what data science is what being a data scientist is about. Expose the whole data science journey.
Raise Awareness. Break the Stereotypes. Be a Role Model.
Last night’s discussion is all but a glimpse of the tip of the iceberg, the very beginning of the conversations our society should be having around the question of diversity in the world of AI. As Tracey says “Women shouldn’t be apologetic for being ambitious”. We, both men and women, have a responsibility to foster confidence in young females and encourage them to celebrate their innovations and successes.
As the event came to a close, Ingrid Marsh, founder of Women With Voices reminded us:
“Do not underestimate the magnitude of the problem. We are just human beings and we shouldn’t beat ourselves up too much. We simply need to correct our mindsets.”
What do you think? Share your thoughts with us on Twitter (@cognitionx_X).
At CognitionX, our mission is to constantly move the conversation forward by bringing together the world’s AI thought leaders from a wide range of backgrounds. Our upcoming 2-day Innovation Exchange (CogX), held on June 20-21, will dig deep into the Impact of AI across 18 topics, ranging from the ethics and social good of AI, to the impact of AI in specific industries like insurance, legal services and manufacturing. Join us at this event to hear more from leading industry experts.
A huge thank you to our venue partner Rathbones, one of the UK’s leading private client investment management firms, for making this event such a success.Published in