A Values-First Approach to AI and Product

Generative AI tools like ChatGPT entered the mainstream less than two years ago, but they’ve already reshaped the online experience. These tools have also introduced the concept of “AI” to the mainstream. While the technology has been around in some forms for decades, never before has it been so visible and accessible to the general public.

Almost a third of younger Americans now use ChatGPT at work, according to Pew research, a number that’s increasing. Meanwhile, consumers feel both excited and concerned about the technology. Amid this swift uptake and mixed emotion, however, generative AI is quickly being incorporated into existing products without basic information about how the underlying models are built or perform in various contexts. And paradoxically, while the possibilities of generative AI are marketed as “limitless,” it’s difficult for consumers to choose and switch between different AI models.

So today, Mozilla is adding AI chatbot capabilities to the Firefox sidebar, an experiment we’ll build on in the months ahead. It’s part of our commitment to an internet where choice is a core tenet. This approach will not only offer choice to people who use Firefox — it will also demonstrate how consumer-level control and transparency can be incorporated into consumer technology.

For users who know Mozilla solely for its product work, this may seem like an early foray into AI. In a strictly product sense, that’s true. But for those who know Mozilla for its wider scope of work — research, advocacy, philanthropy — it's clear that this is the next step in an existing, ambitious body of work around AI. With this latest Firefox sidebar feature, we’ll be testing how best to fuse our values around AI in our products to improve both our offerings and others.

In 2020, Mozilla published our white paper on trustworthy AI, formally kicking off our strategic focus on the technology. (We published an update earlier this year.) Then, and in the years since, we’ve been asked, “What does trustworthy mean?” We define ‘trustworthy’ as an AI ecosystem where consumers of AI have a choice, and also where the builders of AI can be held accountable for their technology. These fundamentals have informed our AI work — from policy to product development — since then.

For example, *Privacy Not Included, our consumer tech buyers guide, has turned its attention to AI products. Our most recent edition helps consumers assess the privacy and security of relationship chatbots. We’re currently working on another edition to help consumers assess more general-purpose chatbots, like ChatGPT and Microsoft Copilot. We’re already thinking about how to integrate our findings directly into users’ sidebar experience, giving them more context about the tools they are (or aren’t) using. Because choice isn’t just about having a menu of chatbot options — it’s about understanding how those options align with our values and priorities, too.

Meanwhile, other Mozilla research and advocacy has focused on transparency and researcher access - key levers for making AI more trustworthy. We’ve urged Microsoft to be more open about its AI training data; ranked platforms on their openness to independent scrutiny; and are conducting a sweeping analysis of the AI auditing ecosystem. We’re also starting work on researcher access to models themselves; recently launched 0din, a bug bounty program for Large Language Models; and continue to explore open-source AI with Mozilla.ai.

With today’s sidebar launch, we’re applying all these learnings to our product work. Which, actually, is familiar terrain: user choice and corporate accountability have been Firefox values for decades.

We’re excited to fuse these values into our product design because there’s a void of features that offer people meaningful choices within AI-powered consumer tech. We’re also realistic about bringing this philosophy to the product realm. It’s not just consumers who are still figuring this technology out — it’s builders, too. From big names like OpenAI to smaller projects, from closed-source initiatives to open-source ones: no one has all the answers. Rather than wait on the sidelines, we want to get in the mix and help pioneer solutions - even if that means not having all the answers ahead of time. One issue we’re particularly focused on is interoperability: How can Mozilla ensure smaller, open-source chatbots can also plug into (and, in many cases, improve) the existing ecosystem?

For more than 25 years, Mozilla’s product work has always led with values. Today, we’re continuing this tradition in the realm of AI and bringing principles like agency, accountability, and open-source to an ecosystem that needs more of them.