inclusion and inclusivity

Inclusion and inclusivity

Definition of inclusion and inclusivity

Inclusion and inclusivity represent the practice of not excluding people or groups based on gender, race, class, sexual orientation, age, disability, etc. For AI, inclusivity means including all members of society either in the data used or in the AI design process, ensuring that as many different groups as possible are represented in a model. Furthermore, inclusivity in AI can also mean that the application should be inclusive, meaning that there are no groups excluded from using the application (e.g., facial recognition should work for both white people and people of color; note that this often can be achieved by using inclusive data and an inclusive team).

A closely related concept is diversity. While diversity focuses on including different demographics in a workforce or a dataset, inclusion is a method to achieve this goal. However, one concept does not imply the other. That is, it is possible for something to be inclusive and not diverse and vice versa.
Another closely related concept to inclusion is bias in data. The difference between the two concepts is that bias is the result of non-inclusion. Thus, while bias is something we want to prevent, inclusivity is a method to do so.

Implications of commitment to inclusion and inclusivity

When one commits oneself to inclusion and inclusivity, two main things one has to ensure is to have an inclusive team and inclusive data. For the team, this means that a diverse group of people is desirable (e.g., including people of different ages, races, genders, socialeconomic classes, etc.). A diverse and inclusive team is a good way to ensure that different voices are heard in the design process, resulting in AI applications that are more inclusive as well. For the data, a commitment to inclusivity means that different demographic groups should be represented somewhat equally in the dataset. As a result, the dataset will contain less bias, which translates into a fairer AI application.

Societal transformations required for addressing concern raised by inclusion and inclusivity

A major problem for inclusive data collection is that it is very sparse. For example, most data contain binary gender values (i.e., male vs. female), thereby excluding people that do not fall within the gender binary. Thus, more data on marginalized groups needs to be collected. Furthermore, the inclusivity in AI teams can be improved by better representation of marginalized groups in the field and better education about working in the field of AI. If one wants to attract different demographics to the field, it is important for people from marginalized groups already present in the field to be visible, so that others that might be interested see that this career is an option for them. However, it should be noted that this problem is not easy to solve, as structural oppression of marginalized groups works on many different levels and in many different areas, causing people from these groups to be disadvantaged in many different ways (also implying that reaching these careers might be difficult in many different ways).