September 20, 2018
We get to see and meet many different artificial intelligence teams at Figure Eight. After all, we have hundreds of customers using us as a core piece of their machine learning pipeline, as well as about the same amount of research institutions using our free academic license. And somewhere around 50 of the data science teams I’ve visited have claimed to have the strongest team in AI. It can’t be true that all these 50 AI teams are all the strongest!
While a lot of these teams are quite exceptional, I often have to tell many of them the same thing: if your machine learning team consists of 90% men in hoodies, all within a narrow age range, then you do not have the strongest team in AI.
For AI to truly benefit all of humanity, all of humanity has to participate in creating AI. When Figure Eight was known as CrowdFlower, we were most well known for running the largest marketplace for human annotators to train and evaluate machine learning models. With more than 100,000 people from 150 countries regularly taking part in our marketplace, we provide the ability not only to create training data but to create training data that benefits from a diversity of experiences and points of view. We still run this marketplace, of course, but today I am reporting on our technical team’s internal demographics, something we are doing at Figure Eight for the first time.
The diversity of tech companies, especially AI companies, is problematic. Here, I’ll talk about the technology teams: Engineering, Product, and Machine Learning. Too often, technology companies hide their diversity problems by having greater diversity in non-technical roles. All roles in a tech company are important, but it’s not a truly diverse company if each role is dominated by one demographic. It’s also crucial to be transparent about our leadership team’s demographic breakdown, as leaders are more likely to be role models for junior staff, and as glass ceilings can prevent the right set of role models and diversity of input at strategic levels.
Figure Eight’s gains in gender diversity
When I joined the company about a year ago, I made a commitment to diversity. One year ago, only about 15% of the Technology team were non-male, and none of the technical leadership were. Today, 37% of the Technology team are non-male, as are 50% of our leadership, including two thirds of our executive leadership.
In the graph above:
- “Non-male” is anyone who identifies as female or non-binary
- “Tech Executive” is anyone in technology who is also on the executive team that reports to the board (currently just three of us!)
- “Tech Leader” is anyone who is a manager, director, or otherwise a leader of a function (eg: Senior Program Manager)
In 12 months, we have gone from 0% non-male Technology leadership to 50% today. This is the statistic that I am most proud of. Having a diverse leadership team in place will help develop our talented staff as we continue to grow.
What we did well
We are more diverse: we went from 0% to 50% non-male Technical Leadership in 1 year, including two non-male executives. We went from 16% to 37% non-male technical staff in the same period.
We are more efficient: in raw numbers, the Technical team only grew by 39%, while we grew by more than 100% in the volume of data and number of customers we support. So, our more diverse team is now doing more for more organizations.
We are happier: it is hard to quantify, but I have always found a diverse team to be more cohesive. It’s no different at Figure Eight. People take the time to more carefully listen to their colleagues when they know they are coming from different life experiences, and communication and respect are improved as a result.
What we did okay
More people identify as LGBT in our technology team than at the national (and even San Francisco) rate, both within staff and leadership. However almost all are men. So, I’ll give us an “okay” for LGBT representation today, with room for growth.
For ethnic and geographic diversity, we are also better than most companies, but with room to grow. We have multiple staff members from each major region: Asia, Australia, Europe, the Middle East, North America, South America, and Sub-Saharan Africa.
What we can do better
When someone’s identity makes them feel like an outsider in multiple ways, they need the most support to overcome the inherent biases in our industry. I would rather give them leaders and colleagues that they identify with than give only my words as an ally. So, I aim to create a more diverse technology team in this way.
For combinations of underrepresented demographics, we can do better. This is partially a numbers game, as we are still much smaller than the large tech companies and the percentage of underrepresented demographics becomes smaller. We have individuals who identify as both LGBT and people of color, among both male and non-male Tech Leadership, so I am glad that we have representation and role models.
A few notes for further transparency
We’re not brushing anything under the carpet, so here are some more details about what I shared from our team:
- This omits one person who was here for less than a month and three summer interns, all male.
- These are the stats from the teams that I lead today, Engineering, Machine Learning, and Product, which I started running in January.
- We also use international contractors whose identities I don’t all know. I am certain that ratio of non-male to male are less diverse among our contractors, but I don’t have this historical data. The total number of contractors is less than total number of in-house people in data shown here, so it wouldn’t move the stats too much.
- Figure Eight’s Sales and Success teams also have people with an engineering title. In these teams, 45% of engineering staff are non-male, and the two leaders are both male.
- I only have ethnicity identity information at the company-level, and we are about 50/50 for people who identify as white/non-white.
AI for everyone
There are many more aspects of diversity that we should care about. For example, my personal expertise is diversity of languages in AI. English only makes up 5% of the world’s conversations daily, but makes up more than 90% of Machine Learning in the world. It is related to most other underrepresented demographics: speakers of minority languages are more likely to be from underrepresented ethnicities, and males are more likely to be educated in dominant languages like English.
I encourage every AI company, especially those claiming to have the strongest team in AI, to share similar diversity numbers about their technical staff and leadership.