Black Voices Are Training Big Tech. But Who's Protecting Them?

Forbes

6 days ago

Howard University and Google are partnering under the name Project Elevate Black Voices to release ... More data that can be used by artificial intelligence to better understand African American English.
It was recently announced that Howard University and Google are partnering under the name Project Elevate Black Voices to release data that can be used by artificial intelligence to better understand African American English (AAE). A media release from Howard University explained that researchers traveled across the United States to capture different dialects used within the Black community, collecting 600 hours of vocal data across 32 states in different dialects, dictions and accents. The project's aim is to address the inherent bias baked into many artificial intelligence-driven tools we use to address difficulties interpreting and understanding Black users who have dialects and diction more common in the Black community. Howard University will own the dataset and licensing, making the data available to researchers and institutions at historically Black colleges and universities (HBCUs) and will allow Google to use the dataset to improve its products.
In an Instagram post made by pocintech about the project, there were many who were leery of the partnership stating it wasn't a good idea, with some wondering how this partnership would benefit the Black community. It's understandable for some to be suspicious about this type of partnership. There is a long history of tech being extractive, exploitative and harmful to the Black community. The erasure of Black creators who have started trends and viral dances yet are the least protected on social media, racial bias in facial recognition software used for policing and surveillance, and technology that is more likely to flag Black creator's content as 'offensive,' are just a few examples.
Though the backlash is warranted, we should also consider the different ways that a robust dataset containing different types of AAE can benefit Black workers and help create better technology. Many companies use tech tools like Zoom and Microsoft Teams auto-captioning, AI notetakers, and meeting recording summaries. Having more data about AAE will help to improve these speech recognition tools, which can often misinterpret AAE. In some cases, a dataset like this may make employees less likely to codeswitch because they won't have to worry as much about whether their voice and speech patterns will be understood by workplace AI tools, making these tools more inclusive and accessible. Industries like healthcare, customer service and the legal sector, which may use speech-to-text tools for things like clinical documentation, customer sentiment analysis, and to transcribe legal documents could particularly benefit from Project Elevate Black Voices.
Many companies use tech tools like Zoom and Microsoft Teams auto-captioning, AI notetakers, and ... More meeting recording summaries.
The criticism that can be seen online is not completely unwarranted. There are very real tensions around who will have access to the dataset in the future and some are worried that the data could be extracted and exploited to the detriment of the community that it was designed to benefit. The value of this type of rich data outside of improved speech recognition tools is a preservation and archive of Black American culture during a time when Black history is being actively erased. In the wrong hands though, this type of data could be used in nefarious ways, furthering the use of Black AI influencers and digital Blackface, training AI products without the compensation or credit (and in some cases the knowledge) of those whose data was used to train these products, and expanding the use of surveillance technology that disproportionately harms and targets Black people.
According to Howard's media release about the partnership, 'The Howard African American English Dataset 1.0 will initially be made available exclusively to researchers and institutions within historically Black colleges and universities to ensure that the data is employed in ways that reflect the interests and needs of marginalized communities, specifically African American communities whose linguistic practices have often been excluded or misrepresented in computational systems.' It's understandable to question how this data could be used to further the harm that Black communities already face. While the initial intentions in collecting this data may be benevolent, it's imperative to understand how easy it is for data to be wielded for corrupt and capitalistic purposes.
It is a great move that the data will be available to HBCU institutions, but a word of caution: Black-led, Black-founded and Black-owned organizations and institutions are not immune to harm and extraction—one could argue that within these systems, harm is more insidious and is sometimes more challenging to spot and combat. Organizations and institutions that are primarily Black or led by Black people can still engage in practices that are oppressive and harmful. Audre Lorde taught us that, 'the true focus of revolutionary change is never merely the oppressive situations we seek to escape, but that piece of the oppressor which is planted deep within us.' We must be careful that when we are trying to reclaim equity and repair broken tools that we do not recreate newer versions of the same harmful systems.