A step forward would be the inclusion of diversity efforts at the early stages of any process or project within the AI industry. We can even go as far back as encouraging diversity in academic institutions.
I have experienced the lack of diversity within the technology field. During my MSc in Machine learning and Computer Vision, I could count the number of black students in the whole course. Four to be exact, myself included. There were over 200 students on the course.
The lack of diversity within institutions might not be intentional, but it should be addressed woefully. If we have more individuals with diverse backgrounds gathering data and building AI systems, we could possibly see the inclusion of segments of datasets that might have been previously overlooked.
Throughout this article, I’ve focused on the racial bias that exists within an AI system. But we should also be aware of the other forms of bias that can occur. Take, for example, the study carried out by researchers at Carnegie Mellon University in 2015 that brought to light the gender discrimination that occurred within Google Ads.
Long story short, women were less likely to be presented with high paying jobs adverts when compared to their male counterparts.
Again, the cause of bias within these systems can not be directly pinpointed, but solution efforts can be implemented in all processes associated with the development of an AI product.
A good start will be ensuring that training data are indeed representative of the practical scenarios that these AI systems are utilized in.
Another solution can be observed within the legal and policy-making sector. There are organizational bodies that are driving critical policymakers in the direction of ensuring that measures to reduce algorithm bias are a mandatory measure, as opposed to a choice.
In the majority of the cases, engineers that are building an AI system do not have an inherent internal bias and prejudice against a specific group of people.
Still, due to the lack of exposure to other cultures and walks of lives, there might be a disconnect between the actual reality the developed systems are expected to operate, in and how the creators intend for it to be used.
Ethics education within companies and organization are one of the solutions to reducing algorithm bias. Educating employees on cultural and lifestyle differences can create an awareness of groups within the society that might have been overlooked or not even considered.
Some companies are making active efforts within the AI space to increase the inclusion of underrepresented groups within academic institutions and AI-focused coursed. Take, for example, DeepMind scholarship program.
The scholarship the program presents is aimed at individuals from low-income backgrounds; African or Caribbean heritage or Women.
To solve global challenges, it takes the coming together of great minds and individuals from all background. And this is what we are starting to see.
Toward Trustworthy AI Development is an effort by researchers coming together to define a set of guidelines that enables developers to be responsible in their development of AI systems.