The larger the systems we prioritize, the more data they need as well - this is particularly the case with some of the more recent ‘internet-scale’ models that require ‘internet-scale’ data such as the CommonCrawl dataset amongst others. But, these raise issues of ‘internet-scale’ biases because these datasets tend to suck up the worst of humanity and amplify that through the models that use these datasets for training.

So, what can we do to combat these two negative trends? On the compute infrastructure front, there are calls from various groups in the US, for example, that are demanding a government-funded national research cloud that would make cheap compute accessible to anyone who would like to experiment in building AI systems. This would lower the barrier to participation and make it easier for anyone who would like to develop AI systems. In particular, this would be a boon for all of society since we would have the potential to empower local entrepreneurs and concerned citizens to develop AI systems and services for their own communities.

Data commons and open-source datasets that are well-maintained such as Common Voice from Mozilla are another way to bolster nascent, independent efforts from local communities to build AI systems that help meet their own needs without having to rely on technology that gets exported from a small geographic region to the rest of the world.

But, all of this requires significant technical expertise to make it a reality, and nurturing talent will play a key role if we’re to combat the pernicious concentration efforts that emerge from Big Tech not only being able to offer lucrative compensation, but also large datasets and accompany access to adequate computational infrastructure that makes it exciting for talent to pursue challenging problems that further their own careers.

A counterpoint to that trend is to offer projects that can provide meaning and purpose beyond just the points that are raised above, as a way to draw talent. In addition, offering educational training programs and engaging in capacity building, both through community-driven programs like those offered by Masakhane NLP and open-access programs and books like D2L.ai offer resources that will upskill the next generation of data scientists, machine learning engineers, and software developers who can bring their expertise towards solving society’s grand challenges while using AI capabilities in an ethical, safe, and inclusive manner.

New models of self-organization and community engagement have shown to be powerful levers to enact change in the world around us and being deliberate in our approach to leaning on these vectors of change can allow us to pursue lofty goals while staying true to community roots and achieving the ultimate goal of having a more fair, just, and equitable society for us all while harnessing all that AI has to offer.