Mozilla’s Vision for Trustworthy AI

Mark Surman

By Mark Surman | Dec. 15, 2020


Mozilla is publishing its white paper, "Creating Trustworthy AI."

A little over two years ago, Mozilla started an ambitious project: deciding where we should focus our efforts to grow the movement of people committed to building a healthier digital world. We landed on the idea of trustworthy AI.

When Mozilla started in 1998, the growth of the web was defining where computing was going. So Mozilla focused on web standards and building a browser. Today, the computing -- and the digital society that we all live in -- is defined by vast troves of data, sophisticated algorithms and omnipresent sensors and devices. This is the era of AI. Asking questions today such as ‘Does the way this technology works promote human agency?’ or ‘Am I in control of what happens with my data?’ is like asking ‘How do we keep the web open and free?’ 20 years ago.

This current era of computing -- and the way it shapes the consumer internet technology that more than 4 billion of us use everyday -- has high stakes. AI increasingly powers smartphones, social networks, online stores, cars, home assistants and almost every other type of electronic device. Given the power and pervasiveness of these technologies, the question of whether AI helps and empowers or exploits and excludes will have a huge impact on the direction that our societies head over the coming decades.

The question of whether AI helps and empowers or exploits and excludes will have a huge impact over the coming decades.


It would be very easy for us to head in the wrong direction. As we have rushed to build data collection and automation into nearly everything, we have already seen the potential of AI to reinforce long-standing biases or to point us toward dangerous content. And there’s little transparency or accountability when an AI system spreads misinformation or misidentifies a face. Also, as people, we rarely have agency over what happens with our data or the automated decisions that it drives. If these trends continue, we’re likely to end up in a dystopian AI-driven world that deepens the gap between those with vast power and those without.

On the other hand, a significant number of people are calling attention to these dangerous trends -- and saying ‘there is another way to do this!’ Much like the early days of open source, a growing movement of technologists, researchers, policy makers, lawyers and activists are working on ways to bend the future of computing towards agency and empowerment. They are developing software to detect AI bias. They are writing new data protection laws. They are inventing legal tools to put people in control of their own data. They are starting orgs that advocate for ethical and just AI. If these people -- and Mozilla counts itself amongst them -- are successful, we have the potential to create a world where AI broadly helps rather than harms humanity.

It was inspiring conversations with people like these that led Mozilla to focus the $20M+ that it spends each year on movement building on the topic of trustworthy AI. Over the course of 2020, we’ve been writing a paper titled “Creating Trustworthy AI” to document the challenges and ideas for action that have come up in these conversations. Today, we release the final version of this paper.

This ‘paper’ isn’t a traditional piece of research. It’s more like an action plan, laying out steps that Mozilla and other like-minded people could take to make trustworthy AI a reality. It is possible to make this kind of shift, just as we have been able to make the shift to clean water and safer automobiles in response to risks to people and society. The paper suggests the code we need to write, the projects we need to fund, the issues we need to champion, and the laws we need to pass. It’s a toolkit for technologists, for philanthropists, for activists, for lawmakers.

At the heart of the paper are eight big challenges the world is facing when it comes to the use of AI in the consumer internet technologies we all use everyday. These are things like: bias; privacy; transparency; security; and the centralization of AI power in the hands of a few big tech companies. The paper also outlines four opportunities to meet these challenges. These opportunities centre around the idea that there are developers, investors, policy makers and a broad public that want to make sure AI works differently -- and to our benefit. Together, we have a chance to write code, process data, create laws and choose technologies that send us in a good direction.

Like any major Mozilla project, this paper was built using an open source approach. The draft we published in May came from 18 months of conversations, research and experimentation. We invited people to comment on that draft, and they did. People and organizations from around the world weighed in: from digital rights groups in Poland to civil rights activists in the U.S, from machine learning experts in North America to policy makers at the highest levels in Europe, from activists, writers and creators to ivy league professors. We have revised the paper based on this input to make it that much stronger. The feedback helped us hone our definitions of “AI” and “consumer technology.” It pushed us to make racial justice a more prominent lens throughout this work. And it led us to incorporate more geographic, racial, and gender diversity viewpoints in the paper.

In the months and years ahead, this document will serve as a blueprint for Mozilla Foundation’s movement building work, with a focus on research, advocacy and grantmaking. We’re already starting to manifest this work: Mozilla’s advocacy around YouTube recommendations has illuminated how problematic AI curation can be. The Data Futures Lab and European AI Fund that we are developing with partner foundations support projects and initiatives that reimagine how trustworthy AI is designed and built across multiple continents. And Mozilla Fellows and Awardees like Sylvie Delacroix, Deborah Raj, and Neema Iyer are studying how AI intersects with data governance, equality, and systemic bias. Past and present work like this also fed back into the white paper, helping us learn by doing.

We also hope that this work will open up new opportunities for the people who build the technology we use everyday. For so long, building technology that valued people was synonymous with collecting no or little data about them. While privacy remains a core focus of Mozilla and others, we need to find ways to protect and empower users that also include the collection and use of data to give people experiences they want. As the paper outlines, there are more and more developers -- including many of our colleagues in the Mozilla Corporation -- who are carving new paths that head in this direction.

Thank you for reading -- and I look forward to putting this into action together.