Setting the Standard for Diversity in Artificial Intelligence
Artificial intelligence will be the most disruptive technology since the internet, and yet the work being done today is building in an inherit bias for the world view of white men. The problem with teaching a computer to ‘think’ is not simply one of coding but also one of the algorithms. The algorithms that we use to teach a system to recognize a flower become the algorithms to recognize a face. These, in turn, become the algorithms we apply to the concept of beauty or the risk analysis of criminal recidivism.
This is not an industry that is merely going to allow you to text freely while an autonomous car ferries you to and from that hot new sushi place downtown, artificial intelligence will be in every aspect of your life in a few years. From the assistant in your phone helping you find the information you need to the diagnostic tool aiding your doctor in finding the root of your ailment, to the planning tools that will help governments allocate resources for infrastructure improvement and redevelopment. Take this simple thought experiment; you’ve probably asked your phone a question before, be it Siri or Google, perhaps even another assistant. Imagine if everyone that had ever worked on speech recognition only spoke French and English as a language had only been bolted into the application as a secondary thought, as a translation from the original. Do you believe that the app would serve you well? Let’s take it a step further, lets talk health care. Let’s say that you are a member of the 52% of the world that is female, but the systems that are currently helping physicians triage global health needs are designed nearly exclusively by men. We know how well the US congress’ 82% male membership understands women’s needs, so how much of a voice in design do you think women in machine learning have when they make up less than 15% of the industry?
Bias is More than Just a Bug in the Algorithm
These problems are not potential issues, they have already been found in the programs that are currently being used. From facial recognition tools showing marked poor performance for faces of color due to data engineers rejecting samples that reflected less light, to job posting ads that showed only a 6th of the highest paying job ads to women as compared to men. Fei-Fei Li of Stanford made the salient point to check how algorithms are biased in search returns, the example being; go image search the term grandma and see how far you need to go through the returns to see a non-white person, how much further til you find a second. While it can be said that Caucasian is still the majority in north America, minorities are not the single percentage point that search would suggest.
Where we Stand
As our examples demonstrate the issues of bias being bred into artificial intelligence by an engineering force that is intensely homogeneous are scary. The current state of computer science is not looking good and while the numbers specifically for artificial intelligence and machine learning are only anecdotal they suggest things are significantly worse than the industry over all. According to DataUsa.io 80.8% of computer science degrees awarded in 2015 went to male graduates, while a review of the industry overall showed that the percentage of women in computer science industry has dropped to 26% down 9% from 1990.
Race follows a similarly distressing pattern with 2015 US degrees being earned by whites at almost twice the rate of all other groups combined.
These are not the trends we want to see when the crucial integration of AI into our daily lives is on the line. If we don’t change course the systems that are being trained today will be tailor-made for the world view and experience of white men of the ripe wise age of 35. While we have nothing particularly against 35-year-old white dudes in California I am sure they can recognize that their particular slice of life doesn’t represent 7.48 billion people and make no mistake, AI will effect us all in the coming decades.
Working Against the Tide
While difficult and worthy of concern, these issues are not insurmountable. Accel.AI is leading the effort here in the Bay Area by pushing not just for recognition of the issue but for real diversity in the industry by training a new generation of AI engineers from all walks of life. While many large companies including those most involved in the development of artificial intelligence are talking about diversity they are not making much of a dent in their over all workforce. An interactive infographic by the Wall Street Journal on diversity in tech last updated April of 2016 displays this clearly, some of the biggest players in machine and deep learning, all have less than 30% women in their workforce and all show a heavy bias towards white men. You can play with their infographic here.
Accel.AI on the other hand recently completed our second workshop which had an attendance that was 56% Non-male and 74% Non-white. While we are no Google, these are the kinds of initiatives we would love to see Google strive for in their own conferences and outreach. Our mission: to decrease the barriers to entry in understanding and engineering Artificial Intelligence while fostering community growth, diversity and inclusion, is best reflected in the events we have held and will be at the core of our program to change the identity of AI engineering.
Participants of these weekend workshops self select into introductory or advanced sessions on topics from the building blocks of Machine Learning to the latest research in Deep Learning & AI. They gain hands on practical experience with Python, Numpy, Scikit-Learn, and TensorFlow software packages. They practice research, development, sharing their newly acquired knowledge, and leading others or being mentored in collaborative sessions. They get to discuss their concerns, hopes, goals, and projects with others in the industry at roundtable discussions and an evening of networking. They learn about Business practices and UX design for AI products such as Chatbots. Lastly, they get to hear from a diverse range of industry professionals on the History of the AI industry, Ethics in AI, and how AI could be applied to Finance, Law or Politics, the Non-Profit sector, HR or Hiring Practices, etc.
Our workshops are of course simply the beginning and while we are gearing up to take our first cohort right here in the heart of Oakland California, one of the most diverse cities in the country, our message and leadership is already resonating through out the Bay. The outcomes we have displayed here have inspired the founder of Silicon Valley Artificial Intelligence Organization, Peter Kane, who is hosting a gathering of the minds, in an effort to create a full report on diversity and inclusion at AI events in the industry. We welcome more individuals and organizations to help us shape a diverse future of AI that is inclusive and enriched by many view points for the benefit of all mankind.
Please join us in this conversation on March 5th, 2017 at their Voices in #AI event!
Authored By: Tim McMacken, COO @ Accel.AI
