By Zunaira Uzair
SparkCognition Inc.
If you walk down the aisle at a bookstore today, even a cursory glance at leading magazines or trade journals will reveal that Artificial Intelligence is the hottest developing trend in technology. While they haven’t always been so noticed, over the last few decades, artificial intelligence (AI) inspired algorithms have deeply ingrained themselves into our lives.
Simultaneously, for some of us, AI has taken on the image of a threat… to our jobs, or even our lives. But despite all this, the march of technology adoption continues. While planning our trips we use path finding algorithms that have been enhanced by AI practitioners, or turn to weather forecasting apps that can use modeling techniques and systems like deep learning neural networks to predict which days serve best for our excursions. Social media apps use intelligent computer vision algorithms to instantly detect faces in uploaded photos while book and movie recommendations matching our tastes and styles appear magically on e-commerce sites. Much of this has its roots in algorithms produced by the AI community. And just as the smartphone changed our lives in the late ‘90s by making communications ubiquitous, AI has the capacity to change our world in even more profound ways by making intelligence ubiquitous.
In the massively popular Machine Learning branch of AI, algorithms alter their behavior based on the data fed to them. They analyze information supplied as input, learn from it and make predictions “inspired” by the supplied data. They are, in fact, data-driven algorithms rather than static programs in which decisions are pre-coded by a developer. Since computers that use Machine Learning algorithms learn from the data sets they are provided, their “world-view”, if it can be called that, is based on what they are shown. This has pluses and minuses.
One incident which fits firmly in the latter category is the now famous learning malfunction exhibited by a Microsoft chatbot designed to learn dialog from tweets on Twitter. Sadly, when it found a massive number of insults, racist abuse and general slander to be part of the human-supplied learning set, it added swear words, insults and epithets to its own vocabulary. Suffice to say, the project was shut down quickly, but not before evidencing the importance of providing a well-vetted training set that reflects a more balanced reality; Twitter fights are not the real world!
So where do these important data selection choices come from today? Where do the infinitesimal details - such as initial biases and ensemble weights - come from? Where, indeed, do the problems many of these algorithms are being directed to, come from? People. And at present, overwhelmingly, the people involved with AI represent only half of humanity; the male half.
Of course, there are phenomenal women data scientists, such as Hilary Mason, and AI researchers such as Prof. Kristen Grauman at the University of Texas at Austin, but they are still few and far between. As much as I would like for it to be, unfortunately, as in many other areas of science, women aren’t quite yet half the talent pool.
This needs to change, particularly in areas such as AI, that have the potential to multiply and amplify initial biases and inputs to immense degrees. The output of an AI algorithm can potentially impact hundreds of millions of people in almost no time. With increasingly sophisticated AI systems that will start to exhibit “personalities” it will become even more important to incorporate a balanced and measured reflection of what humanity truly looks like, what it believes in and what it holds dear.
To underscore even further the criticality of encouraging more women in AI, consider that the issue is not simply one of intelligent systems that provide us with potentially biased insights or decisions. The fact is that we humans adapt ourselves to work with intelligent systems too. This reflexive behavior mankind exhibits towards systems has been measured and captured in numerous experiments, for example the evolution of the “thumb tribe”; a generation of smartphone users whose hands have mutated to transform their thumb into the dominant, muscular digit.
Think about that for a moment. The human body has rapidly adapted and mutated to a design decision taken by a small group of smartphone designers. This kind of reflexive impact will only grow with AI and it will not be contained to physical responses. It will include mental responses also.
I hold these views for I know the importance of what is at stake from my own experience. The company I work for, SparkCognition, builds AI systems that make a variety of critical decisions for our clients in areas such as Energy, Finance, and Government. The world is adopting Artificial Intelligence at scale, and it is important for us to think through the issue of balance.
As a woman working in the field of Artificial Intelligence, my hope is that the ideas I’ve presented can serve as a rational call to action; incentive for both men and women to encourage a gender balance in the AI community. Any realization of this hope will come only from women taking a more active interest in science in general, and AI in particular, and men realizing that an active approach to being inclusive has the potential to make our collective work - the systems we build - better. And that is a worthy goal for any scientist of engineer - man or woman - to vigorously pursue.
